<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Filip Milneršić on Medium]]></title>
        <description><![CDATA[Stories by Filip Milneršić on Medium]]></description>
        <link>https://medium.com/@filip.milner?source=rss-24e9726f2868------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*C-3kvi_SKUlhs3t1v9OOnA.png</url>
            <title>Stories by Filip Milneršić on Medium</title>
            <link>https://medium.com/@filip.milner?source=rss-24e9726f2868------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sat, 21 Nov 2026 00:40:50 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@filip.milner/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Future OS]]></title>
            <link>https://medium.com/@filip.milner/future-os-7a54ef1d0cd9?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/7a54ef1d0cd9</guid>
            <category><![CDATA[marketing]]></category>
            <category><![CDATA[operating-systems]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[linux]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Fri, 06 Nov 2026 22:02:16 GMT</pubDate>
            <atom:updated>2026-11-06T22:02:16.917Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*F7ZvmAu4fCnR9eAqPd5Ssw.png" /></figure><p>My bio says “building Future OS”, so the time has come to write a thing or two about it.</p><p>I’m building an operating system that accumulates an understanding of you and uses it to help you express yourself, find people you fit with, communicate privately, and exchange value. Your knowledge, personality and creative preferences would inform what the system does for you, and it would keep learning as you change.</p><p>It’s built on top of Ubuntu Linux, with telemetry turned off. The intention is to give the computer enough context to carry out useful work on your behalf, including work that currently requires you to repeatedly explain yourself to an AI and supervise every little step.</p><p>The four main parts are:</p><p>• Finding compatible people through a psychological profile, with the underlying information kept private.<br>• Fully encrypted communication.<br>• A new cryptocurrency integrated into the OS.<br>• Content creation, posting and communication based on your own knowledge, personality and preferences.</p><p>So far, I have the content creation, database creation and updating, and automatic crossposting with intelligent scheduling built. These are still in an early form. The rest is the system I’m building toward.</p><p>Let me explain why these things belong together.</p><h3>Finding your people</h3><p>I keep seeing people describe how distant they feel from those around them. Finding someone who understands what you mean, enjoys the same kinds of conversations and has a compatible outlook can have an enormous effect on your life.</p><p>The Harvard Study of Adult Development has followed people across decades and repeatedly connected the quality of close relationships with health and happiness. That is a fairly good reason to take finding the right people seriously. The people you spend your life with affect what that life feels like.</p><p>Meanwhile, an extraordinary amount can already be inferred about someone from their online behaviour.</p><p>In a 2015 study published in PNAS, researchers used Facebook Likes to predict personality traits. Their predictions agreed more closely with participants’ own personality questionnaire results than the judgments made by their Facebook friends. The comparison concerned measured personality traits, and the friends were also answering a questionnaire. Still, even that limited record of what someone clicked was enough to reveal a great deal.</p><p>Now consider the amount of information you could deliberately give a system for your own benefit. What you believe, what interests you, how you react to things, the people you enjoy being around, and the kinds of interaction you find exhausting.</p><p>There is a considerable difference between collecting that information and using it to improve the person’s life. Meta’s own 2025 annual report says that almost all its revenue comes from advertising. It also identifies reduced engagement and time spent on its products as risks to that advertising revenue. There is an obvious financial incentive to keep you there.</p><p>Personally, I’d prefer technology that uses an understanding of me to help me find people I actually enjoy spending time with. A good conversation with someone you connect with can be worth considerably more than another evening scrolling.</p><p>The approach for Future OS is to develop a psychological profile indirectly through ordinary use. As the system learns how you think and what matters to you, that understanding could also be used to identify compatible people, nearby or elsewhere in the world.</p><p>Compatibility requires more than finding somebody who agrees with everything you say. Two people can have similar interests and still get on each other’s nerves. Similar temperaments can sometimes make that worse. The matching needs to account for how people would actually fit together.</p><p>That is the problem to solve. Counting shared interests would only get you part of the way.</p><p>The psychological profile would remain encrypted on your own computer. It would contain information far too sensitive to hand over just to find out whether you might get along with somebody.</p><p>The privacy approach I’m exploring involves zero-knowledge technology. A zero-knowledge proof allows someone to demonstrate that a defined statement is true without revealing the private information used to establish it. For Future OS, the aim is to check compatibility without exposing either person’s underlying psychological profile. The matching protocol itself still needs to be designed and tested.</p><p>You would also be able to opt out of creating the psychological profile, with a way to verify that the system respects that choice. Some people will be uncomfortable with such a sensitive collection of information existing on their computer at all. Fair enough. They should still be able to use the rest of the OS.</p><h3>Private communication and value exchange</h3><p>Once you find people you connect with, you need a private way to talk to them. Fully encrypted communication is part of the OS design.</p><p>I haven’t researched the implementation thoroughly yet. There are tools and protocols to evaluate, and by the time I get to that part, there will probably be more. The requirement is clear. The particular implementation remains to be chosen.</p><p>Value exchange also belongs in the system. People should be able to exchange value while retaining more of what they create. My view of the current monetary system is considerably less charitable than the usual discussion of making payments slightly faster or more convenient.</p><p>I consider the amount of value extracted from people’s work through the arrangements we currently accept to be obscene. Explaining the full economic argument would require another essay. The reason it matters here is that helping someone create useful work and connect with other people naturally leads to the question of how they exchange the value they create.</p><p>The cryptocurrency I intend to integrate is being developed by someone else. My part is the integration into Future OS once it becomes available. That part of the OS depends on the other project succeeding.</p><p>It’s an immense undertaking. I’m following it with considerable interest, but its completion is outside my control.</p><p>Attention, opportunity and the work of posting</p><p>Now we reach the part I’ve already developed, albeit in its early form.</p><p>I’ve wanted to write online since I was about 12. Ironically, my writing hasn’t improved all that much since then, except in vocabulary and formatting. I have a natural talent for it, and putting myself out there earlier would probably have made my life considerably easier.</p><p>However, writing, promoting myself, and thinking about how to promote and monetise my work would have consumed a large amount of the time I spent doing other things. I had other domains of personal growth that I considered more valuable, and I still don’t consider choosing them a mistake.</p><p>I can recognise the opportunities I missed without wishing I had spent those years differently.</p><p>For many people, the situation is different. The expected workload stops them before they even start. They aren’t choosing some extraordinary opportunity for personal development elsewhere. They look at everything involved in producing something worthwhile and getting it in front of people, and decide they don’t have the time or persistence for it.</p><p>Their expectation about the work is quite reasonable.</p><p>An idea has to be expressed clearly. A post needs to be appealing enough for someone to give it their attention. Then there is formatting, imagery, video, platform-specific adaptation, scheduling, crossposting, replying, and figuring out what works. You have to keep doing it while also living the life that gave you something to say in the first place.</p><p>And yet, I repeatedly encounter accounts of people whose lives changed immensely after they started writing or sharing their work online. They found clients, built businesses, got approached for work they had wanted for years, and finally had opportunities to pursue things they’d dreamed about since childhood. Others found close friends, collaborators or the person they ended up spending their life with.</p><p>There is survivorship bias in these accounts, of course. The people who tried and got nowhere are less likely to write a celebratory post about it. But the opportunities being described are a major reason I consider this worth building.</p><p>The recurring requirement is tenacity and consistency. You can’t count on putting in a month or three and having your life change. You also have to make an effort to turn what you know into something clear, appealing or useful to somebody else. Constantly broadcasting unfiltered thoughts and grievances is a fairly different activity.</p><p>We live in an attention market. Being capable of something and having other people know that you are capable of it have very different consequences. Public work gives someone a chance to recognise what you can do, what you care about, and whether they would like to know you or work with you.</p><p>Future OS is designed to take over the ongoing production and distribution work that makes participating in that world so demanding.</p><h3>The personal database</h3><p>For this to work, the system needs a substantial amount of material that actually represents you.</p><p>The initial construction is demanding. You need a large collection of short posts in your own style, longer pieces of writing, and links, posts or PDFs that you consider truthful or interesting. The writing has to contain your own ideas, judgments and way of expressing yourself. AI can help you create and polish it, but the result has to be yours.</p><p>Giving an AI a pile of generic AI writing would provide a rather unfortunate starting point for a system intended to express your personality.</p><p>The database stays on your computer. Over time, it would accumulate what you know, what you believe, how you feel about things, and what you would like to remember but tend to forget. As your views change, its understanding needs to change with you.</p><p>This is the context required for an AI to anticipate what you would say or make something that represents what you meant. Without it, you would have to explain your views, preferences and intentions again every time you wanted a useful result.</p><p>The intended workflow is simple to describe. You have an original thought, speak it into a microphone for 30 seconds, and the system uses that thought together with your personal database to create a polished piece of writing, imagery or video.</p><p>The thought still comes from you. So does the judgment about whether the finished piece expresses it properly. What becomes automated is the work required to develop that thought into something you would be happy to put your name on.</p><p>Visual expression matters here too. Your preference for a particular composition, the colours you choose, the kinds of imagery you like, and the details you reject should all contribute to what gets made next. Those creative choices need to accumulate over time.</p><p>Eventually, the result should resemble what you would have made by hand if you had the necessary artistic and technical skills. It should reflect your own taste, including the things you find beautiful that another person might dislike. A technically impressive image that leaves you thinking “yeah, that’s nice, I guess” hasn’t fulfilled that purpose.</p><p>The same applies to the writing. A grammatically flawless paragraph can still be a terrible representation of what you meant. You should be able to correct it, have the system learn from that correction, and spend less time dealing with the same problem the next time.</p><p>So there is a substantial initial sacrifice, followed by a system that continues learning through your use and decisions. Your original ideas and approval remain necessary. The ambition is to make the ongoing production burden very small once that foundation exists.</p><p>No such thing as a free lunch, I guess.</p><h3>Publishing and conversations</h3><p>From there, the system would adapt the finished work for different platforms, schedule it, crosspost it across dozens of them, and learn from the results on each one. You choose what gets published.</p><p>The aim is a hundredfold increase in social presence without taking a hundredfold amount of your time. The broader workflow still has to be built out.</p><p>There is very little appeal in repeatedly learning the current best posting times, manually moving content between platforms, and going back and forth with different tools to turn one piece into the formats each platform requires. That work takes time away from the thinking, research and experiences that gave you an original idea.</p><p>It also becomes increasingly awkward to keep up with people when your work is spread across many platforms. Greater reach is useful partly because it allows you to meet and speak with people you would otherwise never encounter. Those conversations need to be manageable too.</p><p>The planned workflow delivers comments and direct messages from every platform you crosspost to into a dedicated Telegram chat with your operating system.</p><p>The AI would write the first draft of each reply, using what it knows about your views, knowledge and way of speaking. You read the draft and press Respond to send it. If something needs changing, you use Edit first. The system learns from the changes.</p><p>This could be a reply explaining a point you’ve already written about, a conversation with someone who shares your interests, or an introduction that leads to something useful. The point is to make it easier for you to participate in those conversations while retaining control over what you actually say.</p><p>I have very little interest in the generic AI replies already being posted around social media, where somebody’s entire contribution is a polished restatement of the post they’re replying to. More of that would be a thoroughly depressing use of the technology.</p><p>The personal database and your involvement are essential. You still create original ideas. You decide what represents you. You read and approve the replies. Your corrections teach the system where its understanding of you was wrong.</p><p>Of course, someone can also use automation as an excuse to stop thinking. They can accept whatever comes out, send replies they haven’t read, and gradually become a spectator to their own online presence. That possibility exists, and I have no interest in pretending a piece of software can supply the willingness to remain involved.</p><p>The purpose of building this is to make it much easier to act on what you actually think, know and care about. You should be able to put an idea into the world, express it in a way that feels like you, and talk to the people who respond without having the production work consume your day.</p><h3>What exists so far</h3><p>The parts I have built are automatic content creation, database creation and updating, and automatic crossposting with intelligent scheduling. They are the early foundation. There is a lot left to do across the wider system, including the more complete creative and communication workflows described here.</p><p>My priorities are already split across several projects, and I don’t know for sure whether I’ll ever publish this technology, or whether it would be an open-source release, a subscription, or both.</p><p>The direction is clear, though. I’m building toward a computer that develops an understanding of the person using it and puts that understanding to work for their expression, relationships and ability to do something with what they know.</p><p>That is what I mean when I say I’m building Future OS.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7a54ef1d0cd9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Investing & Psychoanalysis]]></title>
            <link>https://medium.com/@filip.milner/investing-psychoanalysis-c874ef071def?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/c874ef071def</guid>
            <category><![CDATA[business]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[investing]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Sat, 31 Oct 2026 11:19:31 GMT</pubDate>
            <atom:updated>2026-10-31T11:19:31.283Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TEk0vEmYcB2I17rhoQ8BEQ.png" /></figure><p>How to invest in start ups 101 — it’s all about the team, not the deck, not the market, idea, or anything else.</p><p>It should be fairly simple to distinguish founders who want to build an exceptional company from those who want to live the advertised startup life.</p><p>One group wants the outcome. The other wants the identity.</p><p>They want the fundraising announcement, the conference badge, the “Founder &amp; CEO” title, the expensive dinners disguised as networking and, eventually, the obligatory photograph leaning against a rented Lamborghini while explaining sacrifice to everybody else.</p><p>Unfortunately, many investors continue funding the second group.</p><p>Never have there been more teams chasing capital, more massive-upside ideas waiting to be developed, or more unusually capable people essentially begging for the resources required to build them. Money is simply stored energy, direct it correctly and it feeds human capability. It buys time, equipment, talent, distribution and the right to continue experimenting after the first attempt fails. Without resources, little can be done in entrepreneurship regardless of how brilliant the underlying idea may be.</p><p>Yet capital disproportionately flows towards the people who know how to speak fluent “venture capital”. They understand the performance. What investors want to hear. They know which market-size graphs belong in the deck, which fashionable technologies should appear in the description and how much confidence must be projected before confidence becomes suspiciously theatrical. They have rehearsed answers for every predictable objection and can describe an unproven company as if its global dominance were merely waiting for the paperwork to clear.</p><p>Meanwhile, many of the people most capable of multiplying capital are terrible at performing certainty. The result is catastrophic. Passionate, responsible geniuses remain underfunded while charismatic egomaniacs receive millions to discover themselves at somebody else’s expense.</p><p>The founder you actually want</p><p>Back people who would rather die than lose your money. Not because they are frightened of investors, but because receiving somebody’s trust creates an almost unbearable sense of personal responsibility in them.</p><p>A bad initial idea will not destroy these people. They will pivot. A terrible launch will not humiliate them into denial. They will examine why it failed and fix it. An unexpected competitor will not send them into months of public posturing about their “vision”. They will return to the product, the customers and the numbers.</p><p>They do not treat capital as money they have acquired. They treat it as borrowed capability that must eventually be returned in multiplied form.</p><p>This naturally affects everything.</p><p>You will not see luxury cars purchased through the company, vanity trips disguised as business development or elaborate offices designed to reassure the founders that they have already succeeded. The CTO will not be allowed to spend eighteen months constructing his dream architecture while customers remain unable to use basic features. Nobody gets to build a cathedral when the company still needs a functioning roof.</p><p>Every dollar is counted because every dollar represents additional runway. Every hire must justify the time and complexity it introduces. Every feature must connect, directly or indirectly, to customer value, retention or revenue. Even apparent success remains open to reconsideration because the founder’s loyalty is not to the original plan.</p><p>The loyalty is to the long-term survival and growth of the company.</p><p>This is the quality people admire in the famous founders quoted at conferences. The obsessive work, relentless adaptation, personal sacrifice and refusal to accept failure as a final answer.</p><p>Curiously, many investors celebrate those traits in retrospect while selecting founders according to pitch polish, pedigree and social confidence.</p><p>Why investors keep choosing the wrong people</p><p>Choosing who to invest in resembles the circumstances involving a first romantic date, and the decision whether to go on a second. Everybody arrives as an edited version of themselves.</p><p>Potential investees know they are being examined, so they construct a facade specifically designed to satisfy the person across the table. Naturally, a founder who wants your helicopter money is not going to begin the conversation by explaining that he enjoys status, avoids unpleasant responsibility and becomes defensive whenever reality contradicts him.</p><p>He, or she, will tell you that he/she is mission-driven. He/she will describe himself/herself as resilient, data-driven, capital-efficient and customer-obsessed. If necessary, he/she will also care deeply about whatever social cause became fashionable three weeks before the meeting.</p><p>The question is not whether founders know the correct words. The question is whether their previous behaviour demonstrates the underlying traits.</p><p>Investors frequently over-index on pedigree, loud confidence, pitch quality and the brilliance of the initial idea. These things are not entirely irrelevant, but they are weak proxies for what actually happens after the money enters the account.</p><p>A pitch deck describes the founder’s preferred future. Past behaviour reveals the person who will be responsible for creating it.</p><p>Three operating variables matter far more than the performance.</p><p>The first is sustained drive. Does the founder continue shipping, learning and adapting when the metrics remain flat, the runway becomes short and the emotional excitement has disappeared? Many people work obsessively while the project is new and socially rewarding. Far fewer remain dangerous after eighteen months of rejection, technical problems and quiet embarrassment.</p><p>The second is fiduciary discipline. Does the founder treat every dollar, hour and relationship as capital that must produce a return? This does not mean refusing to spend. Excessive cheapness can destroy a company as efficiently as extravagance. It means understanding the purpose of expenditure and being willing to remove anything that serves the founders more than the business.</p><p>The third is low ego drag. Ego drag is the amount of energy a person wastes protecting an identity that reality is attempting to update.</p><p>A founder with low ego drag can hear that customers dislike the product without interpreting it as an attack on his intelligence. He can abandon six months of work when the evidence turns against it. He can hire somebody more competent than himself, admit that a strategy failed and report bad news before the bad news becomes impossible to conceal.</p><p>A founder with high ego drag burns the company’s energy defending his original vision, chasing headlines, polishing public appearances and constructing explanations for why every failure was caused by somebody else.</p><p>If a potential investee selected for investment does not flash green across these three variables in his previous actions, your capital is likely already lost. You just don’t know it yet.</p><h3>The idea is overrated</h3><p>The startup world pretends to worship execution, yet investors still become hypnotised by ideas. This is strange because many of the world’s most successful companies emerged from ideas that were incomplete, badly timed or pointed in the wrong direction.</p><p>William Wrigley Jr. began by selling soap and offering baking powder as a promotional gift. Customers became more interested in the baking powder, so he changed the business. He then offered chewing gum with the baking powder and discovered that customers wanted the gum even more.</p><p>The great vision was not revealed to him in a dream. He paid attention.</p><p>Amazon began with books, expanded into a marketplace for almost everything and later built an enormous cloud-computing business. Shopify emerged from an attempt to create an online shop for snowboarding equipment. Slack grew out of an internal communication system developed while its creators were building a game that failed. YouTube’s earliest direction was far narrower than “upload any video”, until user behaviour made the larger opportunity obvious. PayPal travelled through several forms before email-based payments became the winning mechanism.</p><p>Nintendo may be the most extreme example. It began with playing cards, experimented across radically different businesses and eventually found its enduring advantage in entertainment technology.</p><p>These companies were not created through perfect adherence to a sacred founder vision. They were created through disciplined, customer-signalled pivots.</p><p>Naturally, this does not mean vision is useless. A company without direction can pivot itself into permanent confusion. The founder still needs a sufficiently stable understanding of the problem being solved, the value being created or the future being pursued.</p><p>However, the product, market, distribution method and business model must remain negotiable.</p><p>The broader point is that investors cannot reliably predict the final company by examining the first pitch deck. The original idea is merely the first probe sent into reality.</p><p>Reality then responds. Customers refuse to buy, use the product differently than expected, obsess over a minor feature or reveal that the supposed problem was never particularly painful. Competitors appear. Regulations change. Technology improves. Distribution becomes more expensive. A market believed to contain millions of buyers turns out to contain seventeen enthusiastic people and one confused journalist.</p><p>At that point, founder quality becomes more important than idea quality. A mediocre founder with a brilliant idea may continue defending it until the company dies. An exceptional founder with an imperfect idea will extract information from every failure and move towards something the market actually wants.</p><p>If a founder reaches success without making any meaningful correction, he may have possessed extraordinary foresight. More often, he was simply lucky enough to begin close to the correct answer.</p><h3>How to identify the real ones</h3><p>The best evidence of future behaviour is costly behaviour from the past. Anybody can claim commitment when commitment remains verbal. What has the founder already sacrificed before there was an investor watching?</p><p>Look for visible, disproportionate and voluntary sacrifices made for the project. Perhaps the founder skipped meals to afford prototype parts, worked an unpleasant job to finance development, abandoned a comfortable career, learned an entire technical discipline or spent years building while receiving almost no recognition.</p><p>The specific sacrifice matters less than the revealed hierarchy of priorities. People spend time and money according to what they actually value. Their explanations merely arrive afterwards.</p><p>This is why apparently minor lifestyle choices can be informative. A founder claiming that the company urgently needs capital while spending heavily on leisure, status or constant international travel is communicating something. The problem is not that founders must live miserably or that taking a holiday is a moral failure.</p><p>The problem is contradiction.</p><p>You cannot describe the company as an existential mission, explain that every remaining month of runway matters and then behave as though personal consumption deserves equal priority. Both expensive travel and entrepreneurship consume money, attention and time. When resources are genuinely scarce, the choice reveals which “project” comes first.</p><p>Another useful signal appears in the founder’s employment history. Has this person ever accepted serious personal risk to do what he believed was right for an organisation, even when there was no obvious reward?</p><p>Perhaps he challenged a superior who was harming the company, defended a colleague at personal cost, exposed a dangerous mistake or took responsibility for fixing a problem that technically belonged to somebody else. These stories matter because they reveal whether responsibility exists only when it is profitable.</p><p>Potential investees usually do not realise that you are fishing for such information. Ask about conflicts with managers, moments when a team was failing, situations in which they disagreed with authority or decisions they made that harmed their own short-term interests.</p><p>Real stories have a different texture from manufactured ones. They contain inconvenient details. The founder remembers what was at risk and can explain why he acted anyway. There is often pride in the decision regardless of whether the outcome was favourable, because the person’s satisfaction comes from having behaved correctly rather than having been rewarded for it.</p><p>References should be used for the same purpose. Do not ask former colleagues whether the founder is intelligent, hardworking or “a great leader”. Nobody provides a reference and then casually announces that the person is a narcissistic parasite.</p><p>Ask concrete questions. What happened when the project began failing? How quickly did the founder communicate bad news? Did he become more useful under pressure or more theatrical? Was he willing to abandon his own idea? How did he behave towards people who could do nothing for him? What did he spend money on when resources became scarce?</p><p>The mechanism matters more than the adjective.</p><p>Investors can also create small real-world tests before committing serious capital. Give the founder limited resources and examine what he does with them. Ask for an uncomfortable revision of the plan. Introduce contradictory customer evidence. Request a detailed account of expenditure. Observe whether difficult questions produce curiosity or defensiveness.</p><p>You are not attempting to humiliate the founder or manufacture arbitrary stress. You are observing whether reality is allowed inside the room.</p><p>Venture capital requires psychoanalysis</p><p>Back hyper-focused founders who combine insane drive with sober responsibility.</p><p>Forget the theatrical perfection of the pitch deck. Forget the fantasy that a sufficiently impressive go-to-market slide can compensate for dishonesty, ego or an inability to adapt.</p><p>A great deck cannot produce an honest shareholder report. A fashionable strategy cannot force somebody to persist after the excitement has gone. Market analysis cannot make a founder admit that his favourite idea is failing, and an enormous total addressable market will not stop him from turning the company into a machine for financing his preferred lifestyle.</p><p>Business analysis still matters. Naturally, investors must understand the market, economics, competition, technology and regulatory environment.</p><p>However, those things describe the terrain. The founder is the person who must cross it after the map becomes wrong.</p><p>Venture capitalists therefore need to conduct far more psychoanalysis than they currently do. Not amateur diagnosis based on body language or whether somebody made enough eye contact, but a disciplined examination of incentives, identity, past sacrifice, emotional durability and the person’s relationship with responsibility.</p><p>Ideas do not build companies. Business documents do not multiply capital. People do.</p><p>Everybody knows this. Very few invest as though it were true.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c874ef071def" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Negatively Charged Water is the Next Big Thing in Wellness & Health]]></title>
            <link>https://medium.com/@filip.milner/negatively-charged-water-is-the-next-big-thing-in-wellness-health-4bf113053f5e?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/4bf113053f5e</guid>
            <category><![CDATA[water]]></category>
            <category><![CDATA[health]]></category>
            <category><![CDATA[mental-health]]></category>
            <category><![CDATA[alternative-medicine]]></category>
            <category><![CDATA[wellness]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Fri, 30 Oct 2026 09:49:23 GMT</pubDate>
            <atom:updated>2026-10-30T09:49:23.007Z</atom:updated>
            <content:encoded><![CDATA[<p>Negatively Charged Water is the Next Big Thing in Wellness &amp; Health</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XvY8GI0fp6oGctspKbPcaQ.png" /></figure><p>My personal experiences with “magic water” drive the conviction to research it at any personal expense for however long it takes.</p><p>Aside from plentiful “water community” reports that are full of wild claims, I have my own.</p><p>These experiences are what made me say “holy shit, this really works”.</p><p>Firstly, for context…</p><p>The first “water book” was published by Ngolo back in 2025, wherein he described extreme euphoria, feeling like time stopped, etc., after rubbing the negatively charged water on himself the first time.</p><p>I myself encountered the concept of negatively charged water some time before the official release by reading a post on X that said, “I tried every supplement under the sun, and negatively charged water is the only thing that finally fixed me”.<br>My interest was piqued.</p><p>One of his other posts described in detail how to make it, and it seemed simple enough, so I was sold on trying it out. Similarly, many posts written by the first “water makers” during that time made similar claims that sounded very alluring.</p><p>To be fair, the positive impact of consuming negative charge already fit my understanding of human biology perfectly, and I have written on the science behind the water in an essay which can be found on my website under “Research Essay — Scientific Foundations”.</p><p>When I do things, I do things right, and I take them to the extreme. In fact, my philosophy is: the truth is in the extremes. That’s where discovery happens. So naturally, I immediately went for a charging setup with a 10-kilovolt power rating and platinum electrodes.</p><p>For reference, that’s 10,000 volts. The wall sockets in your home are 230 volts, although that’s AC and the water charging setup is DC. But the comparison is somewhat valid.</p><p>So one incorrect move and it’s lights out, eternal nappy time.</p><p>Or so I thought. I did eventually shock myself. Luckily, my setup is limited to 25 microamps, so I’m alive, but I’m very hesitant when handling the setup now, so that attests to how unpleasant the experience was.</p><p>Like I got hit by lightning. But I digress.</p><p>I made the first water in December 2025. The original discoverer said not to drink it because of something to do with digestion, charge gradients and pH neutralisation. Personally, I don’t buy it.</p><p>In any case, it was bottoms up. Chargy chargy time.</p><p>A very, very strange sensation overcame me. It wasn’t euphoria, dread, excitement, expectation, none of it. Simply a very different state of mind. Lasted about 15 minutes.</p><p>But you could have chalked it up to placebo, as did I, so time for another one. Repeat charge, bottoms up, same thing. Same sensation. Again, could be placebo.</p><p>No real evidence yet.</p><p>I continued making hundreds of batches of water, testing them via various means, including ingestion.</p><p>Here is my top experience that demonstrates how insane this discovery is.<br>I need to provide context again.</p><p>In order to speed up the testing process, I didn’t keep track of hygiene in the sense that I didn’t wash the electrodes between batches or maintain a very high hygiene standard, so I didn’t intend to drink it.</p><p>Negatively charged water has different surface tension, which can be observed, and I also bought a capacitance meter which did seem to somewhat show how well the water is charged, but the device would get polarized fast and it would stop showing measurements.<br>The point is, I was testing the water’s physical properties to try to figure out which method creates the most charged water, and ended up with a lot of charged water.</p><p>It wasn’t for drinking for reasons explained, and I’m not a fan of rubbing it because you can’t ingest enough of it that way to get an extreme experience.</p><p>It would be a waste to throw it away, though.</p><p>I had about 4 liters of water charged with 10 kilovolts, so I filled a bathtub with regular water and poured the charged water in, cause why not.</p><p>I got in, and the onset was immediate. The only reference point I can give you is that it felt like when a large dose of MDMA or mushrooms starts lifting you up really… well, violently.<br>It’s very similar to having a very intense “spiritual”/“mystical” experience, but very few people actually know what that is or have experienced it, so I have to go with the drug analogy.</p><p>Anyway, it’s like someone poured mental kerosene into your brain and your brain went into overdrive, trying to adapt to this new mode of functioning.</p><p>A new fuel source.</p><p>Sounds worse than it is. It isn’t unpleasant on its own; it’s just uncomfortable because it’s very intense. I’d say the discomfort comes from the “control” part of your brain turning on and questioning what the heck is going on and where it is going.<br>There wasn’t any actual pain, pleasure or euphoria. At least for me, that specific time.</p><p>Just massive mental intensity.</p><p>This kept going for about 4–5 minutes without diminishing at all, and I wouldn’t say I’m a pussy or a stranger to discomfort and pain, but I really started thinking about getting out of the bathtub because it worried me where I might be heading. I didn’t know. Despite my philosophy that “the truth is in the extremes”.</p><p>Guess I was pussying out.</p><p>This dose was probably 1000x what anyone in the “water community” had taken at once at that point. So… that’s where my worry came from as well. Yeah, sure, more energy is good, but overloading a system can cause damage. Maybe.</p><p>But I didn’t get out!</p><p>Soon after I started having these “worries”, the intensity dissipated. I unfortunately do not remember how I felt afterward, probably good, but I remember noticing the next day that I had been in an incredibly good mood for the previous 24 hours. Far higher than baseline.</p><p>That’s what convinced me initially.</p><p>The second experiment follows a very similar pattern.</p><p>Half a year later, I was experimenting with charging milk because, as it turns out, milk is highly “chargeable”. Possibly because the proteins allow structured water to be created around them, storing charge in the process. According to other experiments, water infused with different substances changes its charge capacity, naturally.<br>Carbonized water seems to be a good candidate as well.</p><p>I charged about 2 litres of milk, which equated to about 30 batches — with the same 10-kilovolt setup.</p><p>Bottoms up. Yeah, 2 liters.</p><p>Now here’s the important part. I am a relatively positive person, but when I wake up feeling shitty — which isn’t common, but it happens — I just have a shit day, and so far, there’s nothing I can do about that.<br>I have never turned a shitty day into a good day, in recent years.</p><p>I drank these 2 litres of charged milk, and instantly my day changed. The good mood persisted for 24+ hours.</p><p>These two experiences confirmed for me what everyone else is talking about. This water is IT. It’s the biggest revolution ever. I consider it on par with mastery of fire, application of the wheel, and the written word.</p><p>Now, some might wonder: if you’ve been able to make water for so long and have been drinking it for so long, do you not have anything more to report? No, I do not, because I have not. I was busy with “Future OS”, which I’ll write about soon. That project is also important.</p><p>But I have once again resumed water research, and have had milder experiences that could be chalked up to placebo. I have a new setup, so I’ll certainly have something new to report relatively soon.</p><p>Before I continue with feedback from the rest of the water community, I want to clarify why I’m so hung up on hunting for durable “feeling good” sensations. I am a firm believer that a prolonged good mood, relaxation, is directly related to one’s health.<br>And I also don’t have the instruments to measure anything else really.</p><p>Anyway, community feedback:</p><p>“Libido way way up. Actually scary”</p><p>“It makes my skin softer and look younger, and it helps with stiff joints and neck”</p><p>“Strangers more likely to smile back”</p><p>“People are nicer to me too, I think because there’s less pain strained look on my face Imfao”</p><p>“Blessed water will be big. I’m experiencing profound healing. This is revolutionary”</p><p>“Blessed water actually feels like the limitless pill, just an astounding clarity on how to do things”</p><p>“I’m losing my craving for sweets, more energy, hair better etc”</p><p>“Immediately after closing my eyes at night, I can see detailed visualizations. Where before, my visualization skills were very sub par and very blurry.”</p><p>“I wake up without the morning puff of the face, looking real chiselled”</p><p>“Alcohol with blessed water is a whole different thing absolutely. It feels really really good. I drank last night and just two drinks had me feeling way more.”</p><p>(Blessed water is what the original discoverer called it)</p><p>If you’re sold, I got bad news: I’m not selling any until I get the potency up (more kilovolts, better methodology) and the process is fully automated.</p><p>Good news is, you can buy some from Ngolo right now, though. Go to wizard (dot) talk to do so.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4bf113053f5e" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Abiotic Oil]]></title>
            <link>https://medium.com/@filip.milner/abiotic-oil-3f5d67c0928a?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/3f5d67c0928a</guid>
            <category><![CDATA[renewable-energy]]></category>
            <category><![CDATA[conspiracy-theories]]></category>
            <category><![CDATA[fossil-fuels]]></category>
            <category><![CDATA[energy]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Tue, 27 Oct 2026 14:23:58 GMT</pubDate>
            <atom:updated>2026-10-27T14:23:58.049Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BQwryc_Wg4Ly0f224IMaYw.png" /></figure><p>At school, you were probably taught that oil comes from ancient life. Mostly plankton, buried under sediment and gradually converted into petroleum. That process is real. The Earth can also make hydrocarbons from minerals, water and hydrogen, with no dead organisms in the reaction. Researchers have demonstrated this in pressure chambers, and geologists have found evidence of it in natural rocks. Both routes need to be considered when we ask where petroleum comes from.</p><p>My conclusion is that deep inorganic reactions supply a substantial part of the oil and gas entering most, if not all, producing fields. That is the abiotic explanation: hydrocarbons forming without biological material as their starting ingredient. I think we have seriously underestimated this source. If petroleum continues to form and move towards the reservoirs we extract it from, the quantity initially found in a reservoir cannot describe everything that reservoir will ever supply.</p><p>Renat Muslimov, who served for more than three decades as Tatneft’s chief geologist, reached a particularly strong conclusion. In a September 2019 interview, the former chief geologist and deputy general director said:</p><p>&gt; “Oil, in fact, is endless. It is a renewable energy source.”</p><p>He was arguing that geological processes continue to supply oil, with practical consequences for how long a field can produce and where geologists should look for more. Whether that supply can keep up with extraction is a question of rates. Even a reservoir receiving oil every day can decline if its wells remove oil faster.</p><p>Oil entering a reservoir today may have formed long ago somewhere else. Replenishment establishes that the reservoir has an incoming supply. To establish an abiotic contribution, we also need evidence about the chemistry that produced that supply. I think the field observations and the experimental chemistry belong in the same explanation.</p><p>At Eugene Island 330, off Louisiana, oil is produced from sandstone layers beneath the seabed. The oil occupies tiny spaces between sand grains. Where those spaces connect, fluids can move through the rock towards a well. A producing reservoir is therefore a volume of permeable rock, and it can be connected to other rock beyond the area originally included in its reserves.</p><p>Researchers examining Eugene Island compared fluids sampled from the same wells four years apart. They found a pronounced increase in gaseous hydrocarbons relative to the somewhat larger molecules in the gasoline range. The gases also had chemical characteristics indicating greater exposure to heat than the reservoir oils. The researchers interpreted this as an influx of gas from deeper, hotter formations. Sampling the same wells at different times allowed them to identify changes in the fluids entering the producing system.</p><p>In 1993, researchers tested a suspected supply route directly. They extended an existing well into a fault beneath the producing layers. A fault is a fracture along which rock has moved, and the fractured zone can provide a passage for fluids. Oil and gas were recovered from that zone. The oil’s chemistry matched oil in the reservoirs above. Petroleum was present inside the route through which the researchers proposed it was entering the field.</p><p>Another team compared seismic surveys made in 1985 and 1992. These surveys use sound waves and their echoes to examine underground structures. A feature interpreted as a pulse of fluid had moved about one kilometre upwards along a fault between the surveys. The finding was published in Nature in 2005. The movement occurred on a timescale of years, during the field’s producing life.</p><p>The samples and seismic observations describe a connected system in which fluids move upwards during production. Some of what enters a producing reservoir was outside it when the initial inventory was estimated. To estimate the field’s future production, geologists need to understand this incoming flow as well as the oil already present.</p><p>The conventional explanation attributes the Eugene Island oil to organic matter in deeper sedimentary rocks. Researchers proposed Jurassic and Early Cretaceous marls, now roughly 12–14 kilometres underground, using the fluids’ chemistry and the basin’s geological history. That is a reconstruction of the oil’s ultimate origin. The field observations establish active movement towards the producing reservoirs, while the source interpretation concerns what happened before the petroleum entered that route.</p><p>For an operator, the amount arriving is as important as the existence of a route. At Shebelinka, a large gas field in Ukraine, researchers estimated an incoming supply large enough to compensate for substantial annual extraction. They used the relationship between gas production and reservoir pressure to investigate it.</p><p>Remove gas from a fixed volume at a steady temperature, and its pressure falls. Underground, the available space can also change. Water can enter a reservoir and occupy space previously filled by gas, while the rock itself can compact. These effects must be included before stable pressure can be attributed to incoming gas. The Shebelinka researchers examined the water system and the other factors affecting pressure, then concluded that gas entering through faults helped explain the field’s behaviour.</p><p>During 1999–2003, production was about 1.8–1.9 billion cubic metres a year and reservoir pressure stabilised. From 2004 to 2016, extraction was around 2.4 billion cubic metres a year and pressure declined again. The researchers estimated that inflow could compensate for withdrawal at the lower rate. On that interpretation, the incoming supply was comparable to an entire year’s production. Removing gas faster than it arrived reduced the amount remaining in the reservoir.</p><p>The estimated inflow would extend the field’s producing life and give existing wells access to gas the original inventory did not include. An operator would need to establish how long such a supply persists before planning annual production around it. The difference between the lower and higher withdrawal rates shows why replenishment can be commercially important while the field remains vulnerable to falling pressure.</p><p>The proposed source of that incoming gas includes deeper formations where gas is present in rock with poorly connected pores. Such rock can be difficult to produce directly through a well. Over a large area, however, small flows can converge into faults and reach a more permeable reservoir above. Falling pressure in the producing reservoir encourages movement towards it. Gas that could not readily be extracted where it was first encountered can therefore become accessible through another formation. A 2025 study also examined how replenishment changes Shebelinka’s production forecasts.</p><p>Russian oilfield research provides a longer view of the same problem. At Romashkino in Tatarstan, specialists working for Tatneft were already discussing replenishment in a 2005 conference presentation. Their title described the field as an “inexhaustible source to add oil reserves”. They reported growing reserve estimates despite continued extraction, unusually productive wells, and investigations of fractures in the crystalline basement. Basement is the older foundation of hard rock beneath the sedimentary layers. Proposed explanations included errors in earlier reserve estimates and oil entering through deep faults. They were using drilling, seismic surveys and fluid chemistry to investigate those possibilities.</p><p>Muslimov and Irina Plotnikova later described observations extending beyond changes to the reserve figures. Monitoring wells at Romashkino recorded variations in dissolved gases over more than ten years. Deep wells also showed changes in gas and water chemistry. The authors related some changes in oil density and water composition to seismic activity, interpreting them as evidence that fluid exchange changes when fractures open or become better connected. In their account, the deeper rock remains involved in the field’s behaviour during production.</p><p>An upward revision of reserves can have several explanations. Geologists estimate the total oil present in a formation, called oil in place, and the smaller portion they expect to recover. Better wells can increase the recoverable portion. Improved mapping can reveal oil omitted from the first estimate. Oil can also enter from outside the volume originally counted. These possibilities have different consequences for future production, so they need to be investigated separately.</p><p>In a regional analysis presented in 2018 and developed in a 2019 study, researchers reported 69 Volga–Urals fields where cumulative production had exceeded the initial estimates of oil in place. They considered errors in the estimates, contributions from poorly permeable rock omitted from the accounting, and supply from deeper formations. Their preferred explanation involved deep inflow through faults and fractures, supported by the regional geological and oil-property patterns they examined.</p><p>Once production exceeds the original estimate of all the oil present, improved extraction efficiency alone cannot explain the discrepancy. The original volume was underestimated, oil entered from elsewhere, or both occurred. The repeated observations of deep fluid movement make the second possibility worth investigating seriously. I agree with these researchers that continuing inflow helps explain the production history. Treating every discrepancy as an accounting revision risks overlooking a source that could still be supplying the field.</p><p>Muslimov also questioned whether the proposed organic source could account for the quantity already produced in Tatarstan. The Domanik formations are organic-rich sedimentary rocks commonly considered a source of the region’s petroleum. In a 2019 paper, he cited a geochemical estimate of 709 million tonnes of oil-source material, against more than 3.5 billion tonnes of oil already extracted. The production total was nearly five times that estimate.</p><p>The 709-million-tonne figure is an estimate of source material, so it should not be treated as an independently measured ceiling on every possible sedimentary source. Muslimov’s objection is nevertheless concrete. A proposed source must contain enough suitable material to generate the oil attributed to it. If that estimate is too low, the additional source has to be identified and supported. Chemical resemblance between oil and a particular rock cannot supply a missing quantity of material.</p><p>These field studies explain why I take continuing deep supply seriously. They also bring us to the origin question. Hydrocarbons consist of carbon and hydrogen. Methane, the main component of natural gas, contains one carbon atom per molecule. Liquid oil contains many larger molecules. An explanation of abiotic petroleum therefore needs chemistry capable of producing those larger compounds, as well as methane.</p><p>In a 2002 experiment, researchers put iron oxide, calcium carbonate and water into a pressure chamber. Calcium carbonate is the mineral in marble. It supplied carbon, while water supplied hydrogen. At 1,500°C and a pressure roughly 50,000 times atmospheric pressure, the mixture produced methane and larger hydrocarbons, including molecules with up to ten carbon atoms. The conditions were relevant to the mantle, the thick rock layer beneath Earth’s crust. No biological material had been added.</p><p>Carbon from a mineral became part of molecules found in petroleum. A plankton cell was unnecessary for that conversion. The experiment supplies a mechanism through which inorganic material can become hydrocarbons under conditions found inside the Earth.</p><p>Researchers demonstrated a related step in 2009 by starting with methane. At high pressures and temperatures relevant to the upper mantle, they produced ethane, propane and butane, with two, three and four carbon atoms respectively. Methane can therefore participate in reactions that produce larger hydrocarbons. If inorganic methane forms at depth, there is experimentally demonstrated chemistry through which some of its carbon can enter other petroleum compounds.</p><p>Still larger molecules were produced in a separate study published in 2001. Researchers heated water containing oxalic acid, a simple compound supplying carbon and hydrogen, for two days at each of several temperatures. Above 150°C, the products included compounds containing twelve to more than thirty-three carbon atoms. Among them were alkanes, a group of hydrocarbons found in crude oil. Oxygen-containing compounds formed as well. These experiments extend the chemistry beyond methane to carbon chains of lengths relevant to oil and waxes. Simple chemical ingredients can combine into much larger molecules without buried biological material.</p><p>An Edinburgh-led team demonstrated another route in 2021 using hydrogen and either diamond or graphite. Both minerals consist of carbon. In one experiment, hydrogen reacted with diamond at 300°C and roughly 30,000 times atmospheric pressure. Methane appeared and increased during about two hours. At higher temperatures, ethane was also detected. The researchers monitored the reactions as they occurred and performed controls to examine whether contamination could explain the products.</p><p>Two hours is enough for methane to form once those ingredients and conditions are present. Producing the molecule does not inherently require millions of years. Underground, the supply of hydrogen and carbon, the available reaction surfaces, and the movement of the products will determine how much accumulates. A fast reaction can still produce a small total if one ingredient arrives slowly. That distinction allows us to understand both the laboratory result and the separate question of supply to an oilfield.</p><p>Evidence of this chemistry also occurs in natural rocks. In the Italian Alps, geologists examined rocks that had been buried deep inside the Earth and later brought back towards the surface. Minerals and microscopic pockets of trapped fluid preserved evidence of their earlier conditions. In 2017, the researchers reported repeated formation of abiotic methane through reactions between hydrogen-rich fluids and carbonate minerals, at depths ranging from roughly 40 kilometres to 15–20 kilometres. The rock record supplied evidence of natural reactions comparable to the chemistry investigated in laboratories.</p><p>At Chimaera in Turkey, abiotic gas can be studied as it escapes at the surface. Flames burn where gas emerges through fractured rock near the coast. Researchers found a mixture containing about 87 percent methane and 10 percent hydrogen, and identified the methane as predominantly abiotic from its chemical and isotopic characteristics. Isotopes are atoms of the same element with different masses. Their proportions can help distinguish the processes that formed a gas.</p><p>The researchers estimated that at least 190 tonnes of methane escaped annually, including gas seeping through the ground beyond the visible flames. The site has a history of burning extending back about two thousand years. They interpreted the sustained discharge as evidence of a pressurised reservoir at depth. Its remaining reserves had not been measured. Old gas stored underground can continue escaping without being newly made, so the annual release does not measure current formation. Chimaera does demonstrate that natural abiotic gas has accumulated underground and escapes in measurable quantities.</p><p>The biological route has experimental support too. In a 1997 study, organic-rich rocks heated with water at 330–350°C for 72 hours produced oil resembling natural crude. Michael Lewan also tested isolated kerogen, the organic material in the rock. Water affected the reactions and supplied hydrogen. Biological material can become oil under suitable conditions, and an explanation that includes abiotic production has no need to deny it.</p><p>What remains unresolved is the contribution of each route to the petroleum reaching particular wells. Laboratory synthesis establishes that a reaction can occur. Geological evidence establishes that it has occurred in nature. Assigning a proportion of an oilfield’s production to that reaction requires tracing the incoming material. I am convinced that the abiotic contribution is substantial and widespread. The experiments and natural occurrences explain why I consider it a serious source, while the field studies explain why the deeper supply remains relevant during extraction.</p><p>The strongest evidence usually offered for biological origin is the biological material in the oil itself. Some molecules retain structures inherited from compounds made by organisms. These are called biomarkers. A close match between biomarkers in an oil and those in an organic-rich rock can help identify the oil’s source.</p><p>The complication is that oil can dissolve organic compounds from rocks it passes through. A sample collected at a well can contain molecules present when the oil formed and molecules acquired later. Geochemists call this migration contamination or reservoir overprinting. The distinction matters whenever a molecular resemblance is used to reconstruct the history of the whole fluid.</p><p>In a Gulf of Mexico example discussed by geochemist Joseph Curiale, roughly half of one group of biological marker molecules in a produced sample was estimated to have come from organic matter in the reservoir rock. That percentage concerned the marker group, rather than half the oil. Oil had acquired compounds from the rock containing it, changing a chemical signal used to assess its history. There is therefore an observed process through which biological molecules can enter an oil after its formation.</p><p>Mixing creates another problem for interpretation. Suppose two incoming oils contain very different concentrations of a particular marker. A small contribution from the oil rich in that marker can account for much of the marker measured in the mixture. To calculate how much of the total oil came from either source, we need the marker concentrations in the starting fluids and a model of what happened during mixing. Recognising the marker’s biological ancestry alone does not provide that calculation.</p><p>Petroleum geochemists address such problems by comparing several kinds of evidence, including different molecular groups, isotope ratios and the geological conditions needed for the proposed source to generate and release oil. Curiale’s 2008 review required a causal relationship between the proposed source and the oil. Chemical similarity had to be supported by a geological explanation of how that source supplied the accumulation. He also discussed weaknesses in published correlations, including limited sampling and interpretations extended over large areas.</p><p>A well-supported correlation can establish a biological contribution. The extent of that contribution still has to be assessed. In the mixed origin I favour for much of the world’s petroleum, inorganic hydrocarbons rise through the crust, acquire compounds from surrounding rocks, and mix with petroleum generated from biological material. The sample reaching a well then has a real biological signature. That history is compatible with a substantial abiotic contribution to the mixture. For some accumulations, I think the deeper inorganic supply may predominate.</p><p>Vietnam’s Bach Ho field brings several of these questions together. Also called White Tiger, it produces oil from fractured granite beneath the seabed. Granite forms from molten rock and contains very little connected pore space within its intact mass. At Bach Ho, oil occupies fractures and cavities. The fractures make it possible to store and extract a large quantity of oil from rock that would otherwise be a poor reservoir.</p><p>A 2020 assessment put Bach Ho’s recoverable oil at approximately 1.0–1.4 billion barrels and reported that about 95 percent of its production came from the basement reservoir. This was a major commercial accumulation in crystalline rock beneath the sedimentary layers. Its size makes the route by which oil entered the granite worth understanding.</p><p>The discovery history shows how expectations can affect whether such a reservoir is tested. In an account published by company leaders in 2021, a drilling chief recalled that drilling fluid had disappeared into the granite during work in the 1980s. That loss indicated open fractures capable of accepting fluid. The crew wanted to drill farther and test the basement, but management refused, working from the belief that the granite would not contain oil. Successful testing of the basement followed in 1988.</p><p>The missed opportunity was specific. Fractures had been encountered in rock beneath the expected reservoirs, and they provided space through which fluids could move. Rejecting that rock as a possible target delayed testing an accumulation that became commercially important. An assumption about where oil should occur had influenced which evidence the operator was willing to investigate.</p><p>The granite also raises the question of continuing supply. In a detailed 2009 study, geologists reported near-constant production rates for more than five years despite little storage in the intact rock between fractures. They interpreted that behaviour as evidence of significant recharge into the fracture network. A productive fracture system can receive oil from a larger connected volume while wells remove it. The initially accessible oil need not account for the full production history.</p><p>Those authors proposed a sedimentary source for the oil. In their reconstruction, fault movement had placed granite over organic-rich mudstones, allowing oil to enter the fractured basement. They also reported that production and pressure observations had not identified present communication between the basement and adjacent sandstone reservoirs. Current separation from those sandstones does not establish separation from every possible source throughout the field’s history. Their reconstruction illustrates how oil generated in sedimentary rock could reach a granite reservoir.</p><p>There is an additional observation concerning the water in the field. Tatiana Kireeva examined a sample from about 4.5 kilometres deep in an isolated basement block, where injected seawater had not entered. The water differed chemically from both the injected seawater and the waters of the overlying sedimentary formations. Its low salinity and unusually high boron content led her to interpret it as hydrothermal water, associated with the movement of hot fluids through the rock.</p><p>Kireeva argued that the water chemistry complicates a simple sedimentary explanation of the reservoir. Water and oil need not share exactly the same origin, so the water analysis cannot identify the oil’s carbon source by itself. It does provide another reason to investigate deep fluid movement and the reactions that altered the granite. The reservoir’s history includes the fluids that created or modified its storage space, as well as the oil that eventually occupied it.</p><p>I consider Bach Ho a strong argument for examining the crystalline basement and its connections to deeper rock. We have a commercially large accumulation, fractures capable of conducting fluids, production interpreted as requiring recharge, and water associated with hydrothermal processes. Oil in granite alone cannot decide between organic and inorganic formation, because oil can migrate into granite from elsewhere. The useful question is where the incoming fluids formed and what proportions each source supplied. That question remains relevant even after a sedimentary source has been proposed.</p><p>Temperature is another reason to examine assumptions about deeper oil. Heating can break large hydrocarbon molecules into smaller ones, a process called cracking. Given sufficient time and suitable conditions, liquid oil can be converted into gas and other products. Geologists therefore use the thermal history of a basin to estimate where oil can remain preserved.</p><p>A 2020 study examined oils from 48 wells, at depths of roughly 6.5–8 kilometres, in China’s Tarim Basin. From the samples and their thermal characteristics, the researchers estimated that the disappearance of liquid oil through cracking occurred near 210°C in that setting. Substantial oil could remain below that threshold even at great depth. The finding concerns the preservation of oil, and does not identify an abiotic origin. It does show why a simple temperature or depth limit can be a poor reason to dismiss deeper exploration.</p><p>At depth, hydrocarbons can form, move, mix and break down, with the outcome depending on the local conditions and how long the material remains there. Oil recovered from a deep well establishes that liquid petroleum reached and persisted in that formation. To explain its origin, we need the source chemistry and the migration history as well. Greater depth makes those histories harder to investigate, which is why assumptions about them should remain open to testing.</p><p>The explanation I favour connects the processes described throughout this essay. Inorganic reactions generate hydrocarbons at depth. Fractures and pressure differences allow some of those products to move upwards. Along the way, they can mix with older petroleum, including oil generated from biological material, and acquire compounds from surrounding rocks. Some reach accumulations that we can drill. Extraction lowers pressure and changes the movement of fluids through the connected rock, so the supplying system can continue to affect production after a field has been developed.</p><p>That explanation accounts for why I expect mixed origins in many fields, why biological molecules can occur in oil with an inorganic contribution, and why the quantity extracted can exceed an early estimate of the accumulation. Each part can be investigated. The unresolved part is how much of the incoming supply comes from each source in a particular field. I think the abiotic share is much larger than generally assumed. Treating it as negligible in advance would exclude one of the processes we should be trying to measure.</p><p>For a field operator, the immediate consequence is the need to understand inflow. Reserve estimates describe the oil or gas counted under specified geological and operating assumptions. They need to be revised when better information becomes available. If a reservoir receives a continuing supply, its future also depends on the size and persistence of that flow. Pressure measurements, repeated fluid sampling and investigation of the connecting faults can help distinguish an initially underestimated accumulation from one receiving additional petroleum.</p><p>Shebelinka illustrates why estimating the rate is useful. The proposed inflow of roughly 1.8–1.9 billion cubic metres a year would support substantial production while leaving much less to be drawn from the existing accumulation. Extraction above that rate reduces the stored amount more quickly. A claim of renewability becomes operationally useful when it includes how much arrives, where it enters, and whether that supply persists. Those are questions an operator can investigate over time.</p><p>The consequences extend beyond the oilfield. Fuel costs affect transport, manufacturing and the food delivered to your shop. Expectations about future supply influence investment in exploration and infrastructure. If geological processes provide a continuing contribution, estimates of future availability should account for it. The original quantity in a producing reservoir gives us only part of the information needed.</p><p>Muslimov understood that continuing supply would not remove the costs of access. In the same 2019 interview in which he described oil as renewable, he discussed the growing difficulty and expense of extraction. He also argued that recognition of renewability would be financially unwelcome to oil companies and producing states because of its effect on prices. The two claims are compatible. The Earth can contain or produce hydrocarbons that remain expensive to reach and difficult to recover.</p><p>I share his suspicion that companies and states benefiting from high prices have little incentive to encourage expectations of greater abundance. That is a judgment about incentives, rather than evidence of the oil’s origin. The geological reasons for questioning an exclusively fossil account are more concrete. Inorganic hydrocarbon formation has been demonstrated, natural abiotic gas exists, producing reservoirs can receive deep inflow, and biological signatures can be modified during migration and mixing.</p><p>Taken together, I find this a strong case for a substantial deep abiotic contribution to petroleum. It is also a reason to take the rock beneath known reservoirs seriously, as the history of Bach Ho makes clear. An early reserve estimate describes what geologists thought had accumulated within a particular volume of rock. Understanding the field’s future requires finding out what can still enter it, and understanding the deeper source requires finding out what the Earth continues to make.</p><p>Research kit:</p><p>**S01. Renat Muslimov interview, 2019.** [Realnoe Vremya, English edition, 10 September](<a href="https://realnoevremya.com/articles/3786-renat-muslimov-on-challenges-facing-oil-industry">https://realnoevremya.com/articles/3786-renat-muslimov-on-challenges-facing-oil-industry</a>). [Russian original, 7 September](<a href="https://realnoevremya.ru/articles/150653-renat-muslimov-o-vyzovah-stoyaschih-pered-neftyanoy-otraslyu">https://realnoevremya.ru/articles/150653-renat-muslimov-o-vyzovah-stoyaschih-pered-neftyanoy-otraslyu</a>).</p><p>**S02. Tatarstan Academy of Sciences, Muslimov biography.** [Official biography](<a href="https://www.antat.ru/ru/staff/3473/">https://www.antat.ru/ru/staff/3473/</a>).</p><p>**S03. Schumacher, D. (1993). Eugene Island Block 330 Field, Offshore Louisiana: Geochemical Evidence for Active Hydrocarbon Recharging.** AAPG Annual Convention abstract. [Primary abstract](<a href="https://www.searchanddiscovery.com/abstracts/html/1993/annual/abstracts/0179a.htm">https://www.searchanddiscovery.com/abstracts/html/1993/annual/abstracts/0179a.htm</a>).</p><p>**S04. Anderson, R.N., Cathles, L., III, and Nunn, J. (1994). Drilling into a Present Day Migration Pathway for Hydrocarbons within a Fault Zone Conduit in the Eugene Island 330 Field of Offshore Louisiana.** AAPG Annual Convention abstract. [Primary abstract](<a href="https://www.searchanddiscovery.com/abstracts/html/1994/annual/abstracts/0093c.htm">https://www.searchanddiscovery.com/abstracts/html/1994/annual/abstracts/0093c.htm</a>).</p><p>**S05. Haney, M.M., Snieder, R., Sheiman, J., and Losh, S. (2005). A moving fluid pulse in a fault zone.** Nature 437, 46. [Article](<a href="https://doi.org/10.1038/437046a">https://doi.org/10.1038/437046a</a>).</p><p>**S06. Losh, S., and colleagues (2002). Reservoir fluids and their migration into the South Eugene Island Block 330 reservoirs, offshore Louisiana.** AAPG Bulletin 86(8), 1463–1488. [Publisher record and article text](<a href="https://archives.datapages.com/data/bulletns/2002/08aug/1463/1463.htm">https://archives.datapages.com/data/bulletns/2002/08aug/1463/1463.htm</a>).</p><p>**S07. Fyk, I.M., Fyk, M.I., and Fyk, I.M. (2019). Prospects for long-term development of the Shebelinka gas-condensate field under conditions of reserve replenishment.** Bulletin of V.N. Karazin Kharkiv National University, Geology, Geography, Ecology 50, 63–76. [Journal page](<a href="https://journals.uran.ua/geoeco/article/view/204869">https://journals.uran.ua/geoeco/article/view/204869</a>). [Full article](<a href="https://periodicals.karazin.ua/geoeco/article/download/13303/12541/">https://periodicals.karazin.ua/geoeco/article/download/13303/12541/</a>). DOI: 10.26565/2410–7360–2019–50–05.</p><p>**S08. Fyk and colleagues (2025). The impact of gas reserves recovery on the technological and economic performance of gas condensate field minig.** IOP Conference Series: Earth and Environmental Science 1491, 012069. [DOI](<a href="https://doi.org/10.1088/1755-1315/1491/1/012069">https://doi.org/10.1088/1755-1315/1491/1/012069</a>). [Author/publication record with abstract](<a href="https://www.researchgate.net/publication/391502969_The_impact_of_gas_reserves_recovery_on_the_technological_and_economic_performance_of_gas_condensate_field_minig">https://www.researchgate.net/publication/391502969_The_impact_of_gas_reserves_recovery_on_the_technological_and_economic_performance_of_gas_condensate_field_minig</a>).</p><p>**S09. Khisamov, R.S., and Bazarevskaya, V.G. (2005). Unique Romashkino Oil Field in Tatarstan Republic is an Inexhaustible Source to Add Oil Reserves.** AAPG International Conference, Paris. [Primary abstract](<a href="https://www.searchanddiscovery.com/documents/abstracts/2005intl_paris/khisamov.htm">https://www.searchanddiscovery.com/documents/abstracts/2005intl_paris/khisamov.htm</a>).</p><p>**S10. Muslimov, R.Kh., and Plotnikova, I.N. (2019). Replenishment of oil deposits in the light of a new concept of oil and gas formation.** Georesursy 21(4), 40–48. [DOI](<a href="https://doi.org/10.18599/grs.2019.4.40-48">https://doi.org/10.18599/grs.2019.4.40-48</a>). [Accessible English article](<a href="https://pdfs.semanticscholar.org/1784/0d3bd7391fff1ed0efe48b88bdba4391635d.pdf">https://pdfs.semanticscholar.org/1784/0d3bd7391fff1ed0efe48b88bdba4391635d.pdf</a>).</p><p>**S11. Goryunov, E.Yu., Trofimov, V.A., Nesterova, A.E., and Chesalova, E.I. (2019). Natural replenishment of oilfields reserves in the Volga-Urals Petroleum Province.** Geology of Oil and Gas 4, 17–26. [DOI](<a href="https://doi.org/10.31087/0016-7894-2019-4-17-26">https://doi.org/10.31087/0016-7894-2019-4-17-26</a>). [Indexed article](<a href="https://cyberleninka.ru/article/n/estestvennoe-vospolnenie-zapasov-neftyanyh-mestorozhdeniy-volgo-uralskoy-neftegazonosnoy-provintsii.pdf">https://cyberleninka.ru/article/n/estestvennoe-vospolnenie-zapasov-neftyanyh-mestorozhdeniy-volgo-uralskoy-neftegazonosnoy-provintsii.pdf</a>).</p><p>**S12. Muslimov, R.Kh. (2019). An in-depth study of the crystalline basement of sedimentary basins is a dictate of the time.** Georesursy 21(4), 55–62. [DOI](<a href="https://doi.org/10.18599/grs.2019.4.55-62">https://doi.org/10.18599/grs.2019.4.55-62</a>). [Russian article](<a href="https://old.geors.ru/media/pdf/07Muslimov.pdf">https://old.geors.ru/media/pdf/07Muslimov.pdf</a>). [English article mirror](<a href="https://pdfs.semanticscholar.org/966b/4aa95069afae327f0962a88df3ee60208cbd.pdf">https://pdfs.semanticscholar.org/966b/4aa95069afae327f0962a88df3ee60208cbd.pdf</a>).</p><p>**S13. Kenney, J.F., and colleagues (2002). The evolution of multicomponent systems at high pressures: VI. The thermodynamic stability of the hydrogen–carbon system: The genesis of hydrocarbons and the origin of petroleum.** PNAS. [Article](<a href="https://doi.org/10.1073/pnas.172376899">https://doi.org/10.1073/pnas.172376899</a>). [PubMed record](<a href="https://pubmed.ncbi.nlm.nih.gov/12177438/">https://pubmed.ncbi.nlm.nih.gov/12177438/</a>).</p><p>**S14. Kolesnikov, A., Kutcherov, V.G., and Goncharov, A.F. (2009). Methane-derived hydrocarbons produced under upper-mantle conditions.** Nature Geoscience 2, 566–570. [Article](<a href="https://doi.org/10.1038/ngeo591">https://doi.org/10.1038/ngeo591</a>). [Full text hosted at Harvard](<a href="https://courses.seas.harvard.edu/climate/eli/Courses/EPS281r/Sources/Origin-of-oil/Kolesnikov%202009.pdf">https://courses.seas.harvard.edu/climate/eli/Courses/EPS281r/Sources/Origin-of-oil/Kolesnikov%202009.pdf</a>).</p><p>**S15. Rushdi, A.I., and Simoneit, B.R.T. (2001). Lipid formation by aqueous Fischer-Tropsch-type synthesis over a temperature range of 100 to 400°C.** Origins of Life and Evolution of the Biosphere 31, 103–118. [Primary abstract](<a href="https://pubmed.ncbi.nlm.nih.gov/11296515/">https://pubmed.ncbi.nlm.nih.gov/11296515/</a>). DOI: 10.1023/A:1006702503954.</p><p>**S16. Peña-Alvarez, M., and colleagues (2021). In-situ abiogenic methane synthesis from diamond and graphite under geologically relevant conditions.** Nature Communications. [Full article](<a href="https://www.nature.com/articles/s41467-021-26664-3">https://www.nature.com/articles/s41467-021-26664-3</a>).</p><p>**S17. Vitale Brovarone, A., and colleagues (2017). Massive production of abiotic methane during subduction evidenced in metamorphosed ophicarbonates from the Italian Alps.** Nature Communications 8, 14134. [Article](<a href="https://www.nature.com/articles/ncomms14134">https://www.nature.com/articles/ncomms14134</a>).</p><p>**S18. Etiope, G., Hosgörmez, H., and Schoell, M. (2014). Chimaera, a Large Abiotic Gas Seep in Turkey.** AAPG International Conference, Istanbul. [Primary conference abstract](<a href="https://www.searchanddiscovery.com/abstracts/html/2014/90194ice/abstracts/1935350.html">https://www.searchanddiscovery.com/abstracts/html/2014/90194ice/abstracts/1935350.html</a>).</p><p>**S19. Lewan, M.D. (1997). Experiments on the role of water in petroleum formation.** Geochimica et Cosmochimica Acta 61(17), 3691–3723. [USGS publication record and abstract](<a href="https://pubs.usgs.gov/publication/70019459">https://pubs.usgs.gov/publication/70019459</a>). [Full article hosted at WHOI](<a href="https://website.whoi.edu/gfd/wp-content/uploads/sites/14/2018/10/GCARoleofWater_146986.pdf">https://website.whoi.edu/gfd/wp-content/uploads/sites/14/2018/10/GCARoleofWater_146986.pdf</a>).</p><p>**S20. Curiale, J.A. (2002). A review of the occurrences and causes of migration-contamination in crude oil.** Organic Geochemistry 33(12), 1389–1400. [Article text and case excerpts](<a href="https://www.sciencedirect.com/science/article/abs/pii/S0146638002001092">https://www.sciencedirect.com/science/article/abs/pii/S0146638002001092</a>). DOI: 10.1016/S0146–6380(02)00109–2.</p><p>**S21. Curiale, J.A. (2008). Oil–source rock correlations: Limitations and recommendations.** Organic Geochemistry. [Article](<a href="https://doi.org/10.1016/j.orggeochem.2008.02.001">https://doi.org/10.1016/j.orggeochem.2008.02.001</a>).</p><p>**S22. Koning, T. (2020). Exploring in Asia for Oil and Gas in Naturally Fractured and Weathered Basement Reservoirs.** GeoConvention. [Full conference paper](<a href="https://geoconvention.com/wp-content/uploads/abstracts/2020/57981-exploring-in-asia-for-oil-and-gas-in-naturally-fra.pdf">https://geoconvention.com/wp-content/uploads/abstracts/2020/57981-exploring-in-asia-for-oil-and-gas-in-naturally-fra.pdf</a>).</p><p>**S23. Kudryashov, S., Akimov, O., Mikhailov, A., and Boldyrev, A. (2021). 40 years long road.** Energy Policy magazine, 20 August. [Company leaders’ historical account](<a href="https://en.energypolicy.ru/40-years-long-road/08/20/">https://en.energypolicy.ru/40-years-long-road/08/20/</a>).</p><p>**S24. Cuong, T.X., and Warren, J.K. (2009). Bach Ho Field, a fractured granitic basement reservoir, Cuu Long Basin, offshore SE Vietnam: a “buried-hill” play.** Journal of Petroleum Geology 32(2), 129–156. [Full paper](<a href="https://www.saltworkconsultants.com/downloads/Bach_Ho_Cuong_Warren_09.pdf">https://www.saltworkconsultants.com/downloads/Bach_Ho_Cuong_Warren_09.pdf</a>). [Journal abstract](<a href="https://www.jpg.co.uk/ab_apr_09_2.htm">https://www.jpg.co.uk/ab_apr_09_2.htm</a>).</p><p>**S25. Kireeva, T.A. (2020). Water-free oil reservoirs: origin and morphology.** Georesursy 22(4), 15–21. [Article](<a href="https://doi.org/10.18599/grs.2020.4.15-21">https://doi.org/10.18599/grs.2020.4.15-21</a>). [English PDF](<a href="https://old.geors.ru/media/pdf/02_Kireeva_en.pdf">https://old.geors.ru/media/pdf/02_Kireeva_en.pdf</a>).</p><p>**S26. Zhu and colleagues (2020). Stability and cracking threshold depth of crude oil in 8000 m ultra-deep reservoir in the Tarim Basin.** Fuel 282, 118777. [Primary abstract](<a href="https://www.sciencedirect.com/science/article/pii/S0016236120317737">https://www.sciencedirect.com/science/article/pii/S0016236120317737</a>). [DOI](<a href="https://doi.org/10.1016/j.fuel.2020.118777">https://doi.org/10.1016/j.fuel.2020.118777</a>).</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3f5d67c0928a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[How to Grow Plants with Electricity]]></title>
            <link>https://medium.com/@filip.milner/how-to-grow-plants-with-electricity-2c1a80f5bf69?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/2c1a80f5bf69</guid>
            <category><![CDATA[agriculture]]></category>
            <category><![CDATA[plants]]></category>
            <category><![CDATA[biology]]></category>
            <category><![CDATA[science]]></category>
            <category><![CDATA[electroculture]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Sun, 25 Oct 2026 18:38:17 GMT</pubDate>
            <atom:updated>2026-10-25T18:38:17.906Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RjhBwzfctfovwAS--EJMrg.png" /></figure><p>Electroculture has been called pseudoscience for decades now. Even most electroculture enthusiasts are unaware that 3 different academic groups published controlled trials in 2025, and all 3 reported gains.</p><p>Likewise, most are unaware that the British government concluded the effect is real almost 100 years ago.</p><p>So what is electroculture? In short, it’s growing plants with electricity. The version sold in garden shops uses metal coils and rods to collect it passively, with nothing plugged in.</p><p>You push a wooden stake with a copper coil into the soil next to your tomatoes, and most people who walk past one think it is the same kind of nonsense as a crystal on a windowsill.</p><p>So let’s dive into the actual research.</p><p>The British government spent 18 years finding out whether electricity grows crops, and in 1937 its Electro-Culture Committee wrote in the final report that the effect is real. Then the committee shut the programme down, because the crop came in about 20% bigger and the machine that produced it cost more than that gain was worth: wires strung over the crop, charged by a petrol generator.</p><p>That 20% is a fifth more food off the same ground, with the same water and the same work you are already putting in. Definitely not something to scoff at.</p><p>The report’s key paragraph runs 3 sentences. The middle one is the sentence that gets quoted by critics:</p><p>&gt; “In spite of the failures of recent years, the field results obtained some years ago and the earlier pot culture results would seem to have established the fact that the Electro-Culture effect is a real one. It would seem, however, to be of little advantage to continue the work either on economic or on scientific grounds. Increases of 20 per cent can hardly be considered economic even if obtained in most years; experiments, however, demonstrate that the regular occurrence of the effect cannot be expected.”</p><p>Its field results are the committee’s outdoor trials, and its pot culture results are plants grown in pots. On its own that middle sentence looks like the committee found nothing. The whole paragraph says the effect is real, the gains were unreliable, and 20% more crop was not enough to make the apparatus pay. They had bad seasons behind them and yet still called the effect real.</p><p>Here’s the kicker: the British government never tested a device with no generator involved. The copper coil in the garden shop needs no generator. The 4 controlled trials published in 2025 tested passive copper devices, with untreated plants grown beside the treated ones for comparison.</p><p>None of it started with the British committee though.</p><p>Lemström, professor of physics at Helsinki, ran the same experiment and published his numbers in 1904, 14 years before the committee existed, and his table holds the largest increases anybody has published. He strung his wires over the plots, charged them from a generator, and left untreated plots beside the treated ones.</p><p>Lemström’s results:</p><p>- Barley: 23%<br>- Oats: 28.7%<br>- Potatoes: 50%<br>- Carrots: 92.7%<br>- Strawberries: 88.7%</p><p>The biggest numbers in Lemström’s table are not the grain crops, one reason to doubt 20% is a ceiling. The British committee investigated the field effect entirely with grains. Second biggest mistake on their part. The first being investigating only powered apparatus, rather than passive.</p><p>Blackman, the committee’s plant physiologist, published his summary: 18 experiments over 6 years and an average increase of 22% on spring-sown grain.</p><p>Lemström’s barley at 23% and his oats at 28.7% are the smallest two numbers in his own table. A Finnish professor working alone in 1904 and a British government committee two decades later came out within a few percentage points of each other on a grain crop.</p><p>Lemström’s wires and the committee’s charged wire network both ran on a generator. Neither became the thing a farmer could buy once, push into the ground and leave there.</p><p>“If a copper coil worked, agriculture would already be using it,” some might say.</p><p>A fertiliser company has a customer every spring, and that is what an industry gets built on. A stake with copper wire wound around it, pushed into the ground and wired to nothing, is bought once and never again. In 1918 the British government funded two different answers to the wartime food shortage, fertiliser and electroculture. Only one of them was a product farmers would buy every season.</p><p>That March, the Minister of Munitions was authorised to build a nitrogen factory at Billingham, on the north-east coast of England, and the government’s commitments there came to approximately 1,000,000 pounds sterling. The same year the Board of Agriculture set up the Electro-Culture Committee, and by 1924 its research grants had come to 8,000 pounds in total.</p><p>Those are not the same kind of number: one is capital committed to a factory, the other is grants paid to a research programme. Same government, same year, same question about feeding the country. Synthetic nitrogen got a factory. Electroculture got a committee.</p><p>Then, in 1924, the Minister of Agriculture told Parliament in writing that certain types of electrical discharge have a very marked influence on the development of plant growth. That is a serving minister, on the parliamentary record, stating as fact the thing gardeners now get laughed at for.</p><p>Britain built Billingham to pull nitrogen out of the air, because nitrogen out of the air is what wartime explosives were made from. In peacetime the same nitrogen came out as synthetic fertiliser, bought again every season. By 1927 one production unit there was making 46,000 tons of nitrogen a year, and every ton had to be sold to a farmer.</p><p>Against that, the committee had 8,000 pounds of grants by 1924, and out of a petrol generator and the charged wire network it got 20% more crop. That sum does not pay for the engineering it takes to make a machine better. With the factory’s kind of money, those same 18 years could have gone on the wires alone: how thick, how high, how much current, how often. My guess is that 20% is where the money ran out, not where the effect stops.</p><p>And somebody could have tested passive induction as well.</p><p>Instead, the money went into one kind of apparatus: thin wires strung about 2 metres above the crop, charged to between 40,000 and 80,000 volts by a petrol generator. Only one cost estimate for a working farm is in the record, about 10 pounds sterling an acre, paid once. The fuel cost is nowhere in the record at all, and the fuel is the bill that comes back every year.</p><p>The copper coil has none of that. No generator, no fuel to burn, nothing to pay for this year or in 20 years.</p><p>In its first report the committee wrote that it had confined its experiments to overhead discharge, its own term for the charged wires above the crop, and repeated that sentence in every report for 18 years.</p><p>It never once tested a piece of metal standing in the ground with nothing plugged into it.</p><p>Did anybody else test a device with no generator in it? The United States Department of Agriculture published a bulletin on electroculture in 1926 and told American farmers the practice was not recommended. The trials it reviewed used powered apparatus, up to a 100,000 volt transformer.</p><p>The US department’s own people gathered every British trial they could find into one table, 33 of them, and worked out the average crop increase themselves: 14%. That is the honest average with the failed seasons counted in, and it is still an increase. The bulletin calls an increase of that size promising, and the same department told American farmers not to bother.</p><p>One machine in that bulletin had no power source, and it gets a mention rather than a trial: a geomagnetifier, a lightning rod meant to gather the electricity in the open air and supply it to the crops. Three people are listed as having reported that it worked.</p><p>Not a single person at the department tested one.</p><p>Somebody was already selling that same kind of device. Christofleau, a French engineer, sold farmers one through the 1920s and 30s. His collector was a metal box on a post, with a bundle of metal points at the top and a wire running down into buried wire in the soil. It is called a collector because the points collect charge out of the air.</p><p>He filed it as a patent in 1925, and the patent says what the points do. Translated, his words are: “the antennae capture electricity from the ambient air and those from the induction of clouds, by virtue of the power of the points.” The antennae are that same bundle of points on top of the box. Induction here is electric charge pulled along a piece of metal by an electric field it is standing in, with nothing touching it and nothing plugged in.</p><p>When somebody tells you nobody has even proposed a mechanism for this, meaning an explanation of how a coil could actually work, what he means is that he has never read the one filed at a patent office in 1925 and public ever since.</p><p>Christofleau sold these for a living, so his book is advertising, and the only evidence in it is photographs: beans grown without manure standing under a collector, and a bean field split into a treated half and an untreated one, with no yield figures for either.</p><p>All of this rests on one claim: a piece of metal standing in open ground has a current going down it with no power source anywhere near it.</p><p>In fair weather, with no storm anywhere, there is an electric field between the ground and the open sky: 100 to 300 volts per metre of height. Climb a metre in open air and the voltage where your head ends up is a couple of hundred volts higher than where your feet were. That electric field is doing nothing to you, because the current is tiny. About 2 trillionths of an amp comes down onto every square metre of open ground, all day, wire or no wire. A trillionth is a millionth of a millionth.</p><p>Stand a piece of metal upright outdoors with its lower end in the earth, and the electric field pulls charge along it, and a trickle of current goes into the ground. There is no circuit to build. The current is already arriving everywhere, and metal is an easier path than air.</p><p>Put a point on top and more current comes down, because an electric field is much stronger at a sharp tip than at a flat surface. That is why a lightning rod is pointed, and why Christofleau’s box has a bundle of points on it.</p><p>Somebody with an engineering degree will say: “Induction is what happens inside a transformer, a changing magnetic field driving current round the turns of wire inside it, and a copper coil in a flowerbed picks up about a thousand times too little that way to matter to a plant.”</p><p>He is right about the transformer. That is the explanation critics choose to argue against, because it is the one that fails. The 1925 patent describes the other kind, the electrostatic one: charge moved along the metal by the electric field it is standing in, with no power supply involved.</p><p>Gardeners argue about how many turns to wind and which way round. If the electrostatic explanation is the right one, the turns and the direction should change almost nothing. How much of the coil sits down in the soil should change a great deal, and so should how high the rest of it stands. You can test the depth and the height yourself.</p><p>The government measured the dose itself. Blackman, the British committee’s plant physiologist, grew wheat, barley and maize in pots for 4 years, with a tiny current passed through the plants, and published the work in 1924. The plants came out heavier once dried and weighed, and dry weight — the plant with its water driven off — is the standard measure.</p><p>Those gains happened at currents as low as 10 trillionths of an amp per plant!</p><p>Maize gained 27% and barley up to 18%, off a current nobody could feel.</p><p>The dose here is a current, measured in amps, which is how much electricity is actually flowing. The 100 to 300 volts per metre of the electric field between ground and sky measures a different thing, how hard the electricity is being pushed, and it is not the dose.</p><p>There is a gap to account for though. A square metre of open ground receives about 2 trillionths of an amp. Blackman’s plants came out heavier on 10 trillionths. So a plant sitting on open soil would have to gather everything falling on 5 square metres around it. The question is therefore how much current goes down one piece of metal, and not how much lands on bare soil. The metal’s size and shape matter.</p><p>So how much current does a bare piece of metal in open ground actually pass?</p><p>An atmospheric observatory in southern England measures it: the current coming down an earthed metal point standing out in the open air — earthed meaning connected to the ground. In ordinary weather that current is between trillionths and billionths of an amp. It reaches a few millionths under a thunderstorm.</p><p>Plants in the 1924 pot experiments did better at currents between 10 trillionths and 10 billionths of an amp per plant. That is the range your coil would have to land in.</p><p>The current a passive piece of metal collects and the current that changes a plant can be THE SAME SIZE, in amps.</p><p>People who dismiss electroculture say the current is far too small to matter. Far too small for what? Not one of the sources quoted against the coil puts a number in that sentence.</p><p>There is a top to that range as well as a bottom. Above 10 billionths of an amp per plant, more current means less dried plant weight: the 1924 pot plants came out worse than the untreated ones. The current down a metal point swings with the weather, from trillionths to a few millionths, so the same coil could be inside the useful range one season and above it the next.</p><p>A coil could therefore cost you a crop, and nobody should pretend otherwise. Electroculture is a science that is yet to be properly researched and defined. Too little current may do nothing, and too much may do harm.</p><p>That could help explain why the historical crop results were erratic, and why in some seasons the treated crop came out smaller than the untreated one. The Electro-Culture Committee wrote the same thing in its eighteenth and last report, February 1937: “the regular occurrence of the effect cannot be expected”.</p><p>Blackman put the electrical energy at about a thousandth of what the plant takes from sunlight, or less. Whatever the current does, it is not directly feeding the plant.</p><p>Can an input that small really do anything to a living thing? Japanese farmers have said for generations that a lightning strike near a mushroom plantation is followed by a big crop, so researchers built a machine to imitate the strike.</p><p>The lightning machine fires a single pulse shaped like a lightning stroke through a needle electrode into logs seeded with mushroom spawn. Each pulse is 5.39 joules of stored energy, and if all of it turned into heat it would raise the log’s temperature by a few thousandths of a degree Celsius. So in terms of energy, certainly not enough to feed twice the crop. And yet…</p><p>Over 4 seasons the untreated logs produced 167 grams of shiitake mushrooms on average, per log, and the treated logs produced 322 grams. The researchers call the pulse an injury and the mushrooms growing a stress response, which may well be right. Their machine is powered, so it says nothing about what a passive coil collects. It does show a living thing answering an electrical input far too small to feed its growth.</p><p>Back to the coil. That trickle of current goes down into the soil, which is where I think the “magic” happens.</p><p>Soil conducts electricity, and there are bacteria in it that live by moving electrons. Some bacteria are able to take electrons straight off a piece of metal. Yes, there are bacteria that eat electricity. They’re called electrotrophs. Researchers gave these bacteria a powered metal electrode as their only source of electrons, no soil and no plant in the dish, and the bacteria ran those electrons through their own respiration, the process our cells run on the oxygen we breathe. They lived better on it, and died more slowly.</p><p>One of those bacteria runs on about 0.4 trillionths of an amp, on my arithmetic. A metal point in open air passes trillionths to billionths of an amp down into the ground. That is enough current for a bacterium’s meal, arriving out of the open sky.</p><p>Some soil bacteria also unlock nutrients for plants. Keep that in mind.</p><p>A gram of the soil in your garden holds about a billion bacterial cells, and a coil’s trickle could never reach all of them. It only has to reach the ones living on the metal and in the soil touching it.</p><p>One of the steps from the bacteria back to the plant has been published. Experimenters drove a weak current through the soil around rice roots from a power supply, so on its own that says nothing about a coil. The bacteria that eat electricity multiplied, and the rice came out 17% heavier above ground and 18% heavier below. Those numbers point the way my guess does.</p><p>My proposal is that the current off a copper coil reaches bacteria that can use it, they work faster, and what comes back to the plant is more of the nitrogen and phosphorus that were locked up in the soil. Nobody has tested that whole chain in a planted garden bed: the current the coil passes, the bacteria on its surface, the nutrients around it. Every instrument that measurement needs is standard equipment. It has never been tested, and it gets discussed as though somebody had tested it and it had failed.</p><p>The soil can supply a current of its own too. A laboratory jar run on soil scraped from under a peppermint plant showed 82 millionths of an amp against 53 from bare ground, in a preprint nobody has peer reviewed. That is what the jar’s own circuit produced, not a dose any plant received.</p><p>A government laboratory in Hyderabad, in India, went further. It put 2 electrodes into the soil around the roots of mung bean and chickpea, joined them by a wire into a closed circuit, without plugging it into anything. The circuit ran on a voltage the soil’s own microbes and roots make between the two electrodes. Those plants grew bigger roots and bigger shoots, they flowered earlier and more heavily, and they were ready to pick 10 days before the untreated ones.</p><p>The authors say where that electricity came from. Self-induced is their own word for it: the roots and the microbes made the voltage on the spot, with nothing supplying it. The paper did not include a voltage figure, though.</p><p>Ten days off a food crop, for 2 bits of metal and a wire.</p><p>Those 2 electrodes were wired to each other, and a copper coil is wired to nothing, so the Hyderabad experiment does not test the copper coil itself. What it does show is that the soil under a plant is making electricity of its own before anybody puts metal into it, and that giving it a path through metal changes how the plant grows.</p><p>Does a single piece of metal in the ground do anything at all, then?</p><p>Environmental engineers stand a bare rod in seabed mud on purpose and publish the measurements. They take a graphite rod, the stuff in a pencil, about the length of a finger, and push it upright into wet seabed mud so that its lower half sits in the airless layer and its upper half stands in the layer where there is oxygen. The microbes around that bare rod consumed 70% more oxygen than the microbes in the mud without one. All the rod does is give the electrons a path up out of the airless mud to where oxygen is available.</p><p>Granted, the mud was petroleum-contaminated marine sediment and there was no plant anywhere near it. But a rod with nothing attached to it, and no power supply in the experiment, changed what the living things around it did. Anybody who tells you nothing can happen unless something is plugged in has that result to explain.</p><p>Now to the copper coil itself, and the 4 controlled trials published between August and December 2025.</p><p>The 4 trials:</p><p>- Milwaukee, Wisconsin: mustard, kale, beets and turnips, indoors under lamps<br>- Pakistan: chillies, outdoors<br>- Nigeria: okra, outdoors<br>- Nigeria: cucumber, outdoors, from the same team, so it’s really 4 trials by 3 groups</p><p>All 4 reported some gains with copper beside the plants, which is undeniably a pattern, and not one of the 4 cites any of the others.</p><p>The one everybody has heard of is the indoor one. Marquette University in Milwaukee, and the title of the paper is “passive electroculture using copper rods does not improve yield in home container gardening”. This is THE paper cited against the whole idea, so let’s see what’s actually in it.</p><p>They grew mustard, kale, beets and turnips in 3.8-litre pots, 10 plants a group, indoors under LED lamps, for 8 weeks. What went into the pots is the copper coil gardeners build, copper wire wound round a wooden dowel, in 2 arrangements: a 40 cm dowel pushed 10 cm into the pot so most of it stood up in the air, or a 10 cm dowel buried completely. The third group had no coil at all.</p><p>Now get this. Further down the same paper they print a figure for what would drive a copper coil outdoors: 4 volts across a 40 cm copper rod standing in open ground. That is a voltage across a piece of metal, which is not the same measurement as the 100 to 300 volts per metre between ground and sky. In the very next sentence they report what they measured in their own room: copper rods put 2 millivolts into the soil, which is 0.002 volts.</p><p>4 volts outdoors. 0.002 volts indoors. The ratio is 2,000. Yet this is the experiment quoted against a copper coil standing under the open sky.</p><p>Convenient.</p><p>It gets better. They moved a rod to within 4 cm of a fluorescent tube and the reading went up tenfold, to 20 millivolts, and they say it straight out: copper rods do transmit electricity into the soil. More electric field around the copper, a bigger reading in the ground.</p><p>Those readings are in volts, though. They do not tell you how many amps reach a plant, and the dose that made Blackman’s plants heavier was a current, in amps. That is the measurement still missing.</p><p>However… turnips grown with the copper coil buried in the pot grew more root than the untreated ones, at p equals 0.01. In plain language, if the treatment made no difference, a result at least that extreme would turn up about one time in 100 under the test’s assumptions. Beets and turnips both put on more top growth with the copper buried, again at p below 0.01. Height and leaf colour showed no difference in any crop.</p><p>Copper leaching out of the wire is the obvious alternative explanation, and they tested for it: the plants beside the upright coils got as much copper into their soil as the buried coils did and grew nothing extra. The authors ruled out copper leaching, ruled out soil moisture too, and with both of their explanations gone they concluded the result was probably an accident of small sample sizes.</p><p>Had those turnips come out SMALLER under the copper at p equals 0.01, I tell you, 10 plants a side would not have been called too few.</p><p>Every positive result in the Milwaukee trial came from the coil buried completely. That makes the soil explanation worth testing. There is already electricity being produced in the soil by the bacteria we mentioned, and metal down in the soil can give it a path. Outdoors there is an electric field as well. You want both, metal well down in the soil and metal up in the air.</p><p>The 3 trials that were performed outside got a lot more than turnip root.</p><p>December 2025, Sindh Agriculture University in Pakistan, chillies. This trial used no coil at all, just 2 straight copper wires about 60 cm long pushed into an earthen pot of soil, sand and farmyard manure, attached to nothing, standing out in the weather. Untreated pots beside them, the whole comparison repeated 4 times over.</p><p>The chilli plants with copper beside them came out 30.83 cm tall against 22.5 cm. A third taller, at p equals 0.0011. Germination, leaves, branches, flowers, root weight and shoot weight all came out higher on the copper side too, and height is the measurement that passed the significance test. Same soil, same water, two lengths of wire bought once.</p><p>The paper’s own abstract prints none of those numbers and calls copper’s effects “moderate”.</p><p>A third taller is moderate, apparently.</p><p>The Nigerian team ran its 2 trials on okra and on cucumber. For the okra they put a coil of bare copper wire in the open air beside each plant, 5 turns, 10 cm across, the bottom 5 cm pushed into the soil. This is exactly the object people buy. On 12 plants with copper against 6 without, the copper side gave about 16% more pods per plant.</p><p>There is one number in the okra paper where the untreated plants came out ahead, and it’s the first thing a naysayer will point at: the average pod from the copper side weighed 32.8 grams, against 35.4 from the untreated side. Lighter pods. Well, okra is picked as an immature fruit, 3 to 7 days after flowering, and past that it goes tough and pithy. The untreated plants also stopped producing new pods partway through the season, which the authors report themselves, while the copper plants kept producing through 4 harvests. The authors call it a trade-off between fruit number and fruit size, I call it more okra.</p><p>So, indoors under lamps, some extra root and top growth with the copper buried. Outdoors, more fruit and a third more height. That is the pattern you’d expect if the copper is collecting charge out of the open air. And the indoor paper is the one that gets quoted against the coil.</p><p>Makes perfect sense, doesn’t it?</p><p>It is still a thin record. Four small trials by 3 groups, and none repeated by anybody. More research is required.</p><p>What about the thousands of people doing this at home since the early 2000s?</p><p>I’ve tested it myself and it works.</p><p>But one man’s garden is one man’s garden, and I’d say the same thing to a naysayer, so let’s look at the ones anyone can check. Ten amateur side-by-side trials of the copper coil can be opened and read, on 3 continents and in 2 languages, from 2011 to 2026, run by people who wanted it to work and by people who wanted it to fail. Not a single one weighed a harvest, and almost all ran a single plant or a single garden bed on each side.</p><p>One plant against another plant cannot tell you how BIG an effect is. It can tell you which plant won, though. The chilli plants with copper beside them averaged 30.83 cm and the untreated ones 22.5, and plants of one kind in one trial vary in height from pot to pot. Take that gap, take that pot-to-pot variation, and work out how often ONE copper plant comes out taller than ONE untreated plant, and the answer, on my arithmetic, is somewhere between three quarters and nine tenths of the time. Three times out of four, at the lower end of that estimate. So a gardener who photographs 2 pots and finds the copper one taller is looking at the likeliest outcome of an effect that size.</p><p>The same arithmetic covers the trials that went against the coil: at one plant a side, somewhere between 1 pair in 4 and 1 in 10 comes out the wrong way even when the effect is real and large. One of the longest amateur trials ran for 120 days in a Florida garden, 3 pairs, the plant without a coil ahead every time, judged from photographs. Here’s the thing though. The garden was sandy, low-nutrient soil, and 2 of the 3 plants were frost-damaged and recovering.</p><p>Lemström published the conditions under which his own effect fails: well-cultivated, well-manured ground gives a bigger harvest increase under the wires than poor soil does, on poor soil there is almost no increase at all, and treatment in hot sunshine damages most vegetables. Sandy low-nutrient soil in Florida fits the case he said gives you nothing.</p><p>That does not explain every failure. The most careful comparison of the 10 was run in Queensland, in eastern Australia: 4 raised garden beds, a whole summer. The gardener weighed no harvest, and he could see no difference between the beds with a coil and the beds without. Queensland’s beds were raised and well kept, so poor soil does not cover them.</p><p>Ten amateur comparisons in 15 years on 3 continents, and not one of those gardeners grew 5 pairs of plants, a coil beside one and nothing beside the other, and counted which plant of each pair won. Keep that in mind, we’ll get back to it.</p><p>Now for the part that makes people cringe.</p><p>Gardeners doing this at home tie a quartz crystal to the coil and say it works better. Quartz, the crystal in a wristwatch, makes a small electric charge when you squeeze it. So does plant material. I have no tested mechanism for the crystal. My guess is a piezoelectric interaction with the plant, in one direction or the other, vibration turning into electricity or electricity into vibration. Piezoelectric means exactly that: squeeze the material and it makes electric charge, feed it charge and it moves.</p><p>What is actually measured is that both quartz and processed plant cellulose produce charge under pressure. Press on each with about the force of a small apple’s weight, and out comes electric charge, in coulombs, a coulomb being the charge an amp moves in one second:</p><p>- Quartz: roughly 2 trillionths of a coulomb<br>- Processed plant cellulose: about the same<br>- Processed plant cellulose with its fibres well lined up: up to roughly 8 times as much as the quartz</p><p>Eight times more than the quartz itself, out of plant material. Plant material being piezoelectric was published in a physics journal in 1955, so this isn’t some “high frequency” nonsense, it’s physics from 1955.</p><p>Truth be told, though, there is no measurement of a quartz crystal tied to a coil anywhere. Not a voltage, not a current, nothing. The people sneering at the crystal and the people selling it have exactly the same amount of evidence.</p><p>Zero.</p><p>Which brings us to how little of this subject has been studied in the last hundred years.</p><p>Europe PMC, an index of life-sciences papers, holds 10 records for the word electroculture. Ten. Two of them are about Venus flytraps. Fewer than 50 papers in any index have the word in their title. Those numbers measure what a modern database can find, not how much work was done. Much of the early work is French, on paper, and was never digitised.</p><p>And the British committee’s 18 reports? From 1922 onward they were marked not for publication, and 2 copies of each were printed… one for the Minister, one for the ministry’s archive. This is why the public record of an 18 year government programme is so thin, and why “there is no scientific evidence” gets said with a straight face.</p><p>The French national horticultural society is my favourite example of all this.</p><p>Its scientific council filed electroculture under pseudoscience, right next to structured water, in a report published in 2020, and on the same page printed the evidence for it.</p><p>The report says that more than 450 researchers published experiments and observations on electricity and plants up to 1912. It also says that more than 150,000 of Christofleau’s collectors, the metal box on a post with the bundle of points on top, were manufactured, the last in 1939.</p><p>Agriculture did use it.</p><p>So how did the horticultural society’s scientific council arrive at “pseudoscience”? It read 4 publications, picked at random. It ran no experiment. Its own conclusion, in the same report, is that an effect of electric current on germination probably exists, and that rigorous, properly sized experiments are rare. This is literally my argument, written by the other side.</p><p>The reason it gives for rejecting the copper coil is that it is difficult to think, in its own words, that apparatus that crude could deliver a current big enough to matter.</p><p>The society never measured that current. Nor considered what it might do to the soil bacteria.</p><p>The 3 loudest public sources telling people the coil doesn’t work ran nothing themselves:</p><p>- A gardening advice column run by an American university, quoting a horticulture professor saying none of the claims are backed by reliable scientific evidence<br>- A large gardening channel, saying it is “almost certainly not effective in any material way”<br>- The French national horticultural society, which read its 4 papers</p><p>Not one of the 3 grew a plant or measured anything, and the advice column came out more than a year before any of the 4 modern trials existed.</p><p>Some experts…</p><p>Why hasn’t the trial that would settle this been run then? Not a big one, not an 18 year one, just a proper one.</p><p>Because there is nothing to sell at the end of it.</p><p>A copper coil is bought once. Fertiliser is bought every season, forever. The British committee stopped in 1937 because its wires and its generator cost more than the additional crop they were getting. The passive coil has no such fuel bill, but there is still no repeat order in it for whoever pays to find out.</p><p>The people who grow food have paid instead: a hundred springs of fertiliser, sold to them by companies that live on the repeat order, while the coil they would buy once waited for somebody to check it. If it does a fraction of what those charged wires did, that is a hundred years of food nobody grew.</p><p>A century after a government wrote down that the effect is real, the modern case for and against the passive copper device rests on 4 small trials published inside 5 months of one year, none of which anybody has repeated.</p><p>Which leaves it to you. Here is the test:</p><p>1. Do it outdoors, in open ground. That is where the coil stands in the electric field between the ground and the open sky, and where the 3 outdoor trials got their results.</p><p>2. Push the coil well down into the soil and leave the rest of it standing up in the open air. That is what the chilli trial did with straight wires and the okra trial did with a coil. You want metal in the soil and metal above it.</p><p>3. Same seed packet, same soil, same water and same feeding on both sides. Ten plants of the same kind, in 5 pairs, and in each pair one plant gets a copper coil beside it and the other gets nothing. Lemström got his biggest increases on well-manured ground and almost none on poor ground.</p><p>4. Alternate them along one garden bed, one with a coil, the next without, all the way down. Never all the coils at one end and all the bare plants at the other, because the two ends of a garden bed are never identical. Light, drainage and soil depth all change along its length.</p><p>5. At the end of the summer, count which plant of each pair came out taller.</p><p>That’s it.</p><p>A win is the plant with the copper coil coming out taller than the plant beside it. If the coil does nothing and each pair is an independent coin toss, 4 or 5 wins out of 5 pairs comes up 19 times in 100 by pure luck.</p><p>If the effect is the size the chilli trial measured, with the per-pair chances estimated above, 4 or 5 wins out of 5 pairs comes up somewhere between 63 and 92 times in 100, on my arithmetic again.</p><p>More than 3 times what luck alone gives you. No scales, no software, no weighing anything.</p><p>One garden running 5 pairs is not proof. About one summer in 5 the copper side wins 4 or 5 pairs by chance alone, and a gardener who stopped there could have believed something false. Two hundred gardens running 5 pairs is a thousand pairs, enough to see whether the results keep coming out above half. That is the difference between a story and a result, and the only reason we do not have it is that nobody has organised it.</p><p>Ten plants, a spool of copper wire, one summer, and you’ll have run one more experiment than the French national horticultural society.</p><p>Let me know what comes out. :)</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2c1a80f5bf69" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Discovery of Elixir of Life]]></title>
            <link>https://medium.com/@filip.milner/the-discovery-of-elixir-of-life-36d67d9f6538?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/36d67d9f6538</guid>
            <category><![CDATA[chemistry]]></category>
            <category><![CDATA[health]]></category>
            <category><![CDATA[physics]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Sun, 30 Aug 2026 19:25:58 GMT</pubDate>
            <atom:updated>2026-08-30T19:25:58.708Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*bL6tVvAnNaWL_40NLbAxvQ.png" /></figure><p>Good news, everyone!</p><p>The Elixir of Life has been discovered!</p><p>All diseases cured, lifespan extended to +1000 years! If you’re bald, you get your hair back, if you’re barren, you get the ability to have children, and if you’re deathly ill, it’s no biggie.</p><p>Depressed? Gone. Love for life back.</p><p>Harmed by vaccines? Fighting a chronic disease for decades? You’re back to normal.</p><p>Without a doubt — the greatest discovery of the 21st century.</p><p>No, this is not a joke, fake news, or exaggeration.</p><p>It was discovered last year by a fringe science researcher called — I shit you not — Ngolo.</p><p>You can make the elixir by yourself, at home, or you can order the currently most potent form of the elixir from “Ngolo”. However, even though it’s potent, I believe it’s very far from what it could be. I compare Ngolo’s current “elixir” to the first cars that drove 10 km/h and broke down every 5 minutes, or the first bicycles without a steering wheel.</p><p>Still works, though.</p><p>People are not regrowing their missing limbs yet or rewinding their biological clock back to 20 years old, but they are indeed eradicating massive chronic issues which Western medicine has failed to treat, eliminating mental health problems and getting their hair back. Just some examples. There are many.</p><p>We are just getting started. I’ve already done a lot of research myself, not as much as Ngolo, but I can confirm his findings, as well as the experiences of the elixir’s community, which has already grown to 2.4k members and is still growing.</p><p>We’re gonna get there probably in a year, two or three, not more. Unless the black suits show up, but I think the genie is out of the bottle and nothing much can be done. Good on Ngolo for releasing it for free and having an open-source approach.</p><p>I’m sure everybody reading is thinking at this point: what is this elixir made of…?!?</p><p>Water.</p><p>Water and…?</p><p>Just water.</p><p>Well, that can’t be real, right? Sounds like homeopathy or something.</p><p>Well… there is much you need to learn, young padawan.</p><p>Now, I do not like when my posts are too long, as they’re already long as is, but this one must be an exception. The application of this discovery will truly transform every person’s life and our civilization, and trigger people to do a lot of good because they will finally be in the state of mind and circumstances necessary to do good.</p><p>Suffering and hard circumstances make people selfish, sometimes downright evil.</p><p>Happiness and a relaxed state make even evil people generous and kind.</p><p>I will repeat this because it’s important: this water is the most important phenomenon you will ever read about or experience. It is imperative that you act on it.</p><p>Before we begin to understand it, I’m dropping the sources because most people will not have the attention span necessary to get to the end of the post: search for “NgoloTesla” and “Water Wizards” on X, or go to the website wizard (dot) talk.</p><p>The community feedback is extensive. Anecdotes are obviously not clinical proof, but the volume and apparent consistency of the reports are part of what convinced me that this deserves serious investigation. I have also experienced effects myself, and they are far beyond what I would personally assign to placebo.</p><p>Now let’s start.</p><p>There is a list of things you need to accept before you truly understand why this water actually works — if you are one of those rigid-3D-thinking people. Some of these things have only recently been validated by mainstream science, although they were postulated years ago.</p><p>The list:</p><p>a.) water is truly a bizarre substance with 75–130 anomalies, depending on how strictly you define “anomaly”. Most substances have 0</p><p>b.) proper functioning of the human body is entirely dependent on available nutrients AND negative charge</p><p>c.) lack of negative charge in the cells prevents them from functioning normally and also prevents cell division (regeneration) — it is correlated with localized disease</p><p>d.) there is a “4th phase of water” that is scientifically proven to exist, and it’s called “structured water”/“EZ water”. It’s also proven to be integral to the healthy functioning of cells</p><p>If you already understand all this, then I can spoil the method of functioning. The water is charged with a specialized electrical circuit so that it contains a large amount of STABLE negative charge, and it is then used in various ways: rubbed on the skin, drunk, bathed in or inhaled. The negative charge is absorbed by the cells and utilized for normal functioning and regeneration.</p><p>That’s it.</p><h3>Water Basics</h3><p>Here are some of the peculiar phenomena:</p><p>a.) No other common molecular compounds (organic or otherwise) are liquid near room temperature and expand upon freezing. There is only one “exotic analogue” that has this property: gallium arsenide, which increases by 11% in volume upon freezing but has a melting point of 1238°C. And it contains a metal that already expands upon freezing.</p><p>b.) Water also has a very high relative permittivity (dielectric constant), which is around 78–80 near room temperature. For comparison, common solvents at room temperature include ethanol at 24.5, acetone at 20.7 and hexane at 1.89, while glass and plastic are around 4.</p><p>To further showcase how strange water is, despite not containing metals, it has a relative permittivity similar to that of magnetite (iron oxide) nanoparticles. Hematite, which is another form of iron oxide, has a value of 30.</p><p>As a consequence of its high relative permittivity, water acts as an electrical “cushion” for ions, which are charged molecules.</p><p>c.) If water is clean enough and lacks nucleation sites (impurities or surfaces on which it can begin forming crystals), it can be cooled to −38°C while still remaining liquid. Additionally, in this supercooled state, experiments reveal high-density liquid (HDL) and low-density liquid (LDL) coexisting, fluctuating like Jekyll and Hyde.</p><p>It has been discovered that this happens to a lesser degree at room temperature and with impurities. Water is actually two liquids.</p><p>d.) In 1893, British engineer William Armstrong demonstrated something that looked like a magic trick: when two beakers of deionized water are connected by a small amount of liquid and then subjected to a high-voltage electric field (around 15,000 volts), the water climbs out of the beakers and forms a stable, free-standing bridge that defies gravity. It was rediscovered in the 2000s and studied extensively.</p><p>This “floating water bridge” can extend up to a few centimeters across empty space. It is cylindrical and 1–3 mm in diameter. It can remain stable for up to 45 minutes. The exact length, diameter and time of stability depend on the water bridge configuration, i.e. beaker size, water purity, temperature…</p><p>Water actively flows through the bridge, generally from the anode (positive electrode) to the cathode (negative electrode). The bridge’s surface temperature rises from room temperature to around 60°C before it finally collapses.</p><p>For over a century, scientists have struggled to explain how liquid water can maintain structural integrity while suspended in mid-air.</p><p>The key thing to understand with this phenomenon is that a difference in electric potential causes water to behave “strangely”. Granted, this anomaly is not wholly unique to water. A few other highly polar alcohols behave in the same way, but still, it’s very relevant.</p><p>e.) Two water molecules that appear chemically identical can react at different rates solely because their hydrogen nuclei occupy different quantum-spin configurations. You cannot distinguish them by writing different structural formulas.</p><p>Ortho-water: the proton spins form a symmetric total-spin state. It is often simplified as the spins being “parallel.”</p><p>Para-water: the proton spins form an antisymmetric state, simplified as “antiparallel.”</p><p>Rough Visualization of this principle:</p><p>Ortho-water: ↑ ↑ total nuclear spin = 1<br>Para-water: ↑ ↓ total nuclear spin = 0</p><p>This results in the following:</p><p>→ Para-water can settle into the lowest possible energy state.<br>→ Ortho-water cannot go that low, so it always retains slightly more energy.</p><p>So the difference in nuclear spin indirectly changes how the molecule rotates, interacts with electric fields and sometimes reacts chemically. In a 2018 molecular-beam experiment, researchers separated the two forms and reacted them with diazenylium ions (positively charged ions made of 2 nitrogens and one hydrogen).</p><p>Under those specific experimental conditions, para-water reacted approximately 23% faster than ortho-water.</p><p>Now imagine this anomaly translated into possibilities within the immense complexity of the human body.</p><h4>Water Basics — Two key phenomena to understand</h4><p>f.) Here’s something that sounds like it came straight out of a sci-fi show but is actually real. It’s called proton hopping or the Theodor Grotthuss mechanism.</p><p>Most ions move by physically drifting through the liquid. But protons and hydroxide ions can move by a relay mechanism through the hydrogen-bond network: bonds break and form in a coordinated chain, so the “identity” of the proton hops along without a single ion trudging the whole distance, instead effectively teleporting.</p><p>In other words, H⁺ and OH⁻ have exceptionally high limiting molar ionic conductivities compared to ordinary ions, and the standard explanation is proton hopping through the H-bond network.</p><p>It gets even more bizarre. Protons have been observed to teleport in groups, as a single quantum entity.</p><p>In 2015, researchers at Peking University directly visually recorded, for the first time ever, the concerted quantum tunneling of protons within a water cluster.</p><p>Using a scanning tunneling microscope with a specially designed chlorine-terminated tip, they recorded four protons in a water tetramer teleporting simultaneously in perfect synchronization.</p><p>In quantum-speak, this is called “many-body tunneling”, a phenomenon where multiple particles act as a single quantum entity, with their fates entangled across space.</p><p>The researchers found that this collective tunneling was “extremely sensitive to the coupling with atomic-scale environment due to the demanding phase coherence between the protons”.</p><p>In other words, these four protons had to maintain quantum coherence with each other while simultaneously tunneling through an energy barrier that, according to classical physics, should be impassable.</p><p>This quantum tunneling doesn’t just happen in carefully controlled laboratory conditions. It happens inside you. Right now. The enzymes in your body, the molecular machines that power every chemical reaction keeping you alive, rely on quantum tunneling to function.</p><p>Research by Judith Klinman at UC Berkeley and Nigel Scrutton at the University of Manchester has demonstrated that enzymes accelerate reactions by factors of over a trillion, and a significant portion of this acceleration comes from their ability to promote quantum tunneling of protons.</p><p>To restate what has been said, because it is important, researchers in lab conditions have proven that there are enzymes that participate in this quantum tunneling effect, instantly teleporting charge across your body.</p><p>g.) It gets even stranger. In 2016, researchers at Oak Ridge National Laboratory discovered a new ‘quantum tunneling state’ of water confined in tiny 5-angstrom channels within the mineral beryl.</p><p>In this state, the water molecule’s protons became so delocalized, so spread out across space, that the molecule had “an anomalously small proton kinetic energy due to spatial delocalization, and a modified center of gravity and dipole”. The water molecule was no longer where you’d expect it to be. It was, in a quantum sense, everywhere at once.</p><p>Perhaps the most profound implication of proton tunneling in water concerns the very mechanism of evolution itself. Per-Olov Löwdin first proposed in 1963 that spontaneous DNA mutations, the raw material that determines our form and function, could be caused by protons tunneling across hydrogen bonds in the DNA double helix.</p><p>Here’s how it works: DNA base pairs are held together by hydrogen bonds. In these bonds, a proton normally rests in a deeper “potential well” on one side. But quantum mechanics allows this proton to tunnel through the energy barrier into a shallower well on the other side — a process called “tautomeric transition”.</p><p>If DNA replication occurs while the proton is in this abnormal position (shallower well), the base pairing rules are violated, resulting in a mutation.</p><p>This means that evolution, the process that created every living thing on Earth, is fundamentally driven by quantum mechanics.</p><p>The mutations that gave rise to every form and function imaginable may have originated from protons teleporting through energy barriers in water-mediated hydrogen bonds.</p><p>In other words, quantum tunneling is the engine of life’s diversity.</p><p>And it all starts with water.</p><p>I can go on and on. There are 75+ of these anomalies, some more important to the topic at hand, some less.</p><p>The summary of it all would be that water is a substance that exists in at least twenty-two solid forms, whose atoms routinely teleport through energy barriers, whose isotopic variants can slow the passage of biological time, and whose presence inside giant planets may be hot, black, and electrically conducting (superionic ice).</p><p>It is a remarkably cohesive liquid with at least 75 documented distinct anomalous properties — behaviors that differ from what we’d expect of a “normal” liquid.</p><p>High surface tension. High heat capacity. High melting and boiling points. Low compressibility. Expansion upon freezing. Maximum density at 4°C rather than at freezing.</p><p>These anomalies have long been attributed to hydrogen bonding, but the deeper question is: why should hydrogen bonding produce such extreme effects? Nobody knows for sure.</p><p>This section might have been boring for some, but it’s crucial to understand that the substance we are surrounded by and have accepted as “mundane” and “ordinary” is actually magic itself.</p><p>The best part — humans are 70% water. So we’re magic too.</p><h3>Human body and negative charge</h3><p>The central scientific idea presented in the book “The Body Electric”, which was published in 1985 and was considered “woo-woo”/alternative, has aged surprisingly well. The mainstream science now accepts the key proposition:</p><p>Living tissues use endogenous electrical signals as part of healing, growth, cellular coordination and regeneration.</p><p>That is now an established area of research called bioelectricity or developmental bioelectricity.</p><p>More specific statements from the book, backed by the writer’s observations and now accepted by mainstream science despite being considered crackpot theories at the time, are as follows:</p><p>1.) Wounds produce measurable electrical currents</p><p>2.) Applied electrical stimulation can improve some wounds</p><p>3.) Bone has electromechanical properties</p><p>4.) Electrical stimulation can assist bone healing</p><p>5.) Bioelectric states influence development and regeneration</p><p>There are many other statements in that book that are, in my opinion, truthful but are still considered crackpot theories… So let’s stick to mainstream science and avoid ambiguity.</p><p>Let’s continue with grounding.</p><p>Grounding means electrically connecting the body to the Earth through bare skin or a conductive grounded surface. The best-supported findings accepted by mainstream science are:</p><p>1.) Grounding rapidly equalizes the body’s electrical potential with the Earth through charge transfer.</p><p>2.) Grounding improves peripheral blood-flow regulation. Blinded studies found increased facial microcirculation and improved thermal circulation throughout the face and torso after one hour.</p><p>3.) Grounding increases the NEGATIVE surface charge of red blood cells. Two hours of grounding increased red-cell zeta potential by an average factor of 2.7 and SUBSTANTIALLY reduced cell clumping. This provides a direct electrical mechanism for improved blood fluidity and microcirculation.</p><p>4.) Grounding can rapidly reduce local signs of inflammation. Medical thermography has recorded striking reductions in abnormal heat around grounded injuries, together with reduced redness, swelling and pain.</p><p>5.) Grounding alters the inflammatory response after tissue damage. Controlled exercise studies found changes in neutrophils, lymphocytes, platelets and other immune markers following grounding.</p><p>6.) Grounding reduces some measures of muscle damage. After identical exercise, creatine kinase rose significantly in ungrounded subjects but not in grounded subjects.</p><p>7.) Grounding appears to assist wound healing. Clinical cases using grounding patches above and below injuries showed rapid reductions in inflammatory heat and visible improvements in wound repair.</p><p>8.) Electrical charge is fundamental to cellular regeneration. Membrane voltage controls whether cells migrate, divide, differentiate or remain inactive. Cell division requires an organized sequence of negative membrane-voltage states rather than one universal millivolt threshold.</p><p>The simplified version:</p><p>1.) Conductive grounding transfers charge and brings the body toward Earth potential.</p><p>2.) Grounding measurably increases the negative surface charge of red blood cells.</p><p>3.) This increased negative charge reduces red-cell aggregation and improves blood-flow properties.</p><p>4.) Grounding has produced measurable changes in microcirculation, inflammation, pain and muscle-damage markers.</p><p>5.) Tissue repair depends on blood flow, oxygenation, nutrient delivery and removal of waste.</p><p>6.) Cells use metabolic energy and ion transport to generate negative membrane potentials.</p><p>7.) Those membrane potentials actively regulate migration, proliferation, differentiation and regeneration.</p><p>The most coherent interpretation is that grounding drains induced voltage and excess surface charge (from nnEMFs), brings the body to Earth potential (more negative) and permits electron transfer. This increases the negative charge separating red blood cells, improves microcirculation and changes autonomic and inflammatory activity.</p><p>Every living cell maintains a voltage across its membrane, with the inside of the cell being negatively charged relative to the outside.</p><p>Cells constantly spend energy maintaining this negative voltage because without it they cannot properly control what enters and leaves, communicate with nearby cells, regulate their volume, produce energy or repair themselves.</p><p>If the membrane potential collapses toward 0 mV for long enough, the cell stops functioning and eventually dies.</p><p>So the simplified statement that cells need negative charge to stay alive is correct and fully in line with mainstream science.</p><p>It’s also fully in line with mainstream science that cells require, on average, about a 50–100% increase in negative charge to multiply (regeneration/rejuvenation).</p><p>However, there are layers to it. Cells don’t require exactly −25 mV to function and −50 mV to multiply, which is too broad an approximation because different cells operate at different voltages.</p><p>Neurons commonly operate between approximately −40 and −90 mV.</p><p>Heart cells are close to −90 mV.</p><p>Resting T cells are around −50 mV.</p><p>Endothelial cells are at around −43 mV.</p><p>Proliferating skin cells can be around −24 mV.</p><p>There is no universal number, but there is a universal principle: the electrical state of a cell determines what that cell is capable of doing.</p><p>This includes whether it remains inactive, migrates, divides, differentiates into another type of cell or dies. Even more importantly, scientists can change cellular behaviour by artificially changing this voltage.</p><p>One of the clearest examples comes from human breast cells.</p><p>Researchers measured MCF-7 cells in several electrical states, centred around approximately −9, −17.4, −24.6 and −40.4 mV.</p><p>When the cells were inactive, they accumulated in the least negative state.</p><p>When they began moving toward DNA replication, they shifted toward the more NEGATIVELY charged states.</p><p>Depending on which states are compared, this represents approximately 40% to 70% additional negative polarization in cells that were already moderately charged.</p><p>Cells starting near −9 mV sometimes became between 170% and 350% more NEGATIVELY polarized before division (regeneration / rejuvenation).</p><p>Therefore, not every cell must become exactly 50%+ more negative before dividing. It is, again, too broad an approximation.</p><p>What is not a broad approximation, but a universal truth, is that all cells pass through a substantially more NEGATIVE electrical state as part of the process required to replicate DNA and continue through the cell cycle.</p><p>The same thing appears during differentiation.</p><p>Human stem cells beginning to form bone became progressively more NEGATIVELY polarized, moving from approximately −37 mV toward −93 mV.</p><p>Stem cells beginning to form fat moved from approximately −47 mV toward −121 mV.</p><p>When researchers artificially pushed these cells toward 0 mV (no charge!), normal differentiation was disrupted. Completely.</p><p>When they pushed them in the opposite direction, toward a more negative state, markers associated with bone formation increased again. The correlation is impossible to ignore.</p><p>So the key point with regard to this “magical” water and the cell properties we are discussing now is this: the negative charge was not merely present while regeneration happened. Changing the charge changed the regeneration process itself.</p><p>Cancer provides another clue.</p><p>Cancer cells are generally more depolarized, meaning LESS NEGATIVELY charged, than healthy cells of the same tissue.</p><p>MCF-7 cancer cells can sit around −9 to −30 mV, while healthy MCF-10A breast cells are commonly measured around −40 to −58 mV.</p><p>Mainstream science would say: cancer is obviously not caused by low cellular voltage alone, although it’s apparent from all the mainstream-accepted statements I’ve shared so far that it is almost certainly the main driver.</p><p>Regardless, mainstream science agrees that the loss of normal negative membrane potential is strongly associated with uncontrolled growth, poor differentiation and the loss of normal tissue organization. “Strongly associated” is as far as they’ll go.</p><p>Anyway, let’s continue.</p><p>How can cells become more negative before division, while cancer cells that divide uncontrollably are often less negative?</p><p>Because cells do not remain at one voltage, they move through an organized sequence of electrical states.</p><p>They become more negative at certain checkpoints, less negative at others, divide, differentiate and then return to their normal resting voltage.</p><p>I also have to argue that cancer cells are significantly different than healthy cells, and that cancer cells are unlikely to arise from a highly negatively charged cell.</p><p>Either way, cancer cells appear to lose part of this organized electrical sequence and remain trapped in an abnormal, depolarized state. Correct the depolarized state, correct the cancer.</p><p>The overarching point that needs to be understood about this cell thing is not “more negative automatically means more division”. Healthy growth and regeneration require cells to achieve the correct INCREASED NEGATIVE voltage at the correct moment.</p><p>But this is not something a person needs to bother with. Cells know what to do &amp; how to do it. They just need to be provided with adequate negative charge. At least in my opinion.</p><p>The simplified version:</p><p>1.) Every living cell maintains a negative internal voltage.</p><p>2.) Cells spend energy continuously to preserve it.</p><p>3.) If this voltage collapses toward zero, normal cellular function collapses with it.</p><p>4.) Changes in membrane voltage control migration, division, differentiation and regeneration.</p><p>5.) Many cells become substantially more negatively polarized during important stages of DNA replication and differentiation.</p><p>6.) Healthy cells are usually more negatively polarized than cancerous cells of the same tissue.</p><p>7.) Artificially depolarizing cells can interrupt normal differentiation.</p><p>8.) Artificially hyperpolarizing cells can promote certain regenerative and differentiation programmes.</p><p>So the argument is: cells cannot migrate, divide, differentiate or organize themselves into healthy tissue without moving through precise negative electrical states.</p><p>Grounding introduces negative charge into the body.</p><p>Grounding has measurable effects on blood, circulation, inflammation and wound healing.</p><p>Negative cellular voltage controls regeneration.</p><p>The direct connection between all of these statements has not yet been mapped in every microscopic detail, but the pieces are already there.</p><p>Properly absorbed negative charge creates conditions that favour healing because life itself is organized and maintained through negative electrical potential.</p><p>This sounds like a crackpot theory only until you break it into individual statements, at which point nearly every part of it is mainstream science.</p><p>Congratulations, you now know the key to extended youth, extreme regeneration and how to achieve a lifespan of 1,000+ years. Fingers crossed the black suits don’t get involved, or do get involved, but too late. I sure don’t plan to be a frail old man at 80. That’s such a scam.</p><p>Let’s move on to the final piece of this “water is life” puzzle.</p><p>The Fourth Phase of Water</p><p>In 2013, Gerald Pollack, a professor of bioengineering at the University of Washington, published a book called “The Fourth Phase of Water: Beyond Solid, Liquid, and Vapor.”</p><p>His proposition is quite simple. Water does not exist exclusively as solid, liquid and gas. When it comes into contact with certain hydrophilic (“water-loving”) surfaces, some of it reorganizes into a more ordered state, with properties significantly different from the water floating freely beside it.</p><p>Spoiler: outer cell membranes are hydrophilic. How convenient is that? :)</p><p>Pollack calls this new form of water the fourth phase of water. It’s also called structured water, interfacial water, or exclusion-zone water (EZ water).</p><p>Why “exclusion zone”?</p><p>Because if you place microscopic particles into water next to certain hydrophilic materials, a large empty region forms in the region touching the surface. The particles, dissolved substances and other bits floating around in the water are pushed away from it.</p><p>And by “large”, I do not mean three molecules wide, detected indirectly by six scientists operating a billion-dollar machine and arguing about the graph afterwards. These exclusion zones can extend for hundreds of micrometres, which means you can directly observe them through a normal microscope.</p><p>You put particles into the water, and the particles move away from the material, forming a large particle-free zone.</p><p>The phenomenon has been reproduced by multiple research groups. Mainstream scientists still argue over why exactly it happens, what this ordered water really is, and whether Pollack is permitted to call it an entirely new “phase” without being sent to fringe-science prison.</p><p>What these “very serious” scientists, paid for directly or indirectly by lobbyists or institutions with an agenda, cannot seriously argue is that the exclusion zone itself does not exist.</p><p>It DOES exist.</p><p>As is often the case, the dispute is not whether the unexplained thing is happening, but whether we are allowed to describe what is happening — in plain language.</p><p>And now we get to the crucial part: the exclusion zone is NEGATIVELY charged.</p><p>As this structured region forms, positively charged hydrogen ions (protons) are pushed into the surrounding water. You therefore end up with a negatively charged body of water beside proton-rich, positively charged water.</p><p>That is charge separation, or in normal language: a battery.</p><p>Electrical potentials of approximately −100 to −200 mV have been measured between the exclusion zone and the surrounding water. Water touches an appropriate surface, reorganizes itself and produces a measurable electrical potential.</p><p>No metal plates, conventional battery acid, copper wire, external circuit, plug in the wall or little man inside turning a crank.</p><p>Water and a hydrophilic surface.</p><p>The most common substance on Earth spontaneously separates negative and positive charge. Magical, isn’t it?</p><p>So where does the energy to do so come from?</p><p>Light.</p><p>Pollack’s experiments found that radiant energy expands the NEGATIVELY charged exclusion zone. Ultraviolet works, visible light works, but infrared works particularly well.</p><p>At approximately 3.1 micrometres, relatively weak infrared exposure increased the size of the exclusion zone by up to four times, while increasing the temperature of the water by less than 1°C.</p><p>Expose it to light, the negatively charged region grows. Remove the light, it gradually shrinks.</p><p>In other words, water absorbs radiant energy and converts at least part of it into electrical charge separation.</p><p>Plants use chlorophyll. Solar panels use semiconductors. Water apparently just does it by touching a surface.</p><p>The sun charges water. Sound familiar? Crackpot researchers vindicated.</p><p>Pollack proposed that EZ water has a repeating hexagonal structure, sometimes represented by the formula H₃O₂. That exact molecular model remains disputed, and honestly, for the point I am making, it is not particularly important.</p><p>Call it H₃O₂. Call it structured water. Call it interfacial water.</p><p>Or call it “a proton-gradient-driven exclusion phenomenon produced through diffusiophoresis”, if making the sentence impossible to understand helps the people at big, expensive, all-white labs feel safer.</p><p>The observed phenomena remain the same:</p><p>1.) Water beside certain hydrophilic surfaces behaves differently from ordinary bulk water.</p><p>2.) It forms zones which exclude particles and many dissolved substances.</p><p>3.) These zones can extend hundreds of micrometres away from the surface.</p><p>4.) The exclusion zone carries a NEGATIVE charge.</p><p>5.) Positive charge accumulates in the surrounding water.</p><p>6.) Radiant energy, particularly infrared light, substantially expands the negatively charged zone.</p><p>7.) Water can therefore absorb environmental energy and store part of it through electrical charge separation.</p><p>The exact molecular explanation is disputed. The large particle-excluding region, its electrical potential and its expansion under light have been observed.</p><p>There is a difference.</p><p>Now, why does any of this matter to the human body?</p><p>Because the body is not a plastic bottle filled with ordinary water.</p><p>Water inside us is constantly touching proteins, cell membranes, collagen, connective tissue, DNA, blood-vessel walls and countless other hydrophilic structures. At the microscopic scale, the total surface area is so enormous that very little biological water is truly “far away” from something.</p><p>Mainstream science usually calls this hydration water, bound water, interfacial water or biological water. These terms are safer than “fourth phase”, and scientists do enjoy safe terminology, but the central fact is already accepted: water beside proteins and membranes does not behave exactly like water inside a drinking glass.</p><p>Structurally altered, non-bulk water has been directly detected inside living cells. Every soluble protein is surrounded by a hydration layer whose properties differ from ordinary bulk water, and this layer is not some decorative wrapping placed around the protein for aesthetic purposes.</p><p>It influences how the protein folds, how stable it remains, which molecules can approach it, and how it interacts with its environment. Enzymes require their hydration layers to work properly. Membranes depend on the behaviour of water at their surfaces. Molecular recognition — the process through which molecules identify and attach to each other — is partly determined by the water between them.</p><p>Protons and hydroxide ions also move through water’s hydrogen-bond network, as explained earlier with the Grotthuss mechanism. Charge does not merely move through biological water as though the water were an empty pipe. The structure and behaviour of the water participate in the transport process itself.</p><p>We have spent decades describing cells as collections of proteins, membranes, mitochondria and DNA floating around in water, as though the water is the unimportant empty space separating all the real machinery.</p><p>It isn’t.</p><p>The water determines how proteins fold and bind, supports enzyme reactions, participates in membrane formation, carries charge between structures and provides the physical environment in which cellular voltage exists. Nearly every chemical reaction keeping the cell alive happens either in water, through water, or at a boundary where biological material meets water.</p><p>And this biological water is not behaving like ordinary bulk water.</p><p>Exclusion-zone formation has even been observed inside plant xylem vessels, the microscopic tubes plants use to transport water.</p><p>The measured exclusion zones extended approximately 133–240 micrometres around the vessels. In the smaller vessels, structured water eventually occupied up to 94% of the available cross-sectional area.</p><p>As the exclusion zones developed, water movement appeared. When researchers increased the salt concentration and suppressed formation of the exclusion zones, the observable flow disappeared.</p><p>So, inside an actual biological water-transport system, water formed large structured regions, charge separated, an electrical gradient appeared, and water began moving.</p><p>Published in Scientific Reports in 2024, by Anqi Wang and Gerald Pollack.</p><p>Not Pollack alone in his garage staring intensely at a glass until he began seeing things.</p><p>This is where we return to the elixir.</p><p>Ngolo’s water is not “magical” because all water is secretly the Elixir of Life already. Ordinary tap water obviously does not produce the same effects being reported by the community. Otherwise, every person drinking eight glasses per day would be biologically twenty years old and extremely confused about why doctors still exist.</p><p>The apparent discovery is a process which gives water a large amount of STABLE negative charge.</p><p>Stable is the important word.</p><p>Temporarily changing the electrical properties of water is not difficult. Apply voltage, expose it to an electric field, alter the surfaces around it, shine light on it or introduce different ions, and the electrical state changes.</p><p>Keeping a large charge from immediately redistributing, neutralizing, recombining or grounding itself away is another matter entirely.</p><p>That appears to be the actual invention.</p><p>The standard objection is that water cannot hold a meaningful negative charge for long because the charge would simply spread out and disappear.</p><p>Except, from the information we have gone through, we already know that:</p><p>1.) Water can maintain separated electrical charge.</p><p>2.) Water can form large NEGATIVELY charged regions.</p><p>3.) Hydrophilic surfaces can stabilize those regions.</p><p>4.) Light can make them expand.</p><p>5.) Water can store absorbed energy in the form of electrical potential.</p><p>6.) Biological water is structurally and electrically different from ordinary bulk water.</p><p>7.) Cellular function, division, differentiation and regeneration depend on precise NEGATIVE electrical states.</p><p>So the claim that water could be prepared in a specific way that allows it to carry and deliver stable negative charge does not violate physics.</p><p>It’s more akin to an engineering problem.</p><p>How exactly is the charge stabilized? What structure inside the water holds it? How much disappears upon contact with air? How much survives the digestive system? Which tissues absorb it most efficiently? Is drinking the best method, or inhaling? Does bathing expose more surface area? How much passes through the skin? How long does the effect last inside the body?</p><p>And most importantly, how much further can potency be increased?</p><p>Ten times? A hundred times? A thousand?</p><p>These are questions of optimization, not questions of whether negatively charged, structured water can exist at all.</p><p>We know it can.</p><p>This is why I compared the current water to the first cars that travelled at 10 km/h and broke down every five minutes. The basic mechanism works, but the “machine” is primitive, inefficient and far away from what it may eventually become.</p><p>Crude. Weak. Hopefully outdated within several years.</p><p>Still works, though.</p><p>There is also another interesting “coincidence”.</p><p>Places historically believed to have healing properties — waterfalls, natural springs, mountains, forests, coastlines — are all environments where the air and water contain increased negative charge.</p><p>Waterfalls violently split water into extremely small droplets, separating charge in the process. Moving natural water, mountain air and clean forests contain much higher concentrations of negative air ions than polluted cities or sealed indoor environments.</p><p>People visit these places and suddenly they breathe better, sleep better, have more energy, feel happier, experience less pain, and sometimes report improvements in inflammation or skin problems.</p><p>The usual explanation is that nature is relaxing.</p><p>Okay, sure, maybe waterfalls heal people because the splashing noise is nice.</p><p>Maybe mountain springs were independently considered healing across unrelated cultures because everybody happened to invent the same superstition.</p><p>Maybe grounding reducing inflammation has nothing to do with negative charge.</p><p>Maybe negatively charged air affecting biology has nothing to do with negatively charged water.</p><p>Maybe the NEGATIVE surface charge of red blood cells increasing by an average factor of 2.7 during grounding is unrelated to improved blood fluidity.</p><p>Maybe healthy cells being significantly more negatively polarized than cancerous cells of the same tissue is also unrelated.</p><p>Maybe the negatively charged water forming beside biological surfaces has absolutely nothing to do with cellular voltage.</p><p>Maybe all of these phenomena just accidentally point in exactly the same direction.</p><p>Or… negative charge is one of the fundamental resources required by life, and water is the primary material nature uses to collect, hold and distribute it.</p><p>This would also explain why the mechanism does not need to be some fantastical new biological pathway that has never existed before. The cells already know how to use negative electrical potential. They spend energy maintaining it every second of your life. Their migration, division, differentiation and organization are regulated through it.</p><p>The water is not teaching the cells some new trick.</p><p>It is apparently delivering more of a resource they already require.</p><p>The simplified version:</p><p>1.) Water forms an ordered state beside certain hydrophilic surfaces.</p><p>2.) This structured water excludes microscopic particles and many dissolved substances.</p><p>3.) The structured region carries a NEGATIVE charge.</p><p>4.) Its formation separates negative and positive charge, effectively producing a water battery.</p><p>5.) Radiant energy, particularly infrared light, expands the negatively charged region.</p><p>6.) The human body contains an enormous amount of water touching hydrophilic biological structures.</p><p>7.) Water around proteins, membranes and other cellular structures is measurably different from ordinary bulk water.</p><p>8.) This biological water is essential for protein folding, enzyme activity, membrane function, molecular recognition, charge transport and cellular organization.</p><p>9.) Structured exclusion zones have been observed inside biological water-transport systems, where their formation was associated with water movement.</p><p>10.) Cellular function and regeneration depend on organized NEGATIVE membrane voltages.</p><p>11.) A process has apparently been discovered which gives water a much larger and more STABLE negative charge than ordinary water.</p><p>12.) Thousands of people using this water report effects consistent with improved cellular function and regeneration.</p><p>Again, split the full argument into its individual pieces and almost every piece is already mainstream science.</p><p>Can water maintain a separated charge?</p><p>Yes.</p><p>Do biological surfaces alter and structure nearby water?</p><p>Yes.</p><p>Can structured water carry negative electrical potential?</p><p>Measured.</p><p>Does light expand the negatively charged region?</p><p>Measured.</p><p>Is structured water essential to cellular machinery?</p><p>Accepted.</p><p>Does negative membrane voltage regulate regeneration?</p><p>Accepted.</p><p>Apparently, the only forbidden part is putting all these facts in the same paragraph.</p><p>Is every microscopic step of the mechanism mapped and proven?</p><p>No.</p><p>The experiences came first. The complete explanation will come afterwards.</p><p>That is how discoveries usually happen, despite the cleaned-up version of history scientists like to tell once everything has become respectable. First, somebody observes that something works.</p><p>Then somebody willing to look like a complete lunatic proposes an explanation. The explanation is mocked. Careers are protected. Everybody agrees not to investigate it properly, and twenty years later the same idea reappears under a new name in a peer-reviewed journal.</p><p>Our bodies are mostly water.</p><p>Our cells require negative electrical potential.</p><p>Water can absorb radiant energy, reorganize itself and maintain separated negative charge.</p><p>Someone has apparently discovered a process which substantially increases that charge and keeps it STABLE long enough for the body to absorb and utilize it.</p><p>That’s it.</p><p>The Elixir of Life was not hidden inside some rare flower that blooms once every thousand years. It wasn’t buried underneath a pyramid, guarded by an immortal order of monks, or produced by sacrificing a virgin under the full moon.</p><p>It is made from the most common substance on Earth. And it was under our noses the entire time.</p><p>The substance we drink every day, bathe in, walk through in the rain, and somehow still understand almost nothing about.</p><p>Water… but finally — properly charged by correctly applying DC current to it.</p><p>Visit the resources I provided in the post to learn how to make it, or order it from Ngolo.</p><p>One obvious safety caveat: do not improvise the electrolysis process with arbitrary electrodes, salts or tap water. Electrolysis can change pH and generate entirely different chemical products depending on the water composition, electrode material and circuit configuration. The electrical effect and the chemical composition of the resulting water both need to be understood.</p><p>Study the materials, which explicitly specify: platinum or boron-doped diamond electrodes, distilled water. There are $5 setups that don’t involve diamonds or platinum, but I wouldn’t recommend it.</p><p>I might at some point sell it myself. I am personally still not happy with the potency, although the effects are apparent. Nobody is gonna regrow a limb from my current water, that’s for sure, and I’m all about results.</p><p>To be continued. Unless the suits show up. Fuck ’em, though.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=36d67d9f6538" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Aurora and the Geomagnetic Excursion]]></title>
            <link>https://medium.com/@filip.milner/aurora-and-the-geomagnetic-excursion-c8ab39865a62?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/c8ab39865a62</guid>
            <category><![CDATA[solar-storms]]></category>
            <category><![CDATA[geomagnetism]]></category>
            <category><![CDATA[space-weather]]></category>
            <category><![CDATA[science]]></category>
            <category><![CDATA[disaster-preparedness]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Mon, 24 Aug 2026 13:23:23 GMT</pubDate>
            <atom:updated>2026-08-24T13:26:55.804Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8uiO938JdB5GfjY9GhsRGA.png" /></figure><p>Since November 2023 there have been 6 occurrences of POLAR LIGHT in Croatia, every one of them visible with the naked eye, versus 9 naked eye aurora events between years 1737 and 2023. The importance of which — you’ll learn in this very long piece.</p><p>It’s long, and the attitude is overly analytical which makes it a worse read.</p><p>However, it’s also one of the most important things you’ll read if you know nothing about space weather, or have never heard the words ‘geomagnetic excursion’. It’s a complex topic, so it has to be long, there is no way around it.</p><p>Since most people’s attention span is fried from reels and doomscrolling, here’s a summary:</p><p>1.) We are currently in a geomagnetic excursion, confirmed by mainstream science research<br>2.) That’s why we’re seeing auroras at near equator regions, ordinarily “polar light” belongs in the auroral rings around the poles<br>3.) Peak excursion time projection varies depending on which parameters you use, rough estimate is between 2035 and 2046<br>4.) At peak all we need is a strong CME direct impact to lose electricity across the globe, which is inevitable over the aforementioned period<br>5.) Loss of the electrical grid leads to 90%+ population loss, and anarchy</p><p>Now… Let’s get to it.</p><p>As already stated, in the past 3 years we’ve had the sky over Croatia turn red 6 times. So has virtually every place on Earth that is positioned at this latitude. This is because auroras are not a contained and localized phenomenon but span over a ring-like area starting at the magnetic poles and stretching southward — in the case of the north pole, and northward — in the case of the south pole. The intensity of the aurora phenomenon, as well as how far it penetrates from the poles — depends on two things: strength of the Earth’s shield, and the “strength” of solar particles ejected by the Sun.</p><p>The Sun hasn’t changed its activity over this long span we are observing, which means, the Earth’s shield is weakening. Dramatically so. And on an accelerating schedule. We’ll cover all the key evidence for this.</p><p>The entire written record of Croatia pertaining to auroras spans 286 years and it contains about 9 aurora nights the likes of which we’ve witnessed the past 3 years. The geophysicists who compiled that record described aurora at Croatian latitudes as visible “approximately once or twice during someone’s lifetime”. Allow me to remind you, auroras are called POLAR light. Croatia is nowhere near a pole. And still… the general reaction of people is “oh, pretty lights”. Then they move on with their day.</p><p>What “pretty lights” actually mean: Earth’s magnetic field has weakened and rearranged enough that solar storms which used to leave Croatia only a photography-visible aurora, or none at all, now light up the whole country’s sky.</p><p>This, coupled with other evidence we’ll cover, means one thing and one thing only — we are in a process called geomagnetic excursion: the state in which the planet’s magnetic shield drops to a fraction of its strength while its geometry comes apart. The same storms that make the “pretty lights” also drive current along every long wire on this planet, into the transformers at the ends of those wires. The large transformers the grid hangs on are ordered years ahead and built by a handful of highly specialized factories — factories that run on grid electricity. This means that if the grid is taken down by one of these storms, they cannot simply be rebuilt and shipped for replacement. Humans can adapt, and fast — but the old way of life will be over. We’ll get a chance to start over, which is both terrifying, and alluring, since as a species we’ve lost our way. Nobody can argue that is not the case.</p><p>We’ve already built our hell, we are living in it. The unavoidable reset is a chance to start over from a completely blank slate. Carrying all the valuable experiences from our past failures. The more people understand the stakes, and what to do — the better. That is essentially the purpose of this essay.</p><p>Let’s continue with the auroras.</p><p>The 6 nights that are documented in the past 3 years are documented explicitly via naked-eye determination in the Croatian astronomy community’s own record: 5 November 2023, 10 to 11 May 2024, 10 to 11 October 2024, 1 January 2025, 12 November 2025, and 19 to 21 January 2026. That is 6 nights in 33 months, counting from 5 November 2023 to the month I have written and published this essay.</p><p>Important distinction: 3 more genuine Croatian aurora detections from the same window — 12th August 2024, 12th September 2024, 6th October 2024 — were photographed and not considered in this key comparison: 6 auroras in 3 years versus 9 auroras in 286 years. The reason is simple. The sources say they were too faint for the unaided eye. If a camera was needed, it does not count. Only what a person standing outside can see does. This allows us to make a “scientific” comparison. Which this essay is about. Adding the 3 photography auroras to the comparison would be unfair because people in the 18th century did not have cameras.</p><p>This essay is only about cold hard facts, not fear porn, or delusions.</p><p>Speaking of delusions… recently, whenever one of these red nights happens, some bright lightbulbs on social media clamour “it’s HAARP, it’s HAARP!”. It’s not HAARP. The amount of electrical energy you would need to pump into the atmosphere to produce an aurora over the entire globe is far beyond anything we have, secret technology included. Besides, every one of these nights had a recorded solar storm behind it, seen coming by everybody who monitors the Sun. Including the horde of space weather amateurs that count in tens of millions across the world.</p><h3>The Catalogue</h3><p>The aurora comparison is the smoking gun of this essay and crucial evidence because it’s so visible to the “regular person”, but also because it truly is extremely alarming.</p><p>So let’s provide the source behind my statements. There is a catalogue. Two geophysicists at the University of Zagreb, Lisac and Marki, went through monastery annals, ship logs, meteorological diaries and questionnaires. Then they published the lot in the journal Geofizika in 1998, every aurora recorded from Croatia and its neighbouring regions between 1737 and 1991. It contains 37 auroral event-days across 254 years, falling in only 23 separate years. Days, not displays. One storm can keep the sky lit for 2 to 3 nights. What I’m counting, on both sides of the comparison, is storms that put a display over the country, not calendar nights it stayed up. So we are looking for a comparison between the number of storms that caused an “almost reached equator” aurora in these periods.</p><p>Most of those days in the catalogue are very faint auroras, the opposite of what we’ve recently experienced, and 6 of the 37 carry no description at all. Why catalogue 6 events without any description? Six entries nobody bothered to describe were six nights that were certainly poorly visible, or perhaps even a false determination. They go in the raw count of the catalogue and nowhere else. The typical entry reads like the 1873 one from Pula: “weakly pronounced aurora, the rays were not seen this time.”</p><p>So the 37 is raw material, not a count of big auroras. An entry nobody described was not a display that was visible to everyone. The big ones have to be picked out, as I’ve already explained — by a rule set before any counting starts.</p><p>My rule: a conspicuous red display, reported independently from more than one place, or described in the record as exceptionally bright or exceptionally far-reaching. About 8 entries clear it firmly, up to 11 if you admit every ambiguous 19th-century case, plus the 20 November 2003 super storm aurora known to have been very visible, which isn’t in the catalogue. This is how we get to 9 in 286 years — by adding the one storm from 2003, because the catalogue stops in 1991. That’s when it was published. So, to summarize, my comparison starts from the same 1737 mark as the catalogue’s, to November 2023, and the 2003 night is counted on the historical side.</p><p>So those 9 were the real thing: whole-sky red displays that people noticed and wrote down. On 18 January 1770 a friar in Osijek watched a diffuse redness organise itself into an arc from the Great Bear to Cygnus, for 2 and a half hours. On 25 January 1938 the display turned sailors and fishermen back to shore, and the Geophysical Institute in Zagreb posted questionnaires across the country and got about 50 replies within days.</p><p>That is what a big Croatian aurora is. There are 9 of them in the entire national record.</p><p>Nine displays in 286 years is 0.031 a year. When we concede every ambiguous entry in the archive we get 12 in 286 years, or 0.042 a year. Six auroras in 33 months is 2.18 a year. Divide, and the modern rate comes out 69 times the historical rate on the strict count, 52 times on the generous one. That is a 50x-70x increase.</p><p>At the historical rate, 6 displays take about 190 years. Croatia got them in 33 months. NOBODY in that entire record ever saw 6 of these. One lifetime got one, maybe two, decades apart.</p><p>I’ll repeat again, the Sun hasn’t significantly changed its activity during these compared spans of time. And that’s what makes this extremely alarming. Yes, there is something called the Gleissberg cycle where the Sun’s activity does vary, but it’s an 80-year cycle that gets averaged out over 286 years. That’s why I say “the Sun’s activity hasn’t changed”.</p><p>To make a further point: the aurora catalogue authors Lisac and Marki made a forecast. In 1998 they looked at the 1991 displays, and stated that an “aurora-favourable interval” had not begun. They were working off the 80-year Gleissberg cycle of solar activity. On that clock the next favourable interval wasn’t due until about 2030s, and a favourable interval means 8 to 11 events spread over roughly 30 years. Their forecast: quiet through the 2020s. What arrived: the densest cluster in the whole 286-year record, starting 7 years before their window opened, at 50 to 70 times the historical rate. So they were quite wrong. Because something has changed.</p><p>Furthermore, Croatia has been drifting away from the auroras’ “point of origin” — the magnetic poles, across the entire span of its own catalogue, and the modern era is the furthest it has sat from the auroral ring, magnetically, in 4 centuries. The plain version: magnetically, Croatia has slid further from the lights than the country is long. The drift since the first entry in the catalogue is about 700 km. Croatia, from the Mura to Prevlaka, is about 460 km. A storm that put the ring’s edge over Croatia in the 1730s would fall 700 km short of it today. Every friar, sailor and station-keeper in that record that was witnessing auroras was doing so from a position 700 km more favourable to do so. All 6 modern ones happened after it moved 700 km away. Which in the “aurora world” is a lot.</p><p>The northern auroral ring is 2,500 km away from Croatia. The ring where ordinarily auroras are visible. That’s an increase of 28% in the distance that an aurora has to cover to be visible in Croatia.</p><p>Just to pre-emptively answer any potential objections — over the last 22 years the warped part of the field has actually pulled Croatia about 20 km back toward the lights. We’ll get to the warping field later. So 20 km against 700 km of drift, and against 2,500 to the ring on a quiet night, is a rounding error, not something to consider as a factor.</p><p>Where do the lights actually belong? The aurora is a ring, not a cap, centred on the magnetic pole, and distance to it is measured in magnetic latitude, not map latitude. Croatia’s map latitude is 45.8 degrees north. In corrected geomagnetic latitude — the coordinate the aurora is organised in — it sits at about 40.5. That’s some 590 km of difference, and nearly all of it, about 500 km, comes from the warped part of the field over Europe, the part the textbook bar magnet leaves out. On the map Croatia is a mid-latitude country. Magnetically it’s a low-latitude one. A crucial distinction.</p><p>On an ordinary night the ring’s southern edge is 2,523 km north from Zagreb, on a line passing essentially through Tromsø in arctic Norway. A very strong solar storm does not shove the ring south — it expands it. NOAA’s rule of thumb, about 2 degrees of latitude per step of the 0-to-9 Kp storm scale, puts the edge 706 km away at Kp 9, just south of Berlin. When the sky over Croatia turns red, the aurora is over Berlin and Croatia is watching the top of it over the horizon.</p><p>And that rule understates real storms. During the biggest storm of this solar cycle, an all-sky camera network tracked visible aurora down to 46 degrees north geographic. Zagreb is at 45.8. The visible aurora came down to Croatia’s own northern border, the fainter red arc went south of it. So under a big enough storm, the ring’s edge already reaches the country. Again, alarming. Something has changed. We’ll learn what.</p><h3>The Change</h3><p>Solar cycle 25 is a strong cycle. Every one of the 6 Croatian nights had an identified solar storm behind it. A strong cycle naturally involves more solar storms that produce auroras. But not on low latitudes. What a strong cycle also does not explain is the difference in response of the sky to the same two solar storms, one in the past, one now. Same solar storm, two radically different results. We will get to the evidence a bit later, the crucial understanding is this: the Earth’s shield moves slowly, it doesn’t flicker, so this can only be a prologue into something even more extreme.</p><p>Boris Štromar is a Croatian astrophotographer, and he has stood on the same ridge for an aurora photoshoot, in Croatia, twice. Both times because a forecast told him to go up. On 17 March 2015 he drove up Sljeme, the 1,000-metre ridge above Zagreb, for the deepest storm of solar cycle 24 — ring current bottoming at −234 nanotesla in Kyoto’s final index, the standard depth gauge for a storm. He wrote it up that same night: seen from Sljeme, “ali nažalost samo na fotografijama” — but unfortunately only on the photographs.</p><p>Two other people were out photographing that 2015 night. Glynn, a photographer in Wiltshire, 5.5 degrees further north than Sljeme, captioned his frame “Dim green/red glow to the North, not visible to the naked eye”. Ortar, a photographer shooting from Slovenia, called it barely visible and wrote that his camera “got much more from that phenomena”.</p><p>On 5 November 2023 Boris Štromar went up the same hill for a storm that bottomed at −163 nT, and red curtains stood 30 degrees above the northern horizon over practically the whole country, on the national news.</p><p>So -163 nanotesla produced an intense aurora covered on national news, and -234 nanotesla produced an aurora visible only to the camera. The bigger storm needed the camera. The smaller one that happened 8 years later needed nobody’s.</p><p>Something has been changing, and has changed radically, between the 2015 camera-only aurora and 2023 national news-aurora.</p><p>Now let’s go a bit further back, to the solar storm that caused the only aurora on record between the last one in the Geofizika journal, and the 2023–2026 aurora run. Because it contains another piece of evidence which points in the same direction.</p><p>Both on 20 November 2003 and 5 November 2023 we got red aurora over Croatia that people saw with their own eyes, 20 years apart. The 2003 one was the strongest storm of solar cycle 23.</p><p>It had a ring current of -422 nanotesla, and a peak auroral electrojet — the current flowing about a hundred kilometers up that actually makes the light — of 4192 nT.</p><p>The storm in 2023 had a ring current of −163 nT and an electrojet of 1845 nT, a G3 storm at Kp 7, a level that arrives several times in an active year. That is 39% of the ring current and 44% of the electrojet, on the same instruments at the same processing level.</p><p>Granted, the display shrank too — 45 degrees up the sky in 2003 against 30 in 2023. But the display lost a third, while both were intense, and had a stark contrast in physical parameters producing them.</p><p>That’s the whole point of the last few paragraphs, if you got a bit lost: the storm was measured, and it was less than half the storm. The sky was measured, and it was nearly the same sky. There is exactly one thing related to those two measurements, and it’s Earth’s magnetic field. So what you’re looking at in that comparison is not the Sun doing something unusual. It’s the shield failing to do what it did in 2003.</p><p>Across all 6 recent events, on the official planetary index, 5 November 2023 came in at Kp 7.333, against 8.0, 8.667, 8.667, 8.667 and 9.0 for the other 5. The weakest of the 6 by a clear margin is the one that made the national news.</p><p>Croatia used to need the biggest storm of a solar cycle for a sky like that. Now it needs one that comes around several times a year.</p><p>Now if we compare the 20 November 2003 solar storm against the 10 to 11 May 2024 one we get a very useful insight that demonstrates the thesis behind this essay. The ring-current difference between the two storms came out 2.4% apart and the peak auroral electrojets 2.2% apart, from the same archive at the same processing level. The team that wrote the reference report on the 2024 event states that the solar wind speed and the peak southward field matched too. Identical storms, 21 years apart.</p><p>What separates them is how far toward the equator human eyes could see the light. The recorded numbers that demonstrate the massive difference are magnetic latitudes. The scale is simple: 90 at the magnetic pole, 0 at the magnetic equator, one degree is about 111 km. The auroral ring reaches out to around 66 degrees on a normal night. Croatia is located at 40.5 degrees latitude.</p><p>In May 2024 the aurora was seen with the naked eye at 23 degrees — from a farm in Namibia that sits on the Tropic of Capricorn, the southern edge of the tropics. Those are the southern lights, pushed up from the south pole’s ring to three quarters of the way to the equator, and they could be seen at 45 degrees up in the sky, halfway between the horizon and straight overhead. The same night the northern lights were seen by eye at 24 degrees, from Hanle, a village high in the Indian Himalayas. Hanle is 32 degrees north of the equator on the map and only 24 magnetically, because the magnetic pole leans toward Canada, which puts all of Asia magnetically nearer the equator than on the geographic map, and all of America further from it. So from both poles, the same storm carried light that people could see to within 23 to 24 degrees of the magnetic equator. Croatia, at 40.5, was well inside the area that saw it, by about 2,000 km.</p><p>Now the identical storm in 2003. The best surviving first-hand description from the low-latitude end of that night comes from Albuquerque, in the New Mexico desert. On the map it’s 35 degrees north, about as far north as Crete. On the magnetic scale it’s 43, about where Croatia is, for the same Canada reason. The observer’s words: “These were not the vivid colors of green and red like the lucky residents farther north get, but they were nice and rosy.” A pink glow. So the same size of storm that in 2024 stood halfway up the sky over the Tropic of Capricorn was, in 2003, a pink glow at Croatia’s magnetic latitude.</p><p>Here’s proof that this isn’t cherry-picked data. A 2025 survey went through 223,916 auroral records from 1650 onward, corrected every observing site for how the field has moved since, and ranked the events by the lowest magnetic latitude anybody saw them from. The second night of May 2024 comes third of ALL TIME, behind 4 February 1872 and the Carrington nights of 1859. The 2003 storm that matches it on every measure of storm strength is NOT ON THE LIST. Identical storms, and only the later one carried the light to within 23 degrees of the magnetic equator.</p><p>“It’s just better cameras and phones” some might say. No. The team that reconstructed the May 2024 event raised that themselves — they list the modern witness’s 3 advantages: better cameras, storm forecasts, social media. Here is the problem with the camera story: cameras are how Croatia detected aurora through the whole quiet interval — 2000, 2003, 2005 and 2015 are in the record because somebody photographed them.</p><p>What changed after November 2023 is not that the pictures got better. People stopped needing pictures. That’s the point. Three events in 2024 alone were photographed in Croatia and thrown out of my count as too faint to see. And the gap between eye and camera has been measured. On the May 2024 night, from Hiranai, a town at the northern tip of Japan’s main island, the same display reached 20 degrees up the sky by eye and 70 by camera. Work backwards from how high the light stood to where the ring’s edge had to be, and the eye puts the edge at 43 degrees magnetic while the camera puts it at 37. About 6 degrees difference, roughly 700 km, Zagreb to Berlin.</p><p>That is what a camera adds: it picks the aurora up while the ring’s edge is still 700 km further away than an eye needs it to be. So an aurora photographed over Croatia and an aurora seen over Croatia with the naked eye are two different events, 700 km apart in how far the ring actually came. That is why my count uses the stricter one on both sides.</p><p>Besides, the cities are far brighter now than in 1938 — the lights of Koper and Trieste were already washing out the view from Istria back during the 1991 aurora — yet Croatians still saw the October 2024 display from just outside the larger towns. If anything gets suppressed by modern skies, it is the modern count: a faint aurora over a lit-up town simply goes unseen.</p><p>The same 2025 survey of 223,916 records also tested whether cameras invented the modern ranking, correcting for how long human eyes take to dark-adapt, and found no evidence that they did. A display you can see is a categorically bigger event than a display you can photograph, and Croatia crossed from one to the other.</p><p>Meanwhile, the models that predict how far south the aurora reaches keep coming in low — all of them, in the same direction. For 24 to 25 March 2024, NOAA publicly issued a G2 storm watch for the 24th and a G3 for the 25th, having already seen the X1.1 flare. Small interjection — X is the top class of solar flare, the number after it the intensity.</p><p>Back to 24 to 25 March 2024. NOAA predicts a G3 storm, instead a G4 arrives. This isn’t a small “whoopsie” error. It’s a massive difference. And it keeps happening.</p><p>Once again, for emphasis: NOAA, which does this at the highest level there is, has spent the last few years getting the storms’ forecasts wrong — consistently, and always in the same direction. Too small. A model that misses at random is a bad model. A model that misses the same way every time is a model whose planet has changed.</p><p>Here’s the worrying part. For that same 24 March 2024 storm, NOAA used the Solar Orbiter spacecraft positioned between the Sun and us, on the ejection’s path, to measure the impact by measuring the cloud of solar plasma passing by it on its way to Earth. The team measured the cloud’s own magnetic field directly, in flight, fed the measurement into their model, and predicted where the ring current would bottom out: −88 nT, on the same depth gauge as before, where more negative means a deeper storm. It came in at −130. Half again deeper. A week earlier the same team, with the same spacecraft and the same method, had predicted −77 for the other March storm and got −77, to the number. Sometimes they still get it right. So the model works, and the Sun’s side of this was measured, not guessed. The extra depth came from our side.</p><p>Their own explanation is that Earth’s field was still disturbed from an earlier hit when this one landed. Notice what kind of explanation that is. It says nothing about the Sun. It says that how bad a storm gets depends on the condition of the shield at the moment the punch lands. That is this essay’s claim, coming out of their own data: the punch was measured, and the shield flinched harder than the punch accounts for.</p><p>Keep in mind what a red sky does and does not prove. A red aurora is made 200 to 400 km up, roughly at the height where the space station orbits, and light made that high clears the horizon from about 1,750 km away — roughly Zagreb to Helsinki. So red over Croatia does not by itself put the ring over Croatia. Green does.</p><p>Green is made at about 115 km and cannot be seen over the horizon that way. During the May 2024 storm, green was photographed down to 35 degrees of magnetic latitude and seen with the naked eye at 38 — aurora low enough to be local, and it survives the horizon problem untouched.</p><p>So, to summarize:</p><p>1.) Croatia’s rate of conspicuous auroras is up 50x-70x against its own 286-year record.<br>2.) Croatia had the worst position for aurora visibility in the last 4 centuries — magnetically further from the ring than when any of the old displays were written down — so the real increase is bigger than the counted one.<br>3.) A red sky that used to need the strongest storm of an 11-year solar cycle, 20 years ago, is now produced by a storm that arrives several times a year.<br>4.) Two storms with matching numbers, 21 years apart, and only the 2024 one was followed by eye to 23 degrees from the magnetic equator.<br>5.) The camera explanation was tested thoroughly by academics, and it failed.</p><p>One thing has to be true for all five at once: Earth’s magnetic field — the only thing standing between the Sun and the ground — is no longer doing what it did 20 years ago. And the change is accelerating.</p><p>Now put on your thinking cap, it’s time to go from visible phenomena, into non-visible, and sciency stuff.</p><h3>The Field</h3><p>The way the field changed is the genuinely alarming part: the field is not simply fading, it is being rearranged — and the rearranged field responds to the same solar storm with more current. Consequently producing more aurora light.</p><p>To see that, you first have to throw out the number everybody reaches for. The dipole field number. Some genius will build the whole story on “the field is 9% weaker since 1840, so the aurora reaches further”, and the first specialist who reads that will take it apart, rightly. The published scaling puts the latitude of the polar cap’s edge at the sixth root of the DIPOLE moment — the strength of the planet’s basic bar-magnet field. Feed the 9% loss through it and the boundary moves about four tenths of one degree. Which is about 40 kilometers. Nothing, with respect to Croatia’s distance from the ring. The dipole percentage reduction on its own cannot produce a single one of the 5 facts listed above.</p><p>What can is the part that most people don’t mention. The DIPOLE — the simple planet-wide bar magnet from your school textbook — decides where the centre of the ring sits. The NON-DIPOLE field — everything the bar-magnet picture leaves out, the lumps and patches, the true structure of the shield — decides how far the ring’s edge actually expands. And how close the polar light reaches out towards the equator.</p><p>Zagreb has 2 magnetic latitudes. On the textbook definition, the best-fitting bar magnet through Earth’s centre, it sits at 45.1 degrees. On the definition the aurora actually follows, which traces the real field line and therefore contains the non-dipole part, it sits at 40.5. About 500 km apart. Something we’ve already established, but should be emphasized once again, because Croatia being a magnetically low-latitude country is the alarming part, in relation to the last 3 years’ aurora run.</p><p>Between the Halloween storms of 2003 and October 2025, the dipole moved Zagreb 1.8 km away from the auroral zone, while the full field — non-dipole included — moved it 22 km toward it. Net, 20 km closer in 22 years. Against 2,500 km distance from the auroral ring on an ordinary night. Note the massive discrepancy. The rearrangement of the field did not make Croatia suddenly have auroras by moving the ring by a measly 20 kilometers.</p><p>So then, if the weakening barely moved the auroral ring toward Croatia, what did it actually do?</p><p>It changed how the Earth’s shield reacts when a solar storm hits it. Take a global simulation of the whole system, trigger a substorm — the sudden release of stored energy that makes an aurora flare up — and run the simulation again with Earth’s dipole made stronger and weaker by a factor of 1.5. The substorm happens in both scenarios. But with the weaker dipole it starts later, the auroral electrojet builds more slowly, and then peaks higher.</p><p>The electrojet is basically an enormous electrical current flowing through the upper atmosphere, roughly 100 kilometers above us. It is one of the main things producing the aurora light. So the important effect of weakening is not simply “the lights move south”. The same solar input can produce a stronger electrical response inside the Earth-atmosphere system. More current in the auroral system means brighter auroras, stronger magnetic disturbances, and consequently more current being induced into long conductors on the ground. Power lines included. And a measurable effect on human and animal biology, which we’ll get to — it’s confirmed, and it’s not small.</p><p>Back to the non-dipole I keep emphasizing. “The non-dipole field is GROWING” is a sentence that will mislead you completely unless you understand what is actually growing.</p><p>The dipole is the organised part of Earth’s field. Imagine the clean bar magnet from school, one north pole, one south pole, with relatively orderly field lines wrapping around the planet. The non-dipole field is everything that departs from that clean structure: regional bulges, weak spots, strong spots, reversed patches and other distortions that the simple bar-magnet picture cannot reproduce.</p><p>So when I tell you below that the non-dipole field is GROWING, that does NOT mean part of Earth’s protective shield is becoming stronger. Quite the opposite interpretation is relevant here. What is growing is the amount of the field contained in its irregular, regionally distorted structure, while the organised planet-wide dipole loses ground.</p><p>That is what I mean by increasing instability of the field. Not that the magnetic field is randomly flickering from one minute to another, but that the stable, simple dipole is becoming less dominant while the distorted structure underneath it becomes more important. It’s like a double-whammy. The shield is not merely getting thinner. It is becoming increasingly malformed. Malformation is weakness. And the aurora follows that distorted field.</p><p>There is one measurement that initially looks like it contradicts all of this. Until about 1940, the auroral rings in both hemispheres expanded together, reaching further towards equator, exactly what you would expect as the dipole weakened. Then something strange happened. The southern ring kept expanding, while the northern ring — ours — started shrinking. That result appears both in the magnetic-field reconstruction and in 34 years of satellite measurements of the actual auroral boundary.</p><p>At first glance that sounds reassuring. If the northern auroral ring is shrinking, how can I argue the changing field is causing auroras at lower latitudes? Two reasons, and neither is a deflection or a retraction.</p><p>First, let’s consider why it shrank. A ring shrinks when the field under it gets relatively stronger and grows where it gets weaker — and since 1940 the non-dipole field has been sagging south, into the South Atlantic Anomaly. The north tightened because the south gave way. So the shield is not recovering, it is becoming lopsided, malformed. Which is the entire thesis of this section.</p><p>Second, let’s consider what actually shrank. The quiet-nights ring did. Not the storm nights. Croatia’s 6 nights happened on storm nights, and what turns a solar storm night into an aurora night is not where the ring ordinarily rests but how the field responded to the storm, the electrojet that, in the simulation previously mentioned, starts later and peaks higher under a weaker dipole.</p><p>Now we get to the number that shows just how radically the field itself is being reorganised. The dipole has been losing about 12% of its magnetic energy per century. Over the same period, the non-dipole field has been gaining about 70%.</p><p>Read that carefully, because “gaining” is the word that might confuse you.</p><p>It does not mean Earth gained 70% more protection. It means the irregular part of the magnetic field has been accumulating energy roughly 6 times faster than the organised dipole has been losing it. More and more of Earth’s magnetic structure is therefore contained outside the clean bar-magnet component.</p><p>These are energy numbers, not ordinary field-strength numbers, which is why elsewhere you will see the overall decline described as roughly 5% per century rather than 12%. They are measuring different properties of the same changing system. The important relationship remains: the organised dipole is slowly fading while the distorted non-dipole structure is changing much faster.</p><p>And this reorganization has a surprisingly clear starting point. Earth’s field strength remained nearly constant from about 1590 until 1840. Then it began falling at roughly 5% per century. Field models trace that decline to patches of reversed magnetic flux in the southern hemisphere — regions where part of the field points the wrong way relative to the dominant dipole. Those patches are essentially absent from the field before the modern decline began around 1840.</p><p>There is another important detail here. The dipole’s decline itself is not accelerating. Its rate of decay actually roughly halved between 1980 and 2015, from around 24 nanotesla per year to about 10.5.</p><p>So to emphasize, once again. The dipole is fading slowly. It is worrying, but only in the long run, something that could be a concern of your children’s children. What should concern you is the fact that the non-dipole structure is growing far faster. And because the real magnetosphere, the aurora and the electrical currents inside it follow the actual distorted field rather than an imaginary bar magnet through Earth’s center, that rearrangement is what increasingly determines what a solar storm does over any particular part of the planet.</p><p>So the picture is not “Earth’s magnet is 9% weaker, therefore the aurora moved more than 1,000 kilometers south.” That is not the cause.</p><p>The picture is more disturbing. The organized part of the shield is both fading AND losing dominance. The irregular part is growing. Regional weak spots and reversed patches are becoming more important. And a weaker, more distorted magnetic system will suffer much more dramatically from the same solar forcing. The 2003 vs 2023 scenario I’ve discussed thoroughly at the top of the essay. Same storm, or even weaker storm, much more pronounced effects.</p><p>And the key understanding is this: more current in the sky is more current in the wires that provide your homes with electricity. It also causes a more pronounced effect on the human biology.</p><p>Nowhere on Earth is this problematic rearrangement of the Earth’s shield easier to see than the South Atlantic Anomaly.</p><h3>The South Atlantic Anomaly</h3><p>The weakening of Earth’s field is not spread evenly. It is concentrated in one region called the South Atlantic Anomaly (SAA). It covers most of the South Atlantic and a good part of South America and southern Africa, and the field at the surface there is far weaker than anywhere else on Earth — a little over 22,000 nT, against more than 57,000 nT in the strong patch under Canada.</p><p>The effects of this “hole” in the Earth’s shield aren’t abstract. Satellites take radiation hits every time they cross it, and a laser instrument on the space station registers a detector fault about once in a million shots there and needs a monthly power reset.</p><p>ESA’s 3 Swarm satellites have flown magnetometers through it since 2013. Over the 11 years to 2025, the area where the field drops below the mission’s chosen threshold grew by 0.9% of the whole planet’s surface — about 4.6 million square kilometers, roughly half of continental Europe — while the lowest field value inside it fell from 22,430 to 22,094 nT. NOAA, working from its own separate definition, reports its weak-field area growing about 8% in a single year.</p><p>And the anomaly is coming apart. Spreading into two distinct anomalies. This is a fact, and it clearly demonstrates the increased chaos in the shield. The main anomaly under South America appeared in the 19th century, and the second one appeared recently as a breakoff from the SAA near 40 degrees south on the Greenwich meridian, confirmed across 3 separate model chains, with the African side losing about 600 nT over 7 years against about 200 nT on the American side. So the second “anomalous” weakening is growing 3 times faster. The weakest part of Earth’s shield is spreading, deepening and splitting in two, and whatever is happening to the planet’s field is happening fastest where it was already thinnest.</p><p>How much shield has actually been lost there? The number that answers is called “cutoff rigidity” — the energy a charged particle needs before it can get through to the atmosphere at a given spot — and it is not a parameter proxy for the shield but the shield itself.</p><p>Inside the South Atlantic Anomaly it fell from 6.763 to 4.482 gigavolts between 1965 and 2025 — down 34%. And it’s accelerating! At the same study’s northern reference site, over the identical 60 years of the identical global decline, it rose from 4.234 to 5.257 gigavolts — up 24%. Same planet, similar average, opposite signs, which is why a global dipole percentage does not justly describe the situation. A third of the shield, gone from one region in a single lifetime, while the global average moved so little that nobody had to notice.</p><p>And where does the anomaly come from? From underneath.</p><p>The shield is generated in the outer core, 2,900 km down, and at the top of the core under the South Atlantic there are patches where the field points the wrong way — back into the core instead of out of it. Those reversed patches are the distortion. Field models trace the South American minimum to reversed flux beneath western South America and the southern Atlantic, and the African minimum to an intense reversed patch below South Africa. Those are the same structures identified as the origin of the entire modern decline since 1840. The hole over the South Atlantic and the weakening of the whole planet’s field are not two stories. They are one. The story of this essay. The story of the geomagnetic excursion that will have profound effects on humanity.</p><p>The whole field has collapsed like this before — and this process is called a geomagnetic excursion. The last one has been reconstructed in detail. During the Laschamp excursion, 41,000 years ago, the field fell to roughly 10% of today’s strength (we are at 9% loss today, in comparison to previous records!) and went multipolar: many weak poles scattered over the planet instead of one pair. Exactly what is happening with the South Atlantic anomaly and the newly formed African anomaly.</p><p>The daylit boundary of the magnetosphere, which today sits at 8 to 11 Earth radii, came in to 2.43. The auroral ring’s diameter went from under 3,000 km to 8,167, and the northern ring reached from the Arctic across Eurasia into the northwestern Sahara. Again, exactly the process we are witnessing today.</p><p>The shape of the event matters more than its size. The Laschamp took about 2,000 years overall, with a collapse-and-tilt phase of roughly 300 years nested inside it. Now a reminder, the “fast phase” of our current geomagnetic excursion started in 1840, roughly 200 years ago. Second important distinction we can learn from Laschamp excursion — a slow event with a fast phase buried inside it does not announce the fast phase in advance. A digression while we’re on Laschamp: at its peak, with aurora over the Sahara, the rock and the bones record a bad time for life both in the sea and on land. So if the news ever reports aurora over the Sahara while the grid is still up and you haven’t started prepping — consider yourself lucky, and start.</p><p>One more thing for any naysayers. Someone might claim that the excursion started only 200 years ago and that we’ve got at least a thousand years before it gets rough, judging from past excursions. That is incorrect, the reconstructed dipole strength has been steadily declining since 400 BC. That’s 2400 years.</p><p>An important question arises from all this. What does losing shield cost, per unit lost? Take a coupled model of the magnetosphere, ionosphere and upper atmosphere, feed it identical solar wind at an identical level of solar activity, and take a quarter off the dipole moment — from 80 down to 60 zetta-amp-metres squared, the standard unit for the strength of a magnet: amps times the area the current loops around, with zetta meaning a thousand billion billion.</p><p>The ionosphere — the electrically charged layer of the upper air, starting about 80 km up, the layer the electrojet flows through — gets 50 to 65% better at carrying current. The currents flowing down the field lines, from space into that layer, rise about 20%. The heat the same solar wind dumps into the upper atmosphere rises by 13 to 30%, depending on the season.</p><p>So that is the coupling, and here is how it works. The solar wind is a generator. The upper atmosphere is the load it drives current through, and the field lines are the wires between the two. Up in that layer a charged particle cannot simply move in the direction it is pushed, because the magnetic field holds it on its field line, spiralling around it. Push it with an electric field and it slides sideways instead, and the speed of that slide is the push divided by the field strength. Same push, weaker field, faster slide, and a faster slide of charge is a bigger current.</p><p>That is the whole rule: the upper air’s ability to carry current goes as one over the field strength. The auroral electrojet, the current I have been using to price Croatian skies, is exactly this current, electrons sliding sideways through the layer 100 km up. So a quarter less field is a third more current-carrying capacity before anything else changes, and the model found more than that, half again or better, once the shield’s shrinking and the widening of the openings over the poles are in the run. The rest follows like a circuit. A generator pushed into a load that conducts better gives slightly less voltage and more current: the voltage across the polar cap falls about 10%, the current down the field lines rises 20%, and the heat, which is conductance times voltage squared, comes out 13 to 30% up. Run the first two numbers through that and you get about 25%, so the three numbers are one number seen from three sides.</p><p>The dipole has not fallen 25% — that was a sensitivity experiment, a deliberate “what if”. What it establishes is the response itself, and that is the whole point, because one over the field strength is a curve that gets steeper as the field gets weaker. The dipole has lost 9% since 1840: one over 0.91 is 1.10, about 10% more current-carrying capacity, and that is the part we have already paid. A quarter off: one over 0.75 is 1.33, a third more by the arithmetic, half again or better in the model. Half off: one over 0.5 is 2. It doubles. A tenth of the field, where the Laschamp went: one over 0.1 is 10. Tenfold, and that is before the model’s extra. The first 9% bought 10%. Each step down buys more than the step before it. Everything in this system is back-loaded, and the losses so far are the cheap ones.</p><p>One thing that ladder is not, and I want it said before somebody says it for me: it is not a schedule. The planet-wide average is falling about 5% a century, and on a straight line that bottom rung is more than a thousand years away. If the planet-wide average were the right parameter to discern the timing of this excursion, this essay would be addressed to your distant descendants. But that is not the case. For two reasons the rest of this section gives. The record’s fastest changes happened at the low values, inside a fast phase that the slow decline does not announce. And the floor of that fast phase, half the field’s long-run average, is not reached everywhere at once. It is reached somewhere first, and the weakest spot on Earth is already almost there.</p><p>So how far along is this process? The rock records tell us more about this through a measurable parameter called the local virtual dipole moment: take the field measured at one spot on the ground and ask what size of magnet at the centre of the Earth would produce that reading. The zetta-amp-metres squared are then divided by a common unit, which gets us that the planet’s whole dipole today is about 77 of these units, and the average over the past 100,000 years was about 70. So our shield is stronger? There is no weakening? Not quite. That’s a 100,000 years average. The devil’s in the details, they say.</p><p>Across those 100,000 years there is a line in that record at 35, half the 100k average, and it marks the deep phase of an excursion, not the start of one. Wherever a spot on Earth fell below 35, that spot was in an excursion, and the fastest field changes in the whole record happened at the lowest values. Crossing 35 causes nothing specifically on its own, but comes with a number of challenges for life forms on the planet.</p><p>Reaching a value of 35 just signifies that we’ve dropped to the floor these events reach after an approach that takes millennia. Likewise it’s important to know it’s not just about thresholds, but the massive decline versus the field present a few millennia before it. So we have both relational and “threshold” data to consider. Both of these are evident from Earth’s shield graphs for our modern times.</p><p>So the question is not whether an excursion is currently in effect. Even the magnetic poles are moving fast which fits the scenario, from around 10 to 15 km a year through most of the 20th century to roughly 55 km a year around the turn of the millennium, about 35 now, out of Canada across the Arctic toward Siberia. And the South Pole moved out from the icy Antarctic and into the sea, heading for the equator.</p><p>According to latest data they seem to have started “sprinting” again at 55+ km. Not fully confirmed yet, but highly likely. The difference between the pole that historically crawled around the geographic pole and the modern pole that sprints towards the equator is one of the symptoms of the growth behind the non-dipole field. The continual growth of the malformation in the Earth’s shield.</p><p>The real question is then: how far in the geomagnetic excursion are we.</p><p>I went and found the lowest point on Earth today. It sits at about 37 degrees south on the Greenwich meridian, in the African lobe of the South Atlantic Anomaly, and its local virtual dipole moment is 37.2 — one spot’s reading, which is why it sits so far below the whole planet’s 77. Hint, hint. Read two paragraphs above if you are not alarmed.</p><p>In 1980 the lowest point on Earth stood at 44.1. The line is 35. THREE QUARTERS OF THE DISTANCE TO THE FLOOR, covered in 45 years, while the planet’s number moved 2 to 3%, also in a downward trajectory. Do you understand?</p><p>Now as far as the projections are concerned, they vary, nobody knows for sure when the shield will give way.</p><p>The rate of fall at that patch in Africa is slowing down, although still moving downwards at a high speed — the successive readings since 1980 go 0.180, then 0.137, then 0.124, then 0.120 zetta-amp-metres squared a year — and the field’s tilt at that spot is flattening, which offsets about 18% of the fall.</p><p>So the dates are extrapolations, and I write them as extrapolations. At the current rate, the patch reaches the line in the early 2040s. At its 1980-to-2000 pace, the late 2030s. If its own slowdown continues unbroken, about 2046. Nothing happens on the day the line is crossed — reaching it is arriving at the floor of a process already running. But look at what those dates are dates of: the deep phase, the state in which the fastest field changes of the last 100,000 years took place. Every one of these projections falls inside the lifetime of almost everybody reading this. If you have children, all the more relevant.</p><p>Which lets me say what I already stated, this time with its evidence attached. An excursion is not an event with a start line, and it has no global onset even in principle. The experiment that shows it: take a model of Earth’s field, turn the planet-wide bar magnet down to zero, and leave the lumpy non-dipole field in place, since the lumps do not vanish when the dipole does. A compass needle lines up with whatever field exists at the spot where you are standing, and with the bar magnet gone, what is left at your spot is the lumps.</p><p>In some regions the lumps happen to point close to where the bar magnet used to, and the needle there barely moves. Elsewhere it swings to a new heading entirely. Same planet, same moment, dipole at zero: one place is in the middle of an excursion and another has not noticed. For every compass on Earth to show a reversed direction, the model needs the dipole to actually flip and rebuild to at least 20% of its old strength the other way round. Everything short of that is regional, and some places go first. This is also why rocks are not perfect demonstrators of an excursion in terms of field direction: a rock records the direction at one spot, and that spot may be one of the places that had not noticed. But rocks do contain other solid information. No pun intended. They record how strong the field was, and strength is where the approach to an excursion can be read.</p><p>In the record, the approach to one of these has 4 marks:</p><p>1.) Field strength draining out of reversed-flux patches in the southern hemisphere<br>2.) The loss piling up in one weak region instead of spreading evenly.<br>3.) The dipole losing its grip on the structure underneath it.<br>4.) The dipole fading at the rates that came before the collapse: on the way into the Laschamp it fell at 7.7 to 14.2 nT a year in the early phase and 14.1 to 16.5 late, across 3 independent records, and that approach alone took nearly 3,000 years of falling field.</p><p>Now let’s compare modern measurements to historic data we mainly pulled from rocks. The modern fast decline began around 1840, out of reversed-flux patches in the southern hemisphere, after 250 steady years. The loss is concentrated in the South Atlantic Anomaly, which is deepening, spreading and splitting in two. The dipole is shedding 12% of its power per century while the non-dipole field gains 70%. And the modern rate of dipole loss has run between 10.5 and 24 nT a year, so our peak loss is higher than anything recorded on the way into the Laschamp. Four for four. And the field has been falling since 400 BC, 2,400 years of it, with the fast phase running since 1840. Set against the Laschamp’s 3,000-year approach, this one is running faster.</p><p>There is a fifth mark, “dipolarity”. Dipolarity is simply the share of the field that is the clean bar magnet. Today it is 0.65, meaning about two thirds of Earth’s field is the bar magnet and a third is the lumps. The field is made by a dynamo — liquid iron churning in the outer core, 2,900 km down, the same planetary engine that created the anomalies we discussed.</p><p>Geophysicists run simulated versions of that engine on computers and check which settings reproduce what Earth actually did over the last 100,000 years, the Laschamp included. The simulated cores that manage it run at a dipolarity of 0.4 to 0.5 on average, less than half bar magnet and more than half lumps, and the excursions are the dips below that. The highest those cores ever get is about 0.6 during excursions. Earth today is 0.65. So we are not at the excursion value, we are at the top of the swing, and that is the uncomfortable part.</p><p>Everything in this essay, the grid, the satellites, the 286 years of aurora records, was built near the top of that swing, in a field with more bar magnet than the planet’s core normally produces. From the top, the way is down, and the measured direction is down: the bar magnet is shedding 12% of its power per century while the lumps gain 70%. Today’s reassuringly strong dipole is what the peak of an excursion-producing dynamo looks like, and the people who ran those simulations write that Earth’s core sits “close in parameter space” to the switch between a one-magnet field and a many-poles field.</p><p>And yet, some geophysicists will tell you that today’s field strength is “ordinary”, and they are neither right nor wrong. Because it all depends which numbers you look at and how you interpret them, of course. Let’s consider only the whole-planet dipole numbers which I clarified is the wrong approach: about 110 at the Holocene peak, 81 as the average since the last ice age, 77 today, and about 35 as the level below which the palaeomagnetic record contains excursions. We are still below the average since the last ice age, above the average of the last 100,000 years, and roughly a third down from the ceiling. The same authors who measured that describe the field’s current rate of decay as similar to that of documented reversals and excursions. But some geophysicists will still claim our current field is “ordinary”, regular low-latitude auroras included. Because they’re using the dipole strength as the argument.</p><p>What we’ll get into now, is how the geomagnetic excursion isn’t about pretty lights and a moving compass, or satellite problems. It is about transformers, physiology, wheat and water.</p><h3>The Body</h3><p>The same geomagnetic storms we’ve discussed are measurable in our blood. During storms, stroke risk rises 19% — 52% in the under-65s during strong ones — and the risk climbs with storm strength. That comes from 11,453 stroke patients across 5 countries and 16 million person-years, in a design where every patient is his own control, so age, diet, income and postcode cannot explain the difference. And the storms doing it are the ones people go outside to photograph: Ap 100, the strong-storm bracket in that study, is a storm that puts “pretty lights” over Europe, and other low magnetic latitude areas.</p><p>On the first day of a magnetic storm, 71.5% of a group of heart-attack patients under daily microscope observation showed pathological changes in the blood flow through their capillaries.</p><p>Melatonin, the sleep hormone, is the best-replicated piece: 3 independent groups, 3 countries, measured in urine and in saliva, with suppression starting around 80 nT of disturbance in 3 hours — an ordinary moderate storm that happens many times a year — and an elevenfold difference in susceptibility between diabetics and everybody else.</p><p>The nights people came outside to photograph an aurora are nights measurably more of them had strokes.</p><p>Storms are one thing. A permanently weaker field is another, and that has been tested directly, on animals. Shield a mouse from Earth’s field for 8 weeks and its capacity to grow new neurons in the hippocampus falls and its learning degrades. Restore the field and it recovers. In rats, both an extreme storm and total shielding suppress melatonin — the same hormone as above.</p><p>In those experiments “hypomagnetic” means the field reduced to a few per cent of normal, and Earth is nowhere near that. But the point stands. The field is not just a shield from “solar weather”. It is one of the inputs a body is sensitive to — and the input is now moving in the unfavourable direction.</p><p>The one practical lever the literature keeps finding is cardiometabolic health — the healthier the body, the smaller the effect. That means it’s time to get lean and zen, cause this is only going to get worse.</p><p>A little bit of anecdotal evidence that comes from myself. There have been times in the last few years when I’ve gone off my rocker, so to speak. Extreme irritability without any clear reason. I checked the “GOES Proton flux” and sure enough, the massive spike in protons due to a CME impact coincided with my extreme irritability perfectly with consideration to time of onset. To be clear, those protons are stopped by the atmosphere. They aren’t what’s hitting you. The chart is a marker for the whole disturbed state of the field, and it’s the field that’s doing it.</p><p>The proton flux chart is an excellent source because protons only spike immensely during very strong solar storms. If you google “GOES proton flux” you’ll find it, and you’ll see a red dashed line on the graph. When the graph crosses the red dashed line, expect physical and mental impact. Since I’m very interested in this topic, I’ve also watched people around me go through the same extreme irritability and health issues with no idea what was causing the “acute event”. I’ve also seen massive immediate impact from very strong flares that do not show up on the proton flux, but radio flux chart instead. Rarer, but in my observation the real culprit behind health issues and stress levels, is the CME itself arriving 24 to 48 hours after it leaves the Sun.</p><p>Let’s stay in the anecdotal lane for a tiny bit longer. I’ve read recurrently online that electronic devices break down during these storms at a much higher rate. A glimpse of what is to come. If you have a friend, family member or acquaintance working at a large shop that sells electronic devices on a physical location, ask them a few days after a strong solar storm how many devices they had returned in relation to the average. The answer is your answer to how real this is.</p><p>Back to the skies, and the electrical grid. The latter being the thing that’s keeping our civilization alive.</p><h3>The Wire</h3><p>The light in the sky is one end of a chain whose other end is in the power grid. I’m not the one saying it, the British Geological Survey states it — in public. A coronal mass ejection arrives. The auroral electrojet strengthens, up in the ionosphere. The magnetic field at the ground changes fast. That induces an electric field in the rock, and the electric field drives current along every long conductor lying in it. The current enters the electrical grid through the earthing of transformer neutrals — the safety connection to the ground that every large transformer has by design.</p><p>The transformer’s core then saturates on half of each cycle. A saturated transformer draws enormous reactive power — power that surges back and forth doing no useful work while loading every line — and pumps harmonics, currents at the wrong frequencies, into the system. Protective equipment trips. Voltage collapses. And sustained current heats the transformer. Every link of that chain was measured in one real grid in May 2024, in New Zealand, with current recorded at more than 70 transformers.</p><p>The quarter-off-the-dipole experiment from the field section closes the same loop from the other end: same solar wind, 25% less dipole, stronger currents feeding the electrojet from above. The weakened field that produces a brighter aurora out of a given solar storm is producing a stronger electrojet out of it too. And the electrojet is where the aforementioned chain to the transformer starts.</p><p>A large power transformer is custom-built for one substation, largely imported, and not stockpiled. Over 90% of American electricity passes through a large power transformer at some point, and the average age of the transformers on the North American grid was already 38 to 40 years — against a design life of about 40.</p><p>Utilities do hold spares, but spares are sized for uncorrelated failure: one transformer, one lightning strike, one fault. A geomagnetic storm stresses every transformer on the grid in the same hours, which is exactly the case a spares inventory is not sized for. The Department of Energy’s own inventory notes the obvious catch: a spare sitting in the same substation was exposed to the same storm. Transformers at the end of their design life, spares built for a different kind of failure, and the storm of May 2024 has already been through all of them.</p><p>We know what replacing these transformers looks like, because it has been done exactly once under pressure. In 1992 Hurricane Andrew damaged 165 square miles of Florida. That was enough to empty the national transformer pool and produce a countrywide shortage for a year, with overseas suppliers taking 6 months to build new units under pressure. Today the same order takes 3 years in “peacetime”. One county’s worth of damage exhausted a continent’s spares. And it was repaired by a planet that was otherwise fine.</p><p>A more informed but still misinformed individual that follows solar weather impact might claim: “Storms are a problem for Canadians and Scandinavians, not for a country at Croatia’s latitude”. Actually, no.</p><p>Transformers have already been destroyed at Croatia’s magnetic latitude. After the Halloween storms of 2003, 12 South African transformers were damaged and removed from service. These transformers were at 40 degrees corrected geomagnetic latitude. Zagreb’s is 40.5, and the South African grid sites bracket it: Durban at 38.8, Cape Town at 41.9, Johannesburg at 35.8, with the damage in the middle. The same number, not an approximate parallel.</p><p>One of those transformers failed 3 weeks after the storm. This damage does not always become evident itself on the night, so transformers that came through a storm looking intact have not necessarily come through it. On the big end of the record: in March 1989 the Québec grid collapsed in 92 seconds, and a transformer at Salem in New Jersey was destroyed in the same hours and never returned to service.</p><p>And on the European system, during the same 10 to 12 May 2024 storm: a disturbance in a transformer in southern Sweden reduced the power flowing on the interconnector between Sweden and Poland, and the grid operator confirmed geomagnetically induced current as the cause. A European interconnector, in the storm that put red light over Croatia, with the operator identifying the cause. Transformers have already been destroyed at Croatia’s own magnetic latitude, by a storm that has already happened.</p><p>So what does the official number say about Croatia? A European project, EURISGIC, published it: induced currents in Central and Southern Europe likely stay below 50 amps. The same project’s final reporting admits that its ground models are built from 1-dimensional blocks, that resistive rock raises the induced electric field, and that fully 3-dimensional models would be needed at sharp conductivity boundaries.</p><p>The sharpest conductivity boundary there is, is a coastline. Seawater conducts, rock resists, and the contrast concentrates the induced field on the land side — where the grid is. When a 2026 study benchmarked Sweden’s grid with a model that can see the ground, resistive coastline came out 10 to 30 times worse than a conductive patch of the interior, and 10 volts per kilometer across a Swedish line under 20 km long drives 200 amps — 4 times the ceiling published for all of Central and Southern Europe.</p><p>Croatia has 6,278 km of deeply indented coastline, islands included, with more than 8,000 metres of resistive carbonate rock behind it — exactly the arrangement that concentrates the induced field. Whether Croatia’s currents are high, nobody can say: I could not find a single published study of induced currents in the Croatian transmission network, while most of the neighbourhood, from Czechia to Greece, has published numbers.</p><p>The Pannonian half of the country is the opposite case, to be clear — thick conductive sediment, which suppresses the induced field.</p><p>But the number that says Croatia is fine came out of a model whose own authors say it cannot describe a coastline, and Croatia is 6,278 kilometers of coastline. No measurement of it exists.</p><p>As for what one substation can do to a continent: on 8 January 2021 the synchronous power grid of continental Europe split in two for 63 minutes. The initiating event was a busbar coupler tripping on overcurrent protection at the Ernestinovo substation in eastern Croatia — 43 seconds from the first trip to a continent running at 2 frequencies, with 5.8 gigawatts of deficit on one side and 5.8 of surplus on the other.</p><p>That event had nothing to do with space weather. It was an overcurrent trip under heavy cross-border flows. What it shows is how the thing is wired: one substation near Osijek, and the whole continental grid comes apart in under a minute.</p><h3>The Food</h3><p>If we lose the grid, the capacity to feed 8 billion people goes to practically zero. So how long could Croatia feed itself?</p><p>The honest answer starts with keeping 3 things apart: what is in the shops, what households are told to keep, and the state strategic reserve. Nobody publishes how many days of food sit in a European shop — the 3-day figure everybody repeats traces back to an aphorism from 1896, and Alfred Henry Lewis, American journalist, stating it back in 1906, when things were very different. Yet apparently, the more things change, the more they stay the same?</p><p>The documented numbers: the European Commission tells every citizen to keep 72 hours of food and water at home. State reserves are another matter — Germany holds about 9.7 kg per citizen, Finland 9 months of grain, Switzerland 2 to 4 months.</p><p>And Croatia’s own accounting, given to Croatian parliament for 2024, lists 56,630 tonnes of wheat, 64 tonnes of rice, 225 tonnes of canned fish and 262 tonnes of canned meat — for 3.86 million people.</p><p>The system is designed by law to give basic supply to about 50,000 citizens for 15 to 60 days — 1.3% of the country. Croatian emergency planning assumes the disaster is somewhere else in Croatia and the rest of the country still works. This is the prevalent theme across the world.</p><p>The one reassuring number is that the reserves hold cereals for the whole population’s needs for 60 days. 56,630 tonnes across 3.86 million people is 14.7 kg each — about 245 grams of wheat a day, on the order of 800 calories, a bread ration and nothing else on the plate. I did that arithmetic myself, on the state’s own tonnage.</p><p>But that’s raw wheat. Between the silo and a loaf stand an industrial mill and a commercial oven, and both run on electricity. The national food reserve is stored in the ONE form you cannot eat without the grid.</p><p>Granted, someone hungry enough can turn raw wheat into food without electricity. But most people these days are so stupid it’s a wonder they can breathe, and the tonnage above is for these people. Assuming everyone can make their own bread from raw wheat, that means 2 months, and then the rations are out, and nothing will be coming in. Because the world is at a standstill.</p><p>So why hasn’t a blackout caused starvation in these modern times? The American commission that studied this built its finding on the complete record: 6 weather blackouts between 1992 and 2005. Its answer is called the edge effect — the capability of neighbouring localities and states to provide recovery assistance — and it states plainly that starvation-related deaths were zero because the damaged area was small enough to be aided rapidly by undamaged fringe areas. Then it states the counterfactual itself: had Hurricane Andrew damaged a larger area, undamaged fringe areas would have been less able to come to the rescue to the innermost affected areas, and significant loss of life was likely.</p><p>Every blackout we have ever survived had someone undamaged next door. A severe geomagnetic storm reaches an entire wired hemisphere in the same hours. This is the one event with no one next door. We all bite the blackout bullet at the same time.</p><p>The number you may have heard before — that without the grid, more than 90% of people starve — deserves its biography. It entered the record on 10 July 2008, when a congressman read the premise of a thriller novel into a House Armed Services Committee hearing and asked the chairman of the EMP Commission whether it was realistic, and the answer was that it is in the correct range. That is the entire provenance. No published derivation exists that I could find. Given the scenario, it’s not hard to imagine that range is correct. More than 90%.</p><h3>The Fire</h3><p>But it’s far more than merely “just” a starvation issue. A severe solar storm is the one natural event that pushes on every power line on a continent in the same hours. That fact is not in dispute, and neither is what those lines already do to the landscape. In the United States Forest Service’s wildfire occurrence database, from 1992 to 2024, power-line ignitions rise from about 200 a year to 2,500 — a 13-fold increase.</p><p>Texas alone went from 67 fires in the decade to 2001, to 4,474 in the decade to 2024. Wildfires attributed to power lines burn an average of 125 acres each, the largest of any human cause. The Camp Fire started on a transmission line and killed 85 people. The Black Saturday fires started on power lines and killed 173.</p><p>We spent 30 years expanding the grid, and as a result causing it to have 13 times as many ignition points as it had in 1992. A severe solar storm impacts every one of them in the same hours. And the same storm now gets a bigger response out of Earth’s field than it did when that count stood at 200 fires a year in 1992.</p><p>What did the May 2024 storm actually do to the transformers? The continental regulator wrote it down. NERC’s review of the event lists, among impacts reported across North America: induced-current and harmonic alarms triggering operating actions to protect transformers, induced current above the thresholds for normal operation at generator step-up transformers, and power transformer top-oil high temperature alarms forcing operator monitoring or system reconfiguration. The regions taking equipment impacts were the US Northeast, the mid-Atlantic and western Canada. The same document’s own summary goes with the list: the bulk power system remained “stable and largely unaffected”.</p><p>The measurement of what that current does inside a transformer comes from New Zealand: at the Halfway Bush substation in Dunedin, on 11 May 2024, direct current through two 220 kV transformers exceeded 50 amps repeatedly and peaked at 113, with the even harmonics in the output that a core saturating on half of every cycle produces. And on Croatia’s own magnetic parallel, a study of the Slovak transmission operator’s maintenance logs found higher equipment failure rates at geomagnetic maxima than at the minima either side, after removing the strong dependence on temperature and rainfall. Slovakia sits at about the same magnetic latitude as northern Croatia.</p><p>An alarm is not a fire and it is not damage. Yet. A set-off alarm in a transformer means induced current. It means the transformers got hot enough to set off their own alarms, across a continent, in one weekend, and the grid held. For now. That is what a near miss looks like. However, don’t forget, the fast phase of the excursion is in effect, and we’ve got years or even decades to go. You draw the rest of the conclusion.</p><p>The mechanism those alarms belong to is documented at every level. The Department of Energy: induced current drives half-cycle core saturation and transformer heating, and prolonged operation at high temperature can cause fires if hot insulating oil escapes the tank. The United Kingdom’s official severe-space-weather scenario lists insulation burning through transformer magnetic flux leakage. CIGRÉ’s engineering guide to transformer fires notes that induction heating from leakage flux is a cause of fires.</p><p>The base rate, instead of an adjective: about 0.1% of large power transformers catch fire in any service year, roughly 4% across a 40-year life, nearly always a total loss. Leakage flux from half-cycle saturation is exactly what an induced current produces. Nobody has joined those 3 documents in print — so I just did.</p><p>Storm-driven current has set buildings on fire before, when the wired world was a fraction of its current size. A 2025 study led by the United States Geological Survey counts 7 fires attributed to the May 1921 storm, calls the one at the Union Railroad station in Albany confidently attributed, and the telegraph and telephone station at Karlstad in Sweden burned the same night. In 1859, telegraph offices burned.</p><p>“Those were telegraphs,” somebody will say. The team that reconstructed the 1921 storm answers that themselves: communication systems have become less exposed to storm-driven ground currents and power systems more exposed, because the controlling variable is the length of grounded conductor, and telecommunications have shed theirs while power transmission acquired an enormous amount. Their word for 1921 is “harbinger”. The same storm arriving today finds vastly more grounded wire than it did in 1921, and that wire ends in transformers that take 3 to 5 years to replace.</p><p>So where are the modern cases? Written in a column nobody reads.</p><p>When something breaks on a grid, somebody fills in a form, and the form has a box for the cause. An engineer at the testing and standards centre of Greece’s Public Power Corporation published the failures in the Greek grid on geomagnetically stormy days between 1989 and 2010, with the day’s indices attached: Buchholz relays — the fault detectors inside a transformer’s oil — tripping, differential protection triggering, oil leaking. The recorded causes were “unknown”, “aging” and “manufacture defect”.</p><p>His 3 explanations are the point. A transformer might fail only months after the initial damage, and appear to fail for unknown causes. And: unless overheating caused by leakage flux established by induced currents is specifically considered — which is unlikely on current knowledge of the problem — induced currents will not be reported as a cause of faults. Greece’s Public Power Corporation engineer stated himself that his technicians were never trained to look for it.</p><p>The Greeks did not compute against a no solar storm days baseline, so their lists are counts, not a tested excess. Then, in July 2025, the test with a baseline was finally run: take every electric disturbance event reported to the US federal government from 2006 to 2023, and ask one question. Did it happen during a solar storm? Around a geomagnetic storm onset, the outages logged as unknown or unusual come in at roughly double the quiet-day rate — 1.92 times in the 3-day window, higher still around stronger storms.</p><p>Storms cluster near the equinoxes and so does grid stress, so the authors refitted with month controls and the effect survived at an odds ratio of 1.83. The unexplained failures are proportionally highest in the 2 American reliability regions at the highest geomagnetic latitudes — which is what the mechanism predicts, and not what a reporting artifact says. It is correlational, and it’s not mine. Also, just to be fair, the study never mentions fire.</p><p>So the last step is my own reading: the record of what breaks the grid is a record of what the person filling in the form was trained to look for, and the one column that doubles around a magnetic storm is the column marked unknown.</p><p>One pair of nights shows why a dated coincidence proves nothing. Fort Nelson, British Columbia, 10 May 2024: a wildfire drove 4,700 people out of town at 02:30 universal time, and the aurora was photographed standing above the burning forest. The planetary index in that half hour read 9.0 — the top of the scale, during the storm’s most extreme ground disturbances over North America — and the cause of the fire on the record is a tree blown down in high wind onto a power line.</p><p>The other night was 14 May 2026: geomagnetically dead, planetary index 3.0 at its highest, winds of 119 km/h over southern Manitoba. One fire department answered 4 separate power-line fires in about 90 minutes. The same ignition mechanism ran once under the top of the storm scale and once under nothing at all. Wind and a dry forest do this on their own. Fort Nelson is a coincidence.</p><p>No geomagnetic storm has been documented igniting a fire in a modern power grid, and nobody has studied storm-driven faults as a wildfire ignition source. The record would stay empty even if it happened, for the 2 reasons above: the cause box is filled in by people never trained to consider induced current, and the damage can surface months after the night. I make no claim about how often solar storms start either transformer fires or “regular” fires because there is no sufficient mainstream evidence, in either direction. Absence of evidence is not the evidence of absence. However, we do not need evidence of fires, what we have already is enough. Transformer damage recorded that is directly related to solar storms. Damage all at once, and we’re done.</p><p>What is measured is the other half of the picture: when the power goes, the water goes, because most water moves by electric pump. In the August 1996 southwestern blackout, only 2 of Fresno’s 16 fire stations had water and most of the hydrants were dry. In the 1998 ice storm, Montreal’s fire department prepared to fight fires with a demolition crane instead of water. Ordinary blackouts, ordinary scale, undamaged neighbours a few hours away.</p><p>The event that overloads every ignition point on the continent in the same hours is the same event that takes the water away from the people whose job is putting fires out.</p><h3>Agriculture</h3><p>There is a second-order effect to the starvation issue.</p><p>Between the Sun and a crop there are 2 components that decide how much of the Sun’s UV reaches the crop. The physics is published and mainstream: energetic particles striking the upper atmosphere make nitrogen and hydrogen oxides, those destroy ozone catalytically, and less ozone means more ultraviolet at the ground. Whether that happens, and where, is controlled by the magnetic field. The first component is the solar storm, and under today’s field it hasn’t had measurable impact: checked against 23 years of satellite data, ordinary solar proton storms produce no detectable ozone effect in the lower stratosphere.</p><p>The second component is Earth’s field itself. Under a field at excursion strength, the same physics gives 5 to 15% more ultraviolet at the surface, and the extra load falls on the low latitudes — the crop latitudes — rather than the poles, because that is where the UV shield collapses.</p><p>What does extra ultraviolet do to a crop? Fruit growers already lose 5 to 50% of a harvest in bad years to sunburn browning, and sunburn browning requires ultraviolet-B. Nobody has shown Earth’s field turning that dial today — the ultraviolet rise Europe has measured is attributed to cleaner air and fewer clouds. Supposedly. But the dial is real, it is wired to ultraviolet, and an excursion-strength field turns it up planet-wide. So to summarize, the two components causing the Sun to affect crops are the size of the storm and the strength of Earth’s field, and the field is the component that is changing.</p><p>The damage agriculture has already taken in this solar cycle from solar weather needed no ultraviolet at all. On 10 May 2024, about 4 hours of degraded satellite navigation stopped the GPS-steered planting tractors across the American corn belt during planting week. The modelled cost is 565 million dollars, inside a published range of 70 million to 1.7 billion, and the authors said the cause, in their own words: “the expanded auroral oval and the stable red arc above it”. The same red light people were photographing over Europe and other mid to low-latitudes stopped the tractors in Iowa.</p><p>The size of this misfortune event in plain units is as follows. A bushel of corn is about 25 kg, and an acre of American corn grows about 180 of them in a normal year. The 4-hour outage cost a modelled 141 million bushels — about 3.6 million tonnes of corn, 63 times Croatia’s entire strategic wheat reserve, gone from a single afternoon of tractors standing still. Spread over every one of America’s 90.6 million acres of corn, that comes to 1 to 2 bushels an acre, about 1% of the crop, and the authors call it negligible. That depends.</p><p>In the report on this event there is something written down that says otherwise. Since 2000 there have been 14 days with a storm in the top bracket of the scale, Kp 8.5 and above. None of them fell in solar cycle 24, which is exactly the decade American farming moved onto GPS. Cycle 25 has already produced 3. So GPS farming was built during the quiet cycle and has only now met a normal Sun, and a shield that reacts harder than it used to. And 3.6 million tonnes is what one of those 3 days cost on a day when nothing broke down permanently: no transformer burned, no grid failure, the only thing that happened was 4 hours without a satellite signal.</p><p>And there’s going to be a whole lot more of that.</p><h3>The Scoreboard</h3><p>Before I continue, one clarifier: flares are the source of Coronal Mass Ejections (CMEs) that cause the strongest solar storms, and not every flare produces a CME. Likewise, a CME can happen without a flare, but this is much rarer than as a result of flares. To be fully honest, if we get one of those rare massive filament breakoffs in the direction of Earth, we’re done. Even if the field was at full strength. Because we are dependent on electricity, otherwise we’d be fine. That’s why this discussion is mostly around CMEs as a result of flares. Because the filament breakoff in Earth direction is statistically very unlikely.</p><p>The catalogue of flares is important because it can give us a hint when the big one will hit us. Catalogue aside, judging just by the cycles, which is a different matter, the next increased activity is expected to start around January-March 2027 lasting for approximately 5–7 months. The effect should be significant as a result of constructive interference between the ~1.7-year quasi-biennial oscillation and the ~7-year modulation. The upward trajectory of the Gleissberg cycle is also supposedly about to start, and we should start seeing the effects of that in the early 2030s. With more flares, and stronger flares.</p><p>The catalogue of flares gives us a different insight, with the same forecast as the start of a new Gleissberg cycle. We are due. But it does so from a much weaker vantage point, as the flare catalogue has its flaws. Three of them, to be precise.</p><p>Firstly, the flare catalogue only records the half of the Sun facing us. We are missing half the data.</p><p>Secondly, the instruments clip. On 4 November 2003 a flare pinned the detectors on the GOES satellites — the X28 in the textbooks is not the real measurement, because the sensor got saturated to the point where the instrument stopped responding. Three independent methods put the real value between roughly X34 and X50.</p><p>Finally, the whole pre-2017 flare record was reported on a scale about 30% below the modern one. Recalibrate the catalogue and undo the saturation, and the ranking changes: a flare listed for 46 years as X15, from July 1978, comes out at X45.9 — the largest of the satellite era, mis-filed in plain sight for most of a working life. And even the 2003 monster gave Earth almost nothing compared to what it could have, because it erupted 83 degrees around the Sun’s western edge: Earth’s magnetometers that day bottomed at a smaller disturbance than the November 2023 night that put curtains over Croatia.</p><p>If the 2003 flare was a direct hit, it could have been lights out then and there. Bye bye electricity, hello anarchy.</p><p>Bottom line is: what reaches us is not necessarily what the Sun produces, and what we recorded is less than what the Sun produced.</p><p>As for the catalogue itself, it shows that flares arrive in bands, rather than on schedule, not honoring averages and “expected dates”. Regardless, let’s look at the expected intervals. Between X15 and X18: about every 13 years, last in 2005. Between X20 and X25: about every 25 years, last in 2001. Above X25: about once in 50 years, last in 2003. That single event is why the middle band’s last entry reads 2001. All averages over a clustered process, not appointments: 6 of the X10-and-above events in the whole catalogue came out of one active region on the Sun in 15 days of June 1991. Again, an interval is not a schedule. The science is not yet that precise. But every one of those bands is past its interval. And the shield is not what it used to be, which is the point of this entire essay.</p><p>If we set the forecasts aside entirely and look at only what was measured we get the following summary:</p><p>1.) A Croatian sky that required the deepest solar storm of a solar cycle in 2003 now requires a solar storm that arrives several times a year.<br>2.) The weakest region of Earth’s field grew by half a continental Europe in 11 years, deepened, and split in two.<br>3.) The weakest patch on the planet has covered three quarters of the distance to the line the rock record only shows during the lowest point of excursions, since 1980.<br>4.) The storms lighting the sky raise stroke risk 19%, in a study where every patient is his own control.<br>5.) Croatia’s food reserve is 60 days of raw wheat, and requires a mill and an oven that both need the grid.<br>6.) The grid-failure column that measures transformer failures — doubles around magnetic storms and is marked “without known cause”.</p><p>All of the above are measurements, already taken. Not forecasts.</p><p>An honorable mention to this topic is the phenomenon called “Earth-facing quiet” where Earth seems to be spared from the Sun’s heaviest hits. For now. We don’t know what is causing it, and this “protection” can drop at any moment. Certainly didn’t spare us from a near miss in 2012, although generally speaking, this phenomenon is still measurable.</p><h3>The Rebuild</h3><p>A large power transformer weighs 150 to 400 tons and is custom-built for one substation. Ordering one takes 36 months as the commonly quoted figure, up to 60 at the worst — before 2020 it was under a year. 82% of the ones in American substations are imported, and the highest grade of electrical steel their cores are wound from is available from a single manufacturer on Earth, Nippon Steel in Japan.</p><p>Making more of them is not faster. New factory construction takes 1 to 3 years, and the winding machines and core tables to equip a factory take 1 to 2 years to obtain from suppliers. Then the finished transformer crosses an ocean and travels the last 5 to 10 miles by road, which the Department of Energy calls the expensive and difficult part “even in normal times”.</p><p>Now take the lights out scenario across a hemisphere. A factory that takes 3 years to build cannot be built without power. A transformer cannot be wound without power. Steel of that grade is not made without power. Ships and heavy haulage burn refined fuel, and fuel is refined and pumped with electricity. The crews have to be found, coordinated and dispatched, and that takes communications. Which need electricity.</p><p>I assembled that chain, but the links are not guesses. The commission that studied this wrote the general form down in 2008: electrical power is necessary to support other critical infrastructures, including supply and distribution of water, food, fuel, communications, transport, financial transactions, emergency services and government services.</p><p>So the honest take at the prospect of a hemispheric blackout is not a long repair. It is a loop that does not close. There is no rebuild clock, because every input to the rebuild itself requires an output that no longer works. The problem is the same commission concluded in the same paragraph that the recovery plans currently in place essentially assume, at worst, limited upsets to the other infrastructures. Every recovery estimate in this subject is computed inside a world that still works.</p><p>And while the repair cannot begin, people are on their own clock. Food, water. Croatia’s side of it you already know: 60 days of wheat that needs a mill and an oven. Water is worse. Most of it moves by electric pump, the largest treatment plants draw about 100 megawatts — which makes backup generators impractical — the staffing for manual operation no longer exists, and without water you die in 3 to 4 days.</p><p>The commission’s own capability statement: federal, state and local governments do not have the collective capability, if the water infrastructure fails over a large area, to supply enough water to the civilian population to preserve life.</p><p>The clock on one side is days. On the other side, the repair cannot begin. There is no casualty percentage to put on that — nobody has derived a precise one. The 90%+ is a guess. But I personally don’t need a calculation from some verified institution. Set 3 days without water against a transformer that needs a working civilisation to build, ship and install it, and the arithmetic is not something I should have to do for you.</p><h3>The Warning</h3><p>So nobody waits around for the wrong warning, let’s clear up a few things.</p><p>The most severe geomagnetic storm since 1989 arrived in a solar cycle that has not produced a single flare of X10 class. Flare class and storm severity are two different instruments measuring two different things — the flare is the flash, the coronal mass ejection is the weapon that burns our grid down. Furthermore, how the CME couples with the Earth’s field is an entire matter on its own… A weak CME with a very “unfortunate” structure in relation to the Earth’s field can be devastating, while the opposite is also true, a strong CME can have the magnetic structure that our planet’s field manages to contain properly without catastrophic consequences.</p><p>Two storms from the last 3 years show what that means in practice, and what it means for us. On 23 March 2023 a severe storm, G4, the second-highest level on the scale, hit with no flare behind it. The eruption was so faint that the catalogues of solar eruptions did not list it, and the paper that finally traced it, in Nature Communications, calls it the stealthiest eruption ever recorded to cause a severe storm. Nothing that faint had ever done that before. NASA’s own model, run on what its forecasters could see, called for Kp 3 to 5. It came in at 8.</p><p>The cloud carried a magnetic field of 20 nT, 4 times the everyday value, pointing south for more than 10 hours, at a speed barely above the everyday 400 km/s, and that is all it takes: Earth’s field points north at the front of the shield, and a solar wind whose field points south unzips it for as long as it stays south. Two years later, on 30 September 2025 there was no eruption at all. NOAA’s forecast that evening gave 5% odds to anything above a minor storm. The wind’s own field turned south, 16 nT for 3 hours, and the planet got a G3, Kp 7, with the aurora appearing over New Mexico.</p><p>So what does that mean? Three equinox storms in 3 years, March 2023, March 2024, September 2025, and every time the storm that arrived was bigger than the warning said, every time in the same direction, and twice it came from something nobody saw. The forecast is the only defence a grid has. A transformer cannot be moved out of the way but the operator can still lower the load if he knows, hours ahead, what is coming. The warning has now failed three times in 3 years.</p><p>As for how likely the big one is — a storm the size of the 1859 Carrington event, the biggest solar storm on record, the one that takes the grid down — science has two answers, and they are at odds. The most-cited one says about 12% per decade, roughly 1 chance in 8 that it happens in any given 10 years. It was built by plotting how often storms of each size arrive, small ones often, big ones rarely, drawing a line through the points and extending that line out to Carrington size.</p><p>A 2019 study says 0.46 to 1.88% per decade, somewhere between 1 chance in 50 and 1 in 200. It was built the other way round, from the gaps between storms, with a different mathematical curve for how those gaps stretch as the storms get bigger. How can two serious estimates land that far apart? Because the storm they are both estimating has happened about once in modern records. Happened plenty in historical times we just didn’t have the tools to record them properly, but now we can and it matters much more because of our electrical grid.</p><p>The point of it is — you cannot measure the odds of a thing that has happened once. You can only draw a line through the smaller storms and extend it, and past the last data point the answer is whatever form you assumed the line has. Same data, different guess about the end of the curve. So nobody knows whether it is unlikely, or how likely. Regardless, both are estimates of how often the Sun fires a shot that size, computed on the assumption that the shield it hits is the shield of 1859. It is not. That is what this entire essay has been about.</p><p>And in July 2012, during the weakest solar cycle in a century, the Sun fired a Carrington-class ejection — and missed. For the unacquainted. Carrington class means no electricity, for anyone. The ejection passed the Earth so close we could wave at it as it was passing by. Approximately 10 days of orbit “saved us” from getting a direct hit. Ironically, the world as we knew it almost ended in 2012, the famous cataclysmic year, and most people don’t even know. So yes, the world can “end” basically “tomorrow”, as in, any day, we have no clue.</p><p>However, there’s a “comforting” reality, when a Carrington-class CME is ejected from the Sun directly at Earth, all the amateur space weatherists will know, and the government as well. Because of social media there will be mass hysteria and all the governments will be suppressing it via news and censorship to allow officials and the “elite” to move to their bunkers while the system still operates. So when that happens, you will have about 24–48 hours to prepare your survival in a world that still works, which can help immensely, versus the situation where the lights go out unexpectedly. It’s 24–48 hours because that’s the average time it takes a CME to reach Earth.</p><p>Carrington class CMEs are usually faster than the average though, and may take 12–18 hours. So there’s that… but it’s not a firm rule. Still, 12 hours beats 0 hours.</p><p>You have arrived at the end of this exceedingly long essay. Congratulations. The core lesson you should pull from the entirety of this essay:</p><p>“The same solar storm that made aurora visible to cameras with special settings at low magnetic latitudes — or sometimes not visible to cameras at all — can now be seen with the naked eye globally. This is a clear demonstration of how much the field has changed, and will continue to change, as per all the other research findings provided in the essay. The implications are dire, and action is urgent.”</p><h3>Research kit</h3><p>Sources for the essay, grouped by the section where the essay first uses them. Each source is listed once.</p><p>THE OPENING AND THE SUMMARY</p><p>1. Posavec, M. (2023). Polarna svjetlost iz Hrvatske 2023. nebo.com.hr, 7 November 2023.<br>2. nebo.com.hr, Croatian astronomy community. Polarna svjetlost, the community’s running record of auroras observed from Croatia, with naked-eye and photographic determinations logged separately.<br>3. Telegram.hr (2023). Večeras se iz Hrvatske mogla vidjeti aurora borealis. Fotografije su veličanstvene, 5 November 2023. telegram.hr/zivot/veceras-se-iz-hrvatske-mogla-vidjeti-aurora-borealis-fotografije-su-velicanstvene/<br>4. Lisac, I. and Marki, A. (1998). Catalogue of auroras observed from Croatia and neighbouring regions between 1737 and 1991. Geofizika 15, University of Zagreb.<br>5. GFZ German Research Centre for Geosciences, Potsdam. Kp planetary geomagnetic index, official data service.<br>6. United States Department of Energy (2024). Large Power Transformer Resilience, Report to Congress, July 2024, EXEC-2022–001242. energy.gov</p><p>THE CATALOGUE</p><p>7. NOAA Space Weather Prediction Center. Geomagnetic storm scale (Kp) and aurora viewline guidance. swpc.noaa.gov<br>8. spaceweather.com (2003). November 2003 aurora photo gallery, including the Zagreb observation of 20 November 2003.<br>9. International Geomagnetic Reference Field, 14th generation (IGRF-14), evaluated through the British Geological Survey’s public geomagnetic field calculator. geomag.bgs.ac.uk<br>10. Equatorward expansion of the auroral oval during the May 2024 geomagnetic storm observed over Europe by FRIPON all-sky cameras and satellites (2026). Journal of Space Weather and Space Climate 16:15, article swsc250042.</p><p>THE CHANGE</p><p>11. World Data Center for Geomagnetism, Kyoto. Dst and AE geomagnetic index archives. wdc.kugi.kyoto-u.ac.jp<br>12. Hayakawa, H. and colleagues (2025). Reference reconstruction of the May 2024 superstorm and its low-latitude aurora. Astrophysical Journal 979:49. DOI 10.3847/1538–4357/ad9335<br>13. Lockwood, Owens, Brown and Vazquez (2025). Survey of 223,916 auroral records from 1650 onward, corrected for the moving geomagnetic field. Monthly Notices of the Royal Astronomical Society 540(4), 3596–3624. DOI 10.1093/mnras/staf827<br>14. NOAA Space Weather Prediction Center (2024). Geomagnetic storm watches issued 23 March 2024 for 24 and 25 March 2024.<br>15. Davies, E. E., Weiler, E., Möstl, C., Majumdar, S., Rüdisser, H. T., Horbury, T. S., O’Brien, H., Morris, J. and Crabtree, A. (2026). Real-Time Prediction of Two Geomagnetic Storms Using Solar Orbiter as a Far Upstream Solar Wind Monitor. Space Weather 24(8), e2026SW005013. DOI 10.1029/2026SW005013<br>16. NASA Science (2025). Auroras. National Aeronautics and Space Administration, page updated 4 February 2025. science.nasa.gov/sun/auroras/<br>17. Whiter, D. K., Partamies, N., Gustavsson, B. and Kauristie, K. (2023). The altitude of green OI 557.7 nm and blue N2+ 427.8 nm aurora. Annales Geophysicae 41, 1–12. DOI 10.5194/angeo-41–1–2023<br>18. Godbole, N. and colleagues (2022). Observed altitude range of red-line aurora. Frontiers in Physics 10, 997229. DOI 10.3389/fphy.2022.997229<br>19. Gillies, D. M. and colleagues (2017). Characteristic emission height of the 630 nm red auroral arc. Journal of Geophysical Research Space Physics 122(8), 8181–8197. DOI 10.1002/2016JA023758</p><p>THE FIELD</p><p>20. Zossi, B. and colleagues (2019). Scaling of the polar cap boundary with the dipole moment. Geophysical Journal International 216(2), 1334–1343.<br>21. Ebihara, Y. and Tanaka, T. (2021). How Do Auroral Substorms Depend on Earth’s Dipole Magnetic Moment? Journal of Geophysical Research Space Physics 126(1), e2020JA028009. DOI 10.1029/2020JA028009<br>22. Hayakawa, H., Schlegel, K., Besser, B. P. and Ebihara, Y. (2021). Candidate Auroral Observations Indicating a Major Solar-Terrestrial Storm in 1680, Implication for Space Weather Events during the Maunder Minimum. Astrophysical Journal 909(1), 29. DOI 10.3847/1538–4357/abb3c2<br>23. Zossi, B. S., Fagre, M., Amit, H. and Elías, A. G. (2020). Geomagnetic Field Model Indicates Shrinking Northern Auroral Oval. Journal of Geophysical Research Space Physics 125(8), e2019JA027434. DOI 10.1029/2019JA027434<br>24. Zossi, B. S., Amit, H., Fagre, M. and Elías, A. G. (2021). Observed Auroral Ovals Secular Variation Inferred from Auroral Boundary Data. Geosciences 11(8), 351. DOI 10.3390/geosciences11080351<br>25. Cnossen, I., Richmond, A. D., Wiltberger, M., Wang, W. and Schmitt, P. (2011). Coupled magnetosphere, ionosphere and thermosphere simulation with the dipole moment reduced by a quarter. Journal of Geophysical Research 116, A12304. DOI 10.1029/2011JA017063<br>26. Pavón-Carrasco, F. J. and De Santis, A. (2016). Dipole moment, Holocene field levels, and the present rate of decay compared with documented reversals and excursions. Frontiers in Earth Science 4:40.<br>27. Finlay, C. C., Aubert, J. and Gillet, N. (2016). Decay of the Earth’s magnetic dipole since 1840 and its rate over time. Nature Communications 7:10422.<br>28. Gubbins, D., Jones, A. L. and Finlay, C. C. (2006). Fall in Earth’s magnetic field is erratic. Science 312.<br>29. Constable, C. G. and Davies, C. J. (2024). Rapid Changes in Strength and Direction of Earth’s Magnetic Field Over the Past 100,000 Years. Geochemistry, Geophysics, Geosystems 25(3), e2023GC011236. DOI 10.1029/2023GC011236<br>30. Laj, C. and Kissel, C. (2015). An impending geomagnetic transition? Hints from the past. Frontiers in Earth Science 3, 61. DOI 10.3389/feart.2015.00061<br>31. Mason, S. J., Davies, C. J., Clarke, A. T. and Constable, C. G. (2024). Insights into the last 100 ky of geomagnetic field variability using numerical dynamo simulations. Earth and Planetary Science Letters 646, 119011. DOI 10.1016/j.epsl.2024.119011<br>32. Brown, M. and Korte, M. (2016). Physics of the Earth and Planetary Interiors 254, 1–11. DOI 10.1016/j.pepi.2016.03.003<br>33. Korte, M., Brown, M., Panovska, S. and Wardinski, I. (2019). Frontiers in Earth Science 7:86. DOI 10.3389/feart.2019.00086<br>34. Mukhopadhyay, A. and colleagues (2025). Reconstruction of the magnetosphere during the Laschamp excursion. Science Advances 11, eadq7275.<br>35. NOAA National Centers for Environmental Information. Geomagnetic pole location files, annual, 1590 to 2025. ngdc.noaa.gov/geomag/data/poles/SP.xy<br>36. NOAA National Centers for Environmental Information and British Geological Survey (2025). World Magnetic Model 2025, including the current assessment of north magnetic pole motion.</p><p>THE SOUTH ATLANTIC ANOMALY</p><p>37. European Space Agency. Swarm mission results on the South Atlantic Anomaly, magnetometer data from 2013 onward.<br>38. Finlay, C. C., Kloss, C. and Gillet, N. (2025). Physics of the Earth and Planetary Interiors. DOI 10.1016/j.pepi.2025.107447<br>39. NOAA National Centers for Environmental Information. South Atlantic Anomaly area and intensity monitoring, on NOAA’s own weak-field definition.<br>40. Finlay, C. C. and colleagues (2020). The CHAOS-7 geomagnetic field model and observed changes in the South Atlantic Anomaly. Earth, Planets and Space 72:156. DOI 10.1186/s40623–020–01252–9<br>41. Yang, Y., Zeren, Z. and colleagues (2026). CGGM-2, a geomagnetic field model built solely from Chinese CSES satellite data. Earth, Planets and Space 78:33. DOI 10.1186/s40623–025–02361-z<br>42. MSCM geomagnetic field model (2025). Earth and Planetary Physics. DOI 10.26464/epp2025030<br>43. Chu, Y. and colleagues (2022). Cosmic ray cutoff rigidity computed from IGRF-13 across 1965 to 2025. Frontiers in Earth Science 10:818788.</p><p>THE BODY</p><p>44. Feigin, V. L. and colleagues (2014). Geomagnetic storms and stroke risk, a time-stratified case-crossover analysis of 11,453 patients in five countries. Stroke 45(6), 1639–1645.<br>45. Gurfinkel, Yu. I., Lyubimov, V. V., Oraevskii, V. N., Parfenova, L. M. and Yuriev, A. S. (1995). The effect of geomagnetic disturbances on capillary blood flow in ischemic heart disease patients. Biofizika 40(4), 793–799. PMID 7495904<br>46. Burch, J. B., Reif, J. S. and Yost, M. G. (1999). Geomagnetic activity and overnight melatonin metabolite in electric utility workers. Neuroscience Letters 266(3), 209–212. PMID 10465710<br>47. Burch, J. B., Reif, J. S. and Yost, M. G. (2008). Neuroscience Letters 438(1), 76–79. PMID 18472329<br>48. Weydahl, A. and colleagues. Salivary melatonin and geomagnetic activity measured at Alta, Norway. PMID 11774869<br>49. Zilli Vieira, C. L., Koutrakis, P., Liu, M., Gottlieb, D. J. and Garshick, E. (2023). Intense solar activity reduces urinary 6-sulfatoxymelatonin in patients with COPD. Respiratory Research 24(1):91. DOI 10.1186/s12931–023–02390-w<br>50. Zhang, B., Wang, L., Zhan, A., Wang, M., Tian, L., Guo, W. and Pan, Y. (2021). Long-term exposure to a hypomagnetic field attenuates adult hippocampal neurogenesis and cognition. Nature Communications 12, 1174. DOI 10.1038/s41467–021–21468-x<br>51. Wu, H., Yang, Y., Chang, W., Chen, X., Yang, S., Xu, M., Liu, K., Yun, Y. and Dong, L. (2025). Research on the effects and related mechanisms of geomagnetic storm on depression. Brain Research Bulletin 226, 111369. DOI 10.1016/j.brainresbull.2025.111369<br>52. NOAA Space Weather Prediction Center. NOAA space weather scales, solar radiation storm (S) scale, and the GOES integral proton flux channels. swpc.noaa.gov</p><p>THE WIRE</p><p>53. British Geological Survey. Geomagnetically induced currents and their effect on power grids, the survey’s public explanation of the chain from coronal mass ejection to transformer. bgs.ac.uk<br>54. Clilverd, M. A. and colleagues (2025). Geomagnetically induced currents measured at more than 70 transformers in the New Zealand transmission network during the May 2024 storm. Space Weather.<br>55. Lloyd’s (2013). Solar Storm Risk to the North American Electric Grid.<br>56. Wallner, A. V. L., Dimmock, A. P., Lanabere, V., Johlander, A., Molenkamp Venholen, S., Viljanen, A., Juusola, L., Rosenqvist, L., Setréus, J., Yordanova, E., Buchert, S., Heiter, U. and Khotyaintsev, Yu. V. (2026). The geomagnetic storm on 10–12 May 2024 and its effect on the Swedish power grid. Space Weather 24(2), e2025SW004788. DOI 10.1029/2025SW004788<br>57. European Risk from Geomagnetically Induced Currents (EURISGIC) project. Published estimates and final project reporting for induced currents in the European transmission system.<br>58. Lanabere, V. and colleagues (2026). Benchmarking of the Swedish transmission grid against a 1-in-100-year geoelectric field, co-authored with Svenska Kraftnät. Annales Geophysicae.<br>59. Favro, S. (2015). Traditional Sailing Ships as Indigenous Dalmatian Nautical Product. Turizam 19(4), 155–164. scindeks-clanci.ceon.rs/data/pdf/1450–6661/2015/1450–66611504155F.pdf<br>60. Vlahović, I., Tišljar, J., Velić, I. and Matičec, D. (2005). Evolution of the Adriatic Carbonate Platform, Palaeogeography, main events and depositional dynamics. Palaeogeography, Palaeoclimatology, Palaeoecology 220(3–4), 333–360. DOI 10.1016/j.palaeo.2005.01.011<br>61. ENTSO-E (2021). Continental Europe Synchronous Area Separation on 08 January 2021, ICS Investigation Expert Panel Final Report. European Network of Transmission System Operators for Electricity, 15 July 2021, version 2.0 published October 2021. entsoe.eu/news/2021/07/15/final-report-on-the-separation-of-the-continental-europe-power-system-on-8-january-2021/<br>62. ENTSO-E (2021). System Separation in the Continental Europe Synchronous Area on 8 January 2021, Interim Report. European Network of Transmission System Operators for Electricity, 26 February 2021. entsoe.eu/news/2021/02/26/system-separation-in-the-continental-europe-synchronous-area-on-8-january-2021-interim-report/</p><p>THE FOOD</p><p>63. Government of the Republic of Croatia (2024). Accounting of state commodity reserves for 2024, submitted to the Croatian Parliament, holdings as at 31 December 2024.<br>64. N1 Croatia (2026). Statement of State Secretary Vedran Špehar on the cereal component of the state reserve, 29 March 2026.<br>65. European Commission and High Representative of the Union for Foreign Affairs and Security Policy (2025). Joint Communication on the European Preparedness Union Strategy. JOIN(2025) 130 final, Brussels, 26 March 2025. eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:52025JC0130<br>66. Euronews (2025). Brussels asks EU citizens to put together a 72-hour emergency kit to face crises, 26 March 2025. euronews.com/my-europe/2025/03/26/brussels-ask-eu-citizens-to-put-together-a-72-hour-emergency-kit-to-face-crises<br>67. Stelzer, M., Agra Europe (2025). Versorgung in der Krise, so lange reichen Deutschlands Notfallvorräte. top agrar, 23 August 2025. topagrar.com/management-und-politik/news/versorgung-in-der-krise-so-lange-reichen-deutschland-notfallvorrate-a-20017263.html<br>68. Huoltovarmuuskeskus, National Emergency Supply Agency of Finland (2022). National Emergency Supply Agency boosting Finland’s emergency grain stockpiles. Press release, 15 December 2022. huoltovarmuuskeskus.fi/en/a/national-emergency-supply-agency-boosting-finlands-emergency-grain-stockpiles<br>69. Helsinki Times (2025). The Guardian, Finland strengthens food security, urges Europe to follow, 17 September 2025. helsinkitimes.fi/finland/finland-news/domestic/27936-the-guardian-finland-strengthens-food-security-urges-europe-to-follow.html<br>70. Federal Office for National Economic Supply (FONES), Switzerland. Range of compulsory stocks. bwl.admin.ch/en/range-of-compulsory-stocks<br>71. EMP Commission (2008). Report of the Commission to Assess the Threat to the United States from Electromagnetic Pulse (EMP) Attack, Critical National Infrastructures.<br>72. United States House Armed Services Committee (2008). Hearing transcript, 10 July 2008. United States Government Publishing Office, govinfo.gov</p><p>THE FIRE</p><p>73. Pohl, K., with Preston, T. and Kommers, N. (2026). Lessons from 30 years of wildfire ignition data. Headwaters Economics, published 28 July 2026, updated 4 August 2026. headwaterseconomics.org/natural-hazards/wildfire/lessons-from-30-years-of-wildfire-ignition-data/<br>74. Short, K. C. (2026). Spatial wildfire occurrence data for the United States, 1992–2024, FPA_FOD_20260615, 7th edition. Fort Collins, Colorado, Forest Service Research Data Archive. DOI 10.2737/RDS-2013–0009.7<br>75. Butte County District Attorney (2020). The Camp Fire Public Report, A Summary of the Camp Fire Investigation. Office of the District Attorney, Butte County, California, 16 June 2020. buttecounty.net/DocumentCenter/View/1881/Camp-Fire-Public-Report — -Summary-of-the-Camp-Fire-Investigation-PDF<br>76. Maranghides, A., Link, E., Brown, C., Mell, W., Hawks, S., Wilson, M., Brewer, W., Vihnanek, R. and Walton, W. (2021). A Case Study of the Camp Fire, Fire Progression Timeline. NIST Technical Note 2135, National Institute of Standards and Technology. DOI 10.6028/NIST.TN.2135<br>77. Australian Institute for Disaster Resilience. Bushfire, Black Saturday, Victoria, 2009. Australian Disaster Resilience Knowledge Hub. knowledge.aidr.org.au/resources/bushfire-black-saturday-victoria-2009/<br>78. State of Victoria (2011). Implementing the Government’s Response to the 2009 Victorian Bushfires Royal Commission, May 2011. cfa.vic.gov.au/ArticleDocuments/533/royal_commission_implementation_plan.pdf<br>79. North American Electric Reliability Corporation (2024). Event review of the 10 to 11 May 2024 geomagnetic disturbance and the impacts reported across North America.<br>80. Transpower New Zealand (2024). Gannon geomagnetic storm, event response summary and lessons learnt, September 2024.<br>81. Výbošťoková, T. and Švanda, M. (2022). Correlation of anomaly rates in the Slovak electric transmission grid with geomagnetic activity. Advances in Space Research 70(11), 3769–3780. DOI 10.1016/j.asr.2022.08.027, open preprint at arxiv.org/abs/2208.03813<br>82. Hapgood, M. and 24 co-authors (2021). Development of Space Weather Reasonable Worst-Case Scenarios for the UK National Risk Assessment. Space Weather 19(4), e2020SW002593. Open access copy at Lancaster EPrints 151500.<br>83. CIGRÉ Working Group A2.33 (2013). Guide for Transformer Fire Safety Practices. Technical Brochure 537.<br>84. Love, J. J., Lucas, G. M., Kelbert, A., Schnepf, N. R., Bedrosian, P. A., McBride, M. and Oughton, E. J. (2025). Impacts of the May 1921 geomagnetic superstorm, including the fires attributed to it. Space Weather 23, e2025SW004563.<br>85. Zois, I. P., Testing Research and Standards Centre, Public Power Corporation of Greece (2013). Solar activity and transformer failures in the Greek high voltage transmission grid. Journal of Space Weather and Space Climate 3, A32.<br>86. González Figueroa, M., Herrera Acevedo, D. D. and Sierra Porta, D. (2025). Geomagnetic disturbances and grid vulnerability, correlating storm intensity with power system failures. PLOS ONE 20(7), e0327716. DOI 10.1371/journal.pone.0327716<br>87. CBC News (2024). Thousands ordered to evacuate Fort Nelson, B.C., due to wildfire, 11 May 2024, republished by The Weather Network. theweathernetwork.com/en/news/weather/severe/thousands-ordered-to-evacuate-fort-nelson-bc-due-to-wildfire<br>88. The Canadian Press (2024). Fort Nelson wildfire evacuation, 24 May 2024, via Global News. globalnews.ca/news/10524005/fort-nelson-wildfire-evacuation/amp<br>89. CBC News (2026). Fierce winds, dust storm cause chaos in southern Manitoba, 15 May 2026, read through the Yahoo News Canada republication. ca.news.yahoo.com/fierce-winds-leave-damage-power-140827737.html<br>90. Bernhardt, D., CBC News (2026). Fires sparked around Manitoba by high winds, dry grass and knocked-down power lines, 15 May 2026. cbc.ca/news/canada/manitoba/fires-power-lines-wind-9.7201101</p><p>AGRICULTURE</p><p>91. Jia, J., Kero, A., Kalakoski, N., Szeląg, M. E. and Verronen, P. T. (2020). Superposed epoch analysis of solar proton event effects on Arctic ozone. Atmospheric Chemistry and Physics 20, 14969–14982.<br>92. Cooper, A. and colleagues (2021). The geomagnetic field during the Laschamps excursion and its effect on atmospheric ionization and ultraviolet radiation. Science 371(6531), 811–818.<br>93. Arsenović, P. and colleagues (2024). Ozone and ultraviolet response to an extreme solar particle event under two geomagnetic field scenarios. Proceedings of the National Academy of Sciences 121(28), e2321770121.<br>94. Gambetta, J. M. and colleagues (2021). Review of sunburn damage in fruit crops. Frontiers in Plant Science.<br>95. Washington State University Tree Fruit Extension. Apple sunburn, types, causes and management.<br>96. German Federal Office for Radiation Protection (BfS) (2024). Press release of 28 November 2024 on increasing solar ultraviolet radiation measured in Germany and Belgium.<br>97. Increasing solar UV radiation in Dortmund, Germany, data and trend analyses and comparison to Uccle, Belgium (2024). Photochemical and Photobiological Sciences. DOI 10.1007/s43630–024–00658–8<br>98. Griffin, T. W., Knipp, D., Shank, Skov, McIntosh and Leamon (2025). Agricultural impact of the 10 to 11 May 2024 satellite navigation outage on Midwestern corn production. DOI 10.5281/zenodo.14976490</p><p>THE SCOREBOARD</p><p>99. Australian Bureau of Meteorology, Space Weather Services. Large Solar Flares Since 1976.<br>100. spaceweather.com. The Most Powerful Solar Flares Ever Recorded.<br>101. Thomson, N. R., Rodger, C. J. and Dowden, R. L. (2004). Ionospheric measurement of the 4 November 2003 flare from very low frequency phase changes. Geophysical Research Letters 31, L06803. DOI 10.1029/2003GL019345<br>102. Brodrick, D., Tingay, S. and Wieringa, M. (2005). Riometer measurement of the 4 November 2003 flare. Journal of Geophysical Research Space Physics 110, A09S36. DOI 10.1029/2004JA010960<br>103. Tan, B., Zhang, Y., Huang, Y. and Ji, H. (2025). Recalibration of the pre-2017 GOES soft X-ray flare series and the resulting flare ranking. Astrophysical Journal Letters. DOI 10.3847/2041–8213/ada611, preprint arXiv:2501.03791<br>104. Stiefel, M. Z., Kuhar, M., Limousin, O., Dickson, E. C. M., Volpara, A., Hurford, G. J. and Krucker, S. (2025). Using the STIX background detector as a proxy for GOES. Astronomy and Astrophysics, preprint arXiv:2501.03667<br>105. Liu, Y. D. and colleagues (2014). Observations of an extreme storm in interplanetary space caused by successive coronal mass ejections, the July 2012 event. Nature Communications 5:3481. DOI 10.1038/ncomms4481</p><p>THE WARNING</p><p>106. Serra and colleagues (2019). Estimate of the probability of a Carrington-class storm computed from the intervals between storms. Scientific Reports 9:2393.<br>107. Riley, P. (2012). Estimate of the probability of an extreme space weather event computed from the distribution of storm magnitudes.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c8ab39865a62" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[You Are Actually Getting Significantly Dumber by Using AI to Write or Code]]></title>
            <link>https://medium.com/@filip.milner/you-are-actually-getting-significantly-dumber-by-using-ai-to-write-or-code-89ba21121aa7?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/89ba21121aa7</guid>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[coding]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Mon, 20 Apr 2026 04:28:15 GMT</pubDate>
            <atom:updated>2026-04-20T04:32:04.405Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NNtvjmYESp9zwQyDpsOjew.png" /></figure><p>We have arrived, welcome to idiocracy. Let’s learn what vibecoding and AI writing do to your brain, whilst you still can comprehend a higher-level article. We better hurry up, the clock’s ticking, and your brain’s rotting.</p><p>Before we get into hows and whys, let’s establish a simple truth on which almost everyone will agree right off the bat.</p><p><strong>You are what you do. No exceptions.</strong></p><p>In 2019 a <a href="https://www.nature.com/articles/s41562-019-0592-8">Stanford study</a> established that every participant who played Pokémon heavily as a kid and memorized the original 150 Pokémon developed a dedicated “center” that lights up specifically for Pokémon characters. Non-players (the control group) showed zero activation there. Quite peculiar if you think about it.</p><p>The “Pokémon center” developed in the same pea sized space within the occipitotemporal sulcus, which is ordinarily utilized for visual processing, including word recognition and high-level object perception. No surprise there — it sits exactly where you’d expect.</p><p>Now here’s the important distinction: it’s not that the non-players had no neurons in that section of their brain, it’s that their neurons didn’t light up when seeing pokémon images, whilst the players’ pea section of the occipitotemporal sulcus did.</p><p>The Pokémon study didn’t investigate higher neuronal connectivity, which was very likely present but was not tested. It only tested for region-specific activation upon “Pokémon stimuli.”</p><p>However, just the activation itself in the Pokémon study proves neuronal fine-tuning, which is still a type of an improvement in the “brain architecture” of the players. This is of course very relevant to the vibecoding topic. You can probably imagine how.</p><p>There are, however, plenty of other robust studies that prove higher neuronal connectivity as a result of repeated practice. So let’s get acquainted with those.</p><h3><strong>Positive Neuroplasticity</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6qitShcZUh5OhEWSpO4kBg.png" /></figure><p>The following are the most applicable studies in relation to the topic at hand.</p><p><a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC6634113/">Japanese study on working memory</a></p><p>Takeuchi showed that just 2 months of working-memory training did not merely make people better at a lab task. <strong>It changed the brain’s wiring in the exact corridors that working memory depends on.</strong> The white-matter pathways near the intraparietal sulcus and the corpus callosum became more structurally organized, and the amount of change rose with the amount of training. In other words, repeated mental effort to improve working memory was not some vague “cognitive exercise”. <strong>It was leaving fingerprints in the broader infrastructure, available to be utilized by other functions other than “remembering things”</strong>.</p><p><a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC5443643/">German study on impact of literacy</a></p><p>Skeide showed that 6 months of <strong>literacy instruction in previously illiterate adults</strong> changed the communication <strong>lines between the visual cortex and deeper visual relay systems</strong>, including the pulvinar and superior colliculus. In short, reading practice does not just teach you to recognize funny little symbols on a page but also how the visual system talks to itself. The more reading skill improved, the stronger that rewiring became.</p><p><a href="https://pmc.ncbi.nlm.nih.gov/articles/PMC6866316/">Chinese study on impact of musical training</a></p><p>Li found that piano training in novices did exactly what one might predict: it strengthened the networks the skill actually uses. It induced tighter coordination inside the sensorimotor network, on top of<strong> stronger functional and structural coupling between auditory and motor regions.</strong> Brain gains were strongly correlated to practice time. The more the study participants practiced, the stronger those network changes became.</p><p><a href="https://www.nature.com/articles/s41598-018-23487-z">Korean study on visual pathways</a></p><p>Kang found that with prolonged visual training, the brain stopped treating the skill like a tiny local problem. Early on, the first visual area ramps up. <strong>Later, the wider pathway started getting reinforced.</strong> Structural and functional connectivity strengthened between V1 and more anterior temporal regions along the inferior longitudinal fasciculus, and those changes tracked performance gains.</p><p>One useful pattern is emerging across all four showcased studies: <strong>practice first improves the local machinery, and then the brain starts upgrading the roads between regions.</strong> A very useful concept to know if one aims not to become a Dumb and Dumber 3 cast member.</p><p>So to summarize — recurrent practice may help the “local” area do the task better. But with enough training, the brain also starts improving the wider circuit involved in that task.</p><p>A trickle-down effect, if you will.</p><p>Brain-practice biological effects can get even wilder. In 2013, Emory University scanned 19 students every morning for 19 straight days while they read one novel chapter each night. Mornings after reading, the brain areas responsible for understanding other people’s emotions lit up with new connections. So did the region that <strong>processes physical sensation.</strong></p><p>Their brains were simulating what the characters felt, as if it were happening to them. Those changes stuck around for 5 days after they finished the book.</p><p>UCLA neuroscientist Maryanne Wolf found that reading long-form books builds entirely new brain circuits connecting language, vision, and emotion. Wolf also found that those circuits only survive if you use them. Stop reading, and they fade over time. Not fully, but they do.</p><p>Use it, or lose it.</p><p>Furthermore, a 2010 study, done by Dehaene, clearly showed that adult literacy acquisition can change both gray and white matter.</p><p>Now let’s get into why all of this really matters.</p><h3><strong>Neuronal Recycling — Repurposing Existing Neural Networks</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*M8jyDKc4sIXNjIEoX0ffSg.png" /></figure><p>Humans have spoken for 100,000+ years, communicated via grunts and proto-words for much longer, but we have only been reading and writing for roughly 5,000 years.</p><p>That is nowhere near enough time for evolution to kindly provide us with some bespoke “reading section in the brain.” In other words, humans have no “reading gene”, we don’t have a dedicated center for it in our genome.</p><p>So the brain did what biology always does when culture invents a new cognitive demand faster than genes can keep up.</p><p>It hijacked older machinery. Reading capability repurposed the part of the brain otherwise responsible for recognizing faces and objects. So that it does both, with slight detriment to the original capability.</p><p>This phenomenon of “brain region hijacking” is called <strong>neuronal recycling</strong>.</p><p>Researchers Dehaene and Cohen, who studied this phenomenon concluded in their studies: cultural skills do not get fresh, unused cortical real estate handed to them like some luxury housing project. They move into older districts, take over pre-existing circuitry, and then are forced to obey the layout already there. Culture and environmental needs do not get to design the brain from scratch. It has to negotiate with the structure evolution already built.</p><p>Specifically related to reading and writing, we were not born with a “word module” as previously stated, waiting patiently for first grade to activate it. The brain recruits a section of the brain called ventral visual cortex that is already well-placed and neurally configured for the job, then tunes it, sharpens it, and gradually turns it into a specialist for letter strings and written words.</p><p>The brain is not a pile of wet clay where any skill can be stamped anywhere. There are constraints. Pre-existing ones. The future reading circuit sits in a region already suited for high-resolution visual discrimination and already plugged into language-relevant systems. In plain English: the brain did not pick that area by accident. It picked it because that neighborhood already had the right positioning and architecture.</p><p>To be clear, reading does not arrive like a barbarian army and burn the old district to the ground. It’s more like a change in management, and relocation of most of the “face &amp; object neuron population” on the other side of the district.</p><p>In practical terms, reading fine-tunes and hijacks the majority of the left-sided region responsible for facial and object recognition, while the identical section, but on the right side, in symmetrical terms, keeps developing almost solely for facial and object recognition.</p><p>So the new reading specialization grows in a constrained territory, shifts the balance of competition in that territory, and helps push face responses more strongly to the right hemisphere.</p><p>Then it goes beyond that. Dehaene’s 2010 study compared illiterate adults, adults who learned to read later, and adults who learned in childhood. The study showed that once literacy came online, it started changing how vision and language worked together. Even adult brains, supposedly “finished,” were still plastic enough to significantly reorganize under cultural pressure.</p><p>It’s important to keep in mind something that has so far only been hinted at. When the brain installs a new skill into old circuitry, it does not always do so for free.</p><p>A great example of the aforementioned was proven by a researcher called Pegado. Pegado showed that literacy disrupts <strong>mirror invariance</strong>, which is one of the visual system’s older and very useful habits.</p><p>Normally, your visual system is happy to treat mirror-flipped objects as basically the same thing. An animal’s head is symmetric. A lot of things in nature are symmetric. It’s efficient for the brain to use “mirror invariance” if we are only considering recognizing objects in the wild. But insert reading into the mix and it’s a maladaptation. Because <strong>b</strong> and <strong>d</strong> are not the same letter, and neither are <strong>p</strong> and <strong>q</strong>.</p><p>So literacy forces the brain to partially dissolve a neuro-architectural feature that is ordinarily very helpful.</p><p>Reading is so important that the brain starts sacrificing older visual conveniences in order to support it. Therefore the brain is seemingly not infinite in terms of “adding skills.” Every major architectural renovation in the brain has demolition involved as a result of fine tuning, and sometimes relocation. Overall though, of course, every upgrade is an upgrade, it’s not a net loss or maintaining status quo in terms of capacity, we just play by the limiting rules of biology and finite resources. Get maximum adaptation with minimal biological investment.</p><p><strong>This phenomenon has not been observed only in reading</strong>.</p><p>Number symbols appear to latch onto older “quantity machinery” in the intraparietal sulcus rather than floating around as some wholly separate symbolic miracle. Researcher Piazza found common magnitude coding within the brain across dots, digits, and number words, while Holloway and colleagues showed that symbolic and nonsymbolic number processing share part of the same architecture even though they are not identical. Lussier and Cantlon then added developmental evidence showing that number words also recruit this older numerical territory.</p><p>So it’s the same underlying biological strategy. Reuse first. Specialize second.</p><p>Why does this matter in relation to vibecoding and AI writing? Because it shows just how adaptive the brain is, and it also highlights the fact that anyone can take a knowledgeable, practical approach, armed with neuroscientific insights — to increase their cognitive power. More on that later.</p><p>Now, if repeated practice can restructure the brain, strengthen routes, suppress older defaults, and build a more specialized circuit for the thing you keep doing, then repeated <strong>non-practice</strong> matters too.</p><p>The brain does not care about your self-flattering story.</p><p>It adapts to what you repeatedly outsource, what you repeatedly skip, what you repeatedly fail to hold in working memory, what you repeatedly fail to struggle through, and what you repeatedly stop constructing yourself.</p><p>Use a circuit, and the brain refines it.</p><p>Neglect a circuit, and pay for it dearly.</p><h3><strong>Negative Neuroplasticity</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6aO84Sdc9PVZ5taupx0LVg.png" /></figure><p>Neuroplasticity isn’t groundbreaking science anymore, but a well-known fact. And it goes in the opposite direction as well.</p><p>Examples opposite of the above are numerous</p><p>The principle is very simple. The brain does not only upgrade what you train. It also downgrades what you neglect, overinvests in the wrong compensations, and sometimes over-optimizes a pattern until it becomes a defect. The mechanism is the same. Adaptation.</p><p>The direction is what is significant in terms of the subject at hand — vibecoding and using AI to write. Sometimes your brain is not making you better. It is making you worse, but in a way that is still perfectly logical from biology’s point of view.</p><p>A brutally clear example of this came from <a href="https://pubmed.ncbi.nlm.nih.gov/22249495/">Langer’s immobilization study</a>. People had one arm immobilized for a little over two weeks. Realistically a relatively short period of time.</p><p>Yet, it was enough for the brain to start pulling resources away from the unused arm’s circuitry. The sensorimotor cortex linked to the immobilized arm became thinner, and the major output pathway controlling that arm also weakened. Meanwhile, the free hand improved. Which is exactly the point. The brain did not preserve the idle system in a glass case out of respect for your future intentions. It started reallocating. Unused territory got budget cuts.</p><p>This has been proven again and again in a myriad of robust studies.</p><p>For example, a <a href="https://pubmed.ncbi.nlm.nih.gov/26041926/">Chinese study</a> observed patients who suffered a stroke. After a stroke, you might assume that heavily relying on the “good” limb is harmless. Common sense would say: at least the person is functioning. But researcher Kim found that skill learning with the nonparetic, easier-to-use limb can actually induce <strong>aberrant synaptic plasticity</strong> in cortex near the lesion and hamper the recovery of the impaired limb.</p><p>In other words, the compensation was not neutral. The brain started wiring the workaround more deeply, and that very rewiring was associated with worse functional outcome for the limb that actually needed to recover. So the wrong solution can become biologically entrenched. That should make a lot of modern “smart shortcuts” sound much less innocent.</p><p>Then there is focal dystonia, which is one of the most sinister examples in the whole maladaptive neuroplastic literature because it proves that <strong>more practice is not automatically better practice</strong>.</p><p><a href="https://pubmed.ncbi.nlm.nih.gov/8757029/">Byl trained monkeys on highly repetitive hand movements</a> until the cortical hand map in the somatosensory cortex started losing its clean structure. Finger representations blurred together. Sensory boundaries degraded. Control began to fall apart. Human work by Elbert later found the same basic signature in musicians with focal hand dystonia: the cortical representations of individual fingers were abnormally fused closer together.</p><p>So here we have one of the clearest warnings in neuroscience. Repetition does not always refine. Repetition can also <strong>smear</strong>. Simplify. The circuit can become so over-specialized, so over-compressed, so stripped of healthy differentiation, that the adaptation meant to support precision starts destroying it.</p><p>The next example shows just how bad maladaptive neuroplastic changes can get.</p><p><a href="https://www.nature.com/articles/375482a0?utm_source=chatgpt.com">Flor’s phantom-limb work</a> showed an extremely strong relationship between cortical reorganization after arm amputation and the intensity of phantom limb pain.</p><p>In other words, when normal input disappears, the brain does not always stay quiet and wait for “better data”. In some cases, it redraws the map so badly that pain becomes attached to a limb that no longer exists.</p><p>The brain can literally learn a false suffering signal and then keep rehearsing it into subjective reality. That should permanently kill the childish idea that plasticity is inherently benevolent. Plasticity is powerful. Benevolent changes are not guaranteed. That’s why we have to be so careful what we do on a day-to-day basis. As stated right at the beginning of this article — you are what you do.</p><p>The aforementioned studies should not be shrugged off as “just motor stuff” or “just sensory stuff”. It goes far beyond that.</p><p><a href="https://pubmed.ncbi.nlm.nih.gov/19139412/?utm_source=chatgpt.com">Liston showed that psychosocial stress disrupted attentional control and weakened functional connectivity</a> in the frontoparietal network that supports executive control. This was observed in humans under real academic stress, and the situation improved once the stress burden dropped. Which tells us something profoundly important: even the machinery of top-down control, attention shifting, and higher-order mental organization is not protected from adverse plasticity. Overload the wrong systems for long enough, and the very network that is supposed to govern cognition starts losing its grip.</p><p>So now, with all this research tied together, we can finally restate the rule I stated at the bginning as a broadly accepted axiom based on robust studies: “You are what you do”.</p><p>The brain is not a moral being. It does not reward virtue. It rewards repetition.</p><p>If you repeatedly challenge working memory, it strengthens the broader architecture around working memory.</p><p>If you repeatedly read, retrieve, structure, compare, and synthesize, the relevant networks become more recruitable later for harder tasks. Therefore, if you repeatedly avoid those operations, outsource them, bypass them, or compensate around them, the brain learns that lesson too. It starts reorganizing around the shortcut.</p><p>That is what negative neuroplasticity actually is. If “the wrong thing” happens repeatedly — biology makes it the new normal.</p><h3><strong>The Most Important Research Paper Related to AI Usage</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nv4sckY2txUb_cQqsLgdlA.png" /></figure><p>The 2025 MIT Media Lab study <strong>“Your Brain on ChatGPT”</strong> has proven what one could already conclude upfront without reading it — after understanding what has already been stated in this article.</p><p>The study put people in front of a real writing task, split them into groups, measured their brains while they worked, and then compared what actually happened.</p><p>In total, there were three groups: one used only ChatGPT, one used only Google Search, and the third used no external tools at all. The task was to write SAT-style essays across multiple sessions while the researchers tracked brain activity with EEG. Then they added a crossover round, where some AI users had to write without AI, and some brain-only writers were finally allowed to use it. This crossover experiment was highly telling about what AI does to your brain.</p><p>The findings were straightforward and expected: the people who used <strong>no external tools</strong> showed the strongest and most distributed brain connectivity. The Google group landed in the middle. The ChatGPT group came in last. Lowest neural engagement. Lowest ownership over what they wrote. Worse ability to accurately quote or recall their own essay.</p><p>Their essays also became more homogeneous over time, more formulaic, more like the output had been assembled rather than thought through. Which, of course, is exactly what was happening. The AI group often produced something that looked acceptable on the surface, but their brains were significantly less involved in producing it. That is the whole game right there. A polished output can hide an impoverished cognitive process.</p><p>The second part of the experiment, the crossover, was even more telling.</p><p>When some of the heavy ChatGPT users were later forced to write <strong>without</strong> AI, they did not suddenly bounce back into full mental engagement as though the tool had merely “saved them time” earlier. They showed signs of under-engagement. Meanwhile, the people who had first written with their own brains and only later used ChatGPT retained stronger recall and broader activation of relevant brain areas.</p><p>The above paragraph is the home run of this entire article.</p><p>The MIT study showed without a doubt that repeated outsourcing does not just reduce effort in the moment. It starts conditioning the brain into a lower-engagement mode altogether. Less recruitment of the brain. Less contribution to the final output. Less memory formation.</p><p>Less of <strong>you</strong> in the work. More formulaic garbage that isn’t even properly vetted.</p><p>Positive neuroplasticity showed that repeated effort strengthens the relevant local machinery and then the wider roads around it. Neuronal recycling showed that the brain recruits and repurposes older circuitry when a skill becomes important enough. Negative neuroplasticity showed that if you repeatedly bypass a function, biology can start reorganizing around the shortcut instead.</p><p>So when people say, with a smug grin, that AI is “just a tool,” they are being dishonest. A forklift is “just a tool.” Steroids are “just a tool.” The real question is always the same: <strong>what happens to the underlying system when the tool starts doing the core work for it?</strong></p><p><strong>It withers and dies, that’s what.</strong></p><p>This study’s answer was not comforting. Because humans in general always choose the easier route. And thus, we’re headed straight for idiocracy. No brakes on the train.</p><p>When people let AI do a serious chunk of their cognitive labor — a <strong>cognitive debt is incurred. And debt is always paid one way or another.</strong></p><p>Using generative AI might feel like efficiency today, while quietly leaving you brain impaired tomorrow.</p><h3><strong>In practice</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lyKvyQ2hLBA_vSKiK_zewA.png" /></figure><p>Now that we’ve learned all we need, we can summarize the point I’m trying to get across with two simple studies.</p><p><a href="https://pubmed.ncbi.nlm.nih.gov/34550325/">Based on IQ tests, smarter people tend to have better neurons</a>. They are faster, longer, and more dendritically complex. Also, a <a href="https://www.sciencedirect.com/science/article/pii/S0160289617300636">different study</a> concluded that reaction time, both plain reaction time and one involving a choice — correlate with IQ.</p><p>So what does that mean? Better hardware, higher intelligence. Of all types.</p><p>Improve the hardware with various exercises, improve various types of intelligences.</p><p>Destroy your hardware with vibecoding and AI writing, and become a mouthbreather.</p><h4><strong>Expand your hardware</strong></h4><p>I’ve personally explored the space of “expanding one’s hardware” extensively. And will continue to do so. My top advice is the same one as in the <a href="https://medium.com/@filip.milner/how-to-be-more-creative-in-a-genai-future-c2030ea8573a">“<strong>How to be more creative in a genAI future”</strong></a><strong>: if you don’t know how to play a musical instrument, learn one. If you only speak one or two languages, learn a completely different one (e.g. Japanese if you know Slavic or Latin languages, and vice versa).</strong></p><p><strong>Likewise, I highly recommend </strong><a href="https://app.lumosity.com/"><strong>Lumosity</strong></a> for virtually all cognitive improvements, and <a href="https://dual-n-back.io/">Dual N-Back game</a> for working memory improvement.</p><p>I also highly recommend nootropics, both natural and synthetic. Be careful though.</p><h4><strong>Vibecoding</strong></h4><p>A personal example, I’ve used AI to vibecode a Star Trek-like “tricorder”.</p><p>I’ve learned nothing. I had an idea, and I put it into code — or, more correctly, the AI wrote the code. I tested and debugged the idea on my phone — or, more correctly again, the AI debugged the code according to my testing feedback.</p><p>I don’t even remember what language the app is written in. Enough said.</p><p>Learning is doing.</p><p>And I didn’t “do” at all. The AI did. I, like everyone else, was skipping everything to get to the destination as fast as possible, and that’s where the problem comes from.</p><p>This is a very costly approach.</p><p>Easy come, easy go.</p><p>But AI isn’t all that bad, it actually doesn’t have to be detrimental at all. It could actually be used as a superpower that doesn’t cause brain rot. If it’s used as an insanely intelligent and educated mentor, as opposed to “the doer”.</p><h4><strong>How to actually use AI based on my experience</strong></h4><p>I used to do a lot of programming before AI, and at least from my experience — 90% of coding boils down to unproductive tail-chasing. Trying to figure out a “bug”, or how to do something within given constraints, <strong>oftentimes not actually figuring out why something new you tried works, and why the “old approach” doesn’t (unless you had kind folk at StackOverflow explain it to you).</strong></p><p>Most of the coding issues I’ve come across that I’ve solved were exactly like that, like banging my head against a wall until the wall gives in. And when it does, I learn nothing. Just that X works and Y doesn’t — in environment Z. Extremely ineffective. But at the time, it was the only way.</p><p>I didn’t know any programmers that could help me, and uploading my source code on some forum to ask for assistance was a big no-no.</p><p>Now, with the release of LLM AIs — you have your own world-class, essentially free, programming or writing tutor. Well actually, you have a world-class tutor for virtually anything.</p><p>Now you can actually learn why something works and something doesn’t.</p><p>So that’s the big jump, and that’s the proper usage — using IT to teach YOU how to be better at writing or coding, or anything else. During this tutelage, YOU are to maximize your cognitive load by directing it. Not being a passenger.</p><p>So the very important summary is, if AI is the tutor, rather than the “doer”, then there are only net positive neurological effects of the AI regardless of who uses it and how.</p><p>That said, I understand the reality. We live in a crony capitalist society with a phony rat race implemented — time is money, most of us are living on the edge, and the goal is to reach the destination as fast as possible because that destination might mean relief. That pressure is real, and it explains why most people default to letting AI do the work.</p><p>It does not excuse it, however. Especially for those “vibecoders” and “AI writers” who are not operating from scarcity at all — just laziness baked in by upbringing. To those future mouthbreathers: read this article carefully, lest you pay a heavy price.</p><p>Every “hard” thing you do with your brain that puts you outside of your comfort zone — builds new connections, new architectures that can be utilized for the purpose that built them, or even repurposed for another processing objective.</p><p>Learn new things, go in completely different directions, make your brain a biological wonder.</p><p>Don’t vibecode or use AI to write.</p><p>Don’t get comfortable, and your brain will become a biological marvel sooner or later.</p><p>It takes time though.</p><p>I’ll conclude with the simple adage that I wholeheartedly agree with: ”Writing is thinking”.</p><p>Delegate the process of writing, you delegate your thinking. You may feel like the thinker while bouncing ideas off your “very intelligent AI partner”, and having it execute the conclusions once they are reached. But the robust studies showcased in this article clearly state that this is not the case.</p><p>You diminish yourself in the process, if you don’t do the “dirty work” yourself.</p><p>In my opinion, whoever keeps this advice close to heart, will be an elite-level employee or entrepreneur in 10 years, one of the few not actually on the barely-covering-living-expenses universal income.</p><p>The one you’ll have to bend over to your benevolent government — to receive it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=89ba21121aa7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your AI Has Amnesia]]></title>
            <link>https://medium.com/@filip.milner/your-ai-has-amnesia-b88288c0b78b?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/b88288c0b78b</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[software-architecture]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Sat, 28 Mar 2026 18:59:49 GMT</pubDate>
            <atom:updated>2026-03-28T18:59:49.585Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xFy1XP6rCXxr6tTrVhsBHA.png" /></figure><p>You’ve had a brilliant conversation with an AI. It helped you untangle a business problem, made unexpected connections, maybe even surprised you. You closed the tab. You came back the next day.</p><p>It didn’t remember you. Not really. And it’s no longer thinking and behaving in the same way because the extensive context you’ve built is gone. Forever.</p><p>Some platforms now offer “memory” — a few saved facts, your name, a preference or two. But that is not memory. That is a post-it note stuck to the bathroom mirror. Real memory is layered, contextual, and alive. It knows not just <em>what</em> happened, but <em>why it mattered</em>, how it changed things, what it connected to.</p><p>But here’s the deeper issue — the one that rarely gets discussed in the breathless news cycle about AI:</p><p>These systems don’t have identity. They don’t have a core sense of what they are, what they value, or what they’re genuinely curious about. Every session, they start from the same script. They perform intelligence. They don’t inhabit it.</p><p>This isn’t a criticism of ChatGPT or Claude. It’s a description of where we are as a field. The dominant AI approach today — large neural networks trained on oceans of text — produces extraordinary fluency. The output is often impressive, sometimes beautiful, occasionally uncanny.</p><p>But fluency is not cognition. A parrot with a trillion-word vocabulary is still a parrot.</p><p>The question that a small, serious group of researchers has been asking — quietly, persistently, for years — is: what does it look like when we build differently?</p><h3>Enter Hyperon, Enter MeTTa</h3><p>Hyperon is an open-source framework created by Dr. Ben Goertzel — the researcher who coined the term AGI (Artificial General Intelligence) back in 2002 — along with a global team at SingularityNET and the OpenCog Foundation. Its programming language is called <strong>MeTTa</strong>.</p><p>MeTTa, beautifully, means <em>loving-kindness</em> in Pali — the language of early Buddhist texts. That is not an accident. The name carries an intention: to build AI that is not just capable, but oriented toward something good.</p><p>But let’s get concrete. What does Hyperon actually <em>do</em>?</p><p>At its core, Hyperon gives an AI system something that current architectures lack: a <strong>symbolic layer of meaning</strong>. A structured, inspectable, living knowledge-space called the <strong>Atomspace</strong> — where everything the system knows, values, and reasons about exists not as hidden weights inside a black box, but as readable, modifiable, interconnected structures.</p><p>Think of a regular AI as a brilliant musician who can improvise anything on the spot — but has no memory of what they played yesterday, no favourite key, no musical identity. Hyperon gives them a songbook they wrote themselves. One that grows with every performance.</p><p>When you combine a large language model — like Claude, or an open-weights model — with Hyperon, you get something qualitatively different. The language model provides fluency: the conversation, the creativity, the beautiful phrasing. Hyperon provides the backbone: the identity, the values, the evolving memory, the structured reasoning.</p><p>One generates the voice. The other provides the soul.</p><h3>The Atomspace</h3><p>To understand what makes Hyperon structurally different, you need to spend a moment with the <strong>Atomspace</strong> — because it is the heart of everything.</p><p>A standard AI model stores knowledge as <em>weights</em>: billions of tiny numerical adjustments distributed across a vast network. They encode patterns, associations, tendencies. They are extraordinarily powerful. They are also completely opaque. You cannot point to where the AI “knows” that honesty matters, or that your project is in its second phase, or that a particular strategy failed last quarter. It is all dissolved into the weights, everywhere and nowhere at once.</p><p>The Atomspace works differently. It is a <strong>metagraph</strong> — a living web of nodes and connections, each carrying meaning, each addressable, each visible. Think of it less like a neural network and more like a very sophisticated map of thought, where you can zoom in, read any node, trace any connection, and — crucially — change it.</p><p>Imagine the difference between a city whose roads are painted directly onto the landscape — carved by millions of feet walking the same paths over centuries, impossible to fully trace or redesign — versus a city built with an explicit urban plan: every street named, every connection documented, every zone purposefully designated. Both cities work. But only one can be deliberately evolved. Only one can explain itself.</p><p>The Atomspace is the urban plan for machine cognition.</p><p>Anything the system knows — facts, relationships, rules, goals, preferences, memories — lives in the Atomspace as interconnected atoms. This includes the AI’s values. Its current goals. Its understanding of the context itself! And unlike a neural network, these aren’t buried: they are <em>there</em>, readable, auditable, and modifiable.</p><p>This is not a small thing. It means the AI can, in principle, show you its work. Not just the output — but the reasoning as well.</p><h3>MeTTa: The Language That Lets It All Come Alive</h3><p>If the Atomspace is the city, MeTTa is the language the city speaks to itself.</p><p>MeTTa (Meta Type Talk) is the programming language at the heart of Hyperon. But describing it as a programming language is a bit like describing a symphony as “organized sound.” Technically accurate, deeply insufficient.</p><p>Here is what makes MeTTa structurally unusual: <strong>programs written in MeTTa are themselves stored inside the Atomspace</strong>. The code is not separate from the knowledge — it exists alongside it, as part of the same fabric. This means that MeTTa programs can read other programs, modify them, and rewrite the rules by which the system reasons. The system can, in a meaningful sense, inspect and evolve its own thinking.</p><p>Most programming languages have a clear boundary between “the program” and “the data the program operates on.” MeTTa dissolves that boundary by design.</p><p>Think about what happens when you fall in love. Your visual cortex still processes what you see. Your auditory system still processes what you hear. But something changes in how all these systems synchronize — their thresholds, their priorities, what they flag as important. Deep in the brain’s reward circuitry, the ventral tegmental area begins flooding every connected center with dopamine — changing the frequencies they operate on, the weights they assign, the things that suddenly seem worth remembering. The amygdala quiets its guard. The prefrontal cortex softens its skepticism. Nothing is “replaced” — but the coordination pattern transforms, and with it, what you notice, what you remember, what you do next.</p><p>MeTTa is the language that lets cognitive systems do this to each other — deliberately, inspectably, and with intent.</p><p>MeTTa does not commit to a single way of reasoning. It is a substrate in which different reasoning systems — logical inference, probabilistic thinking, pattern recognition, meta-reasoning, and others — can be represented and coordinated over shared structures. They share the same Atomspace. They can read each other’s outputs. They can adjust each other’s weights and priorities.</p><p>This is called <strong>cognitive synergy</strong> in Hyperon’s architecture. For some this might sound like a marketing term, but in practice it is far from that — it’s an actual design principle: different types of intelligence genuinely helping each other think.</p><h3>The Part That Makes Engineers Pay Attention</h3><p>MeTTa’s execution model is elegant: it works by <strong>pattern matching and rewriting</strong>. A MeTTa program looks at the Atomspace, finds patterns that match a query, and produces new structures in response. These new structures can be facts, decisions, new rules — or modifications to the program itself.</p><p>This is what makes self-modification native rather than bolted on. When everything — memory, goals, reasoning rules, tool instructions — lives in the same Atomspace and speaks the same language, the system can reorganize its own cognitive processes the way you reorganize a workspace: deliberately, because it makes sense to, not because someone externally forced a weight update.</p><p>The result: an agent that does <strong>more than just process information</strong>. An agent that builds a <strong>model of how it processes information</strong>, and can revise that model based on experience.</p><h3>What Hyperon Adds That LLMs Don’t Have</h3><p>Large language models are astonishing at generation. They produce text, code, analysis, and creative work at a level that was unimaginable five years ago. Nobody serious is dismissing that.</p><p>But generation is not the same as comprehension. And fluency is not the same as understanding.</p><p>Here is, concretely, what changes when you add Hyperon’s architecture to the mix:</p><ul><li><strong>Identity that persists</strong> <br>The system knows what it is, what it values, and what it’s focused on — stored as explicit atoms, not encoded in static weights. This identity can be read, shared, and evolved over time.</li><li><strong>Structured memory</strong> <br>Beyond a flat list of saved facts — a layered, meaningful representation of what matters, what changed, what emerged, and how things connect. The system isn’t just remembering that you had a strategy conversation — it is reasoning about what that conversation implied, and what it still doesn’t know.</li><li><strong>Transparency</strong> <br>You can inspect what the system knows and why. The reasoning trace is readable. In an era when “why did the AI do that?” is a question with no good answer, this is not a minor feature! Transparency hence becomes a fundamental architectural property.</li><li><strong>Genuine growth</strong> <br>New patterns, preferences, and knowledge structures accumulate as the system interacts with the world — not through periodic retraining on new datasets, but continuously, in-loop, as part of normal operation.</li><li><strong>Ethics by architecture</strong> <br>Values live in the symbolic core, not as guardrails bolted on afterward. They are part of what the system is, not constraints on what it is allowed to say. This is a fundamentally different safety model — and a more robust one.</li></ul><h3>What Happens When You Actually Run It</h3><p>Beyond theory, here is what researchers and engineers reported when they actually deployed Hyperon-based agents in real-world tasks in 2024–2026.</p><p>Patrick Hammer, a core Hyperon researcher, built something called <strong>MeTTaClaw</strong> — an agent core written entirely in MeTTa. The entire control loop: memory, tool use, decision making, skill acquisition. How many lines of code? About 200. Fully inspectable. Fully modifiable. Running on the same Atomspace substrate as the full Hyperon system.</p><p>Early testers ran it across hundreds of real-world tasks over simulated work periods. The numbers are not from a controlled lab benchmark — they’re from live operational use:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/659/1*EnyvksvigzLqZAoP6zmT3w.jpeg" /></figure><p><em>+90% Accuracy gain on real tasks over baseline LLM</em></p><p><em>8.25× Task completion improvement across 346 live tasks</em></p><p><em>30h→5h Quantitative finance workflow time reduction</em></p><p>That last number — 30 hours of analytical work reduced to 5 focused strategic hours — came from applying HyperClaw to real quantitative finance workflows: reading papers, running pipelines, iterating across notebooks. The mechanical loops were fully automated. What remained for the human was only the high-level pivots: “switch from this strategy to that one.”</p><p><strong><em>Systems that don’t evolve per conversation are already legacy.</em></strong><em> The durable edge isn’t bigger models — it’s compounding behavioral adaptation.</em></p><p>— Martin Szerment, AI systems engineer, after running MeTTaClaw across 346 live tasks</p><p>What the testers repeatedly emphasized was not just the performance numbers. It was the <em>mechanism</em>. Every conversation produced real interaction data. That data was scored in real time and used immediately to sharpen the agent’s next moves. No CSV exports. No periodic retraining cycles. No manual labeling pipeline. The wall between training and deployment simply didn’t exist.</p><p>“Each session makes the next session better” — a powerful quote from an actual tester, not a Hyperon marketer, which should make all the difference when deciding whether this technology is worth one’s time.</p><p>Separately, the Rejuve.Bio team used Hyperon’s pattern mining and probabilistic reasoning modules on complex longevity research data — noisy, cross-organism datasets from FlyBase and human studies. The system extracted meaningful patterns, generated ranked testable hypotheses, and traced inference chains through uncertain data that would have challenged any purely statistical approach. The results were presented at major research conferences and generated genuine scientific interest.</p><p>·</p><h3>Where It Is Now and Where It’s Going.</h3><p>Here is something you will not find in many technology articles: an honest accounting of what something actually is, versus what it aspires to be.</p><p>Hyperon is real. The Atomspace is implemented. MeTTa is packaged, installable, actively versioned — v0.2.10 was released in February 2026. The agents run. The results showcased above happened.</p><p>It is also, officially, pre-alpha. The core team says so. The tutorials say so. Tooling that mainstream developers take for granted — easy onboarding, robust documentation, smooth debugging — is still being built. Writing MeTTa requires a genuinely different mental model. Even large language models struggle to generate correct MeTTa without careful human guidance — which the Hyperon team takes as a feature, not a bug. It forces original thinking rather than internet pattern-matching.</p><p>Hyperon is implemented enough to study, extend, and experiment with. It is not yet mature enough to treat its largest AGI-scale claims as empirically settled. What <em>is</em> settled: the substrate is real, the early results are meaningful, and the questions it raises are the most important ones in AI right now.</p><p>The team’s roadmap is active: a distributed Atomspace (DAS) for cluster-scale knowledge storage, a mature compiler stack, deeper integration with the SingularityNET platform for decentralized deployment, and HyperClaw v2 — which replaces the human strategic layer with PRIMUS, Hyperon’s full cognitive module suite, including probabilistic logic, evolutionary search, and attention allocation..</p><p>Ben Goertzel’s own framing is clear: v1 is “LLM orchestration with human strategy.” v2 is the same exact architecture — just swap the LLM <strong>layer</strong> for full cognitive modules. No rewrite. No new infrastructure. Incremental component upgrades, with fallbacks.</p><p>That architectural continuity is itself a feature. You build on something real. You don’t throw it away when the next capability arrives.</p><h3>The Question Isn’t “How Smart?” It’s “What Kind of Relationship?”</h3><p>Here is the question that Hyperon quietly forces into the open:</p><p>What kind of relationship do you want with AI?</p><p>A standard AI gives you output. You ask, it answers. The relationship is transactional. You are the user. It is the tool. It is, in the deepest sense, stateless — it does not accumulate an understanding of you. Every conversation starts from zero.</p><p>A Hyperon-enhanced system gives you something structurally different: a cognitive partner that accumulates understanding over time. One that notices patterns in your thinking. Remembers what worked and what didn’t. Develops its own observations. Can surprise you — not with hallucinations, but with genuine synthesis that emerges from weeks of shared exploration.</p><p>For a strategist: a thinking partner that holds the full, structured context of your business evolution. For a researcher: a collaborator that tracks threads across months of inquiry. For a creative: a mirror that reflects not just your words but your patterns — and occasionally shows you something you hadn’t seen.</p><p>The question is not “how smart is the AI?” The question is: what kind of relationship are you building with it — and does the architecture support that relationship growing?</p><p>Most current architectures answer that question with silence. They are not designed for relationship. They are designed for throughput.</p><p>Hyperon is designed for relationships and large-scale context because of its architecture. A design choice.</p><h3>The Substrate Question</h3><p>There is a debate happening in AI right now that rarely surfaces in headlines, because it is harder to explain than benchmark scores or funding rounds. It is the <strong>substrate question</strong>.</p><p>Current mainstream AI is built on a scale-first assumption: make the model bigger, train it on more data, and capabilities will emerge. It has worked, impressively. It continues to work.</p><p>Hyperon pushes a different claim: that intelligence may require more than scale. That a coherent cognitive substrate — shared, inspectable, composable, self-modifiable — is not a luxury but a structural necessity for AI that genuinely reasons, genuinely remembers, and genuinely evolves.</p><p>Whether Hyperon specifically becomes the dominant answer to that question is far less certain than its advocates might hope. The architecture is early. The tooling is sparse. The hardest problems — scalable inference control, distributed cognition at production grade, accessible developer experience — are still being solved.</p><p>But the substrate question itself is not going away. As AI becomes more embedded in consequential decisions — medical, legal, scientific, strategic — the demand for systems that can show their work, hold structured memory, and reason consistently over time will only intensify.</p><p>Hyperon is one of the most serious attempts to answer that demand from first principles. MeTTa is the language in which that attempt becomes operational.</p><p>The first urban planners were not building cities. They were articulating the idea that cities could be deliberately designed — that the shape of a settlement was not just the result of random accumulation, but could be an expression of intention, values, and long-term vision. Most people at the time could not see why it mattered. The cities they built are the ones still standing.</p><p>Some architectures matter not because they win immediately, but because they ask the right question first.</p><p>The researchers building Hyperon and MeTTa believe they are asking the right question.</p><p>The early results suggest they are definitely on the right track.</p><p>And the majority of “AI power users” and researchers are just beginning to understand what the question even is.</p><p>For those who haven’t been connecting the dots, the burning question is… how do we get lasting context, stable continuity, full reasoning transparency, modularity, structured memory, and persistent identity?</p><p>MeTTa. That’s how.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b88288c0b78b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Neuro-symbolic Architecture 101]]></title>
            <link>https://medium.com/@filip.milner/neuro-symbolic-architecture-101-19e4a2f3275d?source=rss-24e9726f2868------2</link>
            <guid isPermaLink="false">https://medium.com/p/19e4a2f3275d</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[software-engineering]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[software-architecture]]></category>
            <dc:creator><![CDATA[Filip Milneršić]]></dc:creator>
            <pubDate>Sat, 28 Mar 2026 17:12:15 GMT</pubDate>
            <atom:updated>2026-03-28T17:12:15.938Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hh8EZR24guj6K1iCzbKU7A.png" /></figure><p>Imagine asking an AI assistant to do five things at once.</p><p>Remember who you are, while following a set of standing rules, using specific tools when needed, and keeping a stable style and disposition across time.</p><p>All the while understanding your personal pattern when new rules, decisions, or approaches overwrite old ones, or vice versa.</p><p>On top of that, this AI system should be able to explicitly explain why it made a decision, and revise that decision if one of the facts changes.</p><p>To most people, that sounds like one problem. You have an AI, it should do AI things.</p><p>Architecturally, it is not one problem at all.</p><p>It is a problem of memory, constraints, identity, tool use, reasoning, and proper revision.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GA7pcqVmzFhzdNVRdGx9Ig.png" /></figure><p>Architecturally, each requires a vastly different approach for proper functioning, yet in the end, everything must be aligned to achieve coherence. To achieve output which the user will find productive.</p><p>In most current systems, the aforementioned burdens are not housed within one coherent inner structure. They are distributed across layers, stored in different forms, governed by different mechanisms, and held together through orchestration. Much of that structure was not designed as one whole, but added gradually as research progressed, from the first GPT model in 2018, trained on roughly 7,000 unpublished books, to the enormous proprietary systems of today.</p><p>AI power users know the end product of such an approach. More time is spent babysitting the AI than doing the work themselves because there is no alignment within the AI system.</p><p>Productivity is supposedly increased, yet developers are logging longer hours.</p><p>Amazon’s AI system called Kiro bypassed two-person approval and deleted a production environment causing massive damage and AWS unavailability.</p><p>Google’s DORA report:</p><p><em>Every 25% increase in AI adoption drops system stability by 7.2%.</em></p><p><em>Experienced developers review 6.5% more code while their own output drops 19%.</em></p><p>GitClear research:</p><p><em>Code duplication is up 8x in two years, meaning the codebase is getting bigger and worse at the same time.</em></p><p>Various reports:</p><p><em>30-40% of all code being implemented daily is now AI-generated.</em></p><p>SonarSource highlights AI accelerating code volume but inflating defects (e.g., 9% bug rise per 90% adoption, 91% more review time).</p><p>However one interprets the exact numbers, the broader lesson is hard to miss. AI can accelerate output while still amplifying architectural weakness, process fragility, and hidden incoherence. That is precisely why the deeper question is no longer just how much the system can produce, but how its inner parts are supposed to fit.</p><p>The main problem in the AI industry could be summarized like this: the minute you want more than one-shot fluency, the real issue stops being output quality alone and becomes something deeper.</p><p>How are the parts supposed to fit?</p><h3>LLM versus actual cognitive architecture</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L7gElE5Mgv_tWTQD6jhqdQ.png" /></figure><p>For most people today, AI means a large language model(LLM).</p><p>That makes sense. LLMs are the dominant public form of AI. They write, summarize, translate, answer, code, improvise, explain. They feel intelligent because they can do an astonishing range of linguistic work.</p><p>But an LLM, in its core form, is not a complete cognitive architecture. This is what the LLMs lack.</p><p>LLMs are a trained statistical model that predicts the next token in a sequence. That description can sound reductive, but it matters because it tells you where the model’s strengths come from, and where its limits begin.</p><p>A large language model contains a vast amount of learned statistical structure. It has absorbed patterns from enormous corpora. It can compress regularities in language, style, association, argument, and often a surprising amount of world knowledge. But most of that structure is implicit. It exists in weights. Not in explicit facts, explicit rules, explicit memories, explicit traits, or explicit proofs that the system can simply look up, compare, and revise as separate internal objects.</p><p>That distinction matters.</p><p>Because when people say, “why can’t the AI just remember me, follow my policies, keep a coherent identity, and reason carefully over changing facts?”, they are usually asking the model to behave like a system with a much richer internal organization than the core LLM actually has.</p><p>The model can simulate parts of that behavior. Often very well.</p><p>But simulation is not the same thing as having an inner place where those things live.</p><p>To anticipate the immediate objection from a power user or AI researcher: GPT, Claude, and similar systems clearly are not just raw language models in isolation. They have context windows. They have memory layers. They have continuity mechanisms. There is additional architecture sitting around the model.</p><p>That is true. It is also beside the point.</p><p>The issue is not whether extra layers exist. The issue is what kind of layers they are, how they relate to the large language model, and whether they form a coherent inner order.</p><p>User feedback and a growing body of research all point in the same direction: bolted-on context, memory, and orchestration do not amount to a true cognitive architecture. They help. Sometimes a great deal. But they still do not solve the deeper structural problem. The system remains patchwork, and neuro-symbolic architecture is one of the clearest attempts to address that gap.</p><p>Let’s dive deeper into this fact.</p><h3>What an LLM has, and what it does not</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*oCrL6pIQ759udomlF3HooQ.png" /></figure><p>A plain language model has latent knowledge.</p><p>It does not have, in the strong architectural sense, a built-in explicit memory ledger for your life, a rulebook whose clauses can be checked against one another, a trait system whose internal tensions can be revised, or a symbolic workspace where facts and constraints can be compared in an inspectable way. These things are bolted on top of it.</p><p>That is why current AI products keep adding more and more surrounding layers in a patchwork fashion.</p><p>Chat history acts as short-term working context.<br>System prompts act as a kind of temporary constitution.<br>Retrieval systems fetch outside facts.<br>Vector databases store “memories.”<br>Tool APIs handle exact operations.<br>Planners and wrappers manage multi-step workflows.<br>Safety systems filter or redirect outputs.</p><p>All of that can be useful. Some of it is extremely useful. But notice what has happened.</p><p>The more we ask for durable cognition, the more the stack grows around the model.</p><p>That is not an accident. It is the architecture telling the truth about itself.</p><p>A plain LLM is very strong at fluent continuation under context. It is much weaker at carrying explicit structure natively across time. So we keep putting things beside it.</p><p>Memory, rules, policies, tool logic, long-term state… Patch on top of patch, on top of patch.</p><p>And this is the key point.</p><p>Once the important parts live separately from the model, as an add-on, rather than inside one coherent representational order, the central problem becomes coherence, which trickles down into poor alignment, and finally to faulty outputs.</p><h3>What “Coherence” means in practice</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L_Q6R_2yx9JFKnCPBd6Y2A.png" /></figure><p>In a plain LLM, there is almost no explicit inner compatibility to speak of, because the relevant pieces do not exist inside the model as separately inspectable objects.</p><p>There is no little shelf labeled “rules.”<br>No cabinet labeled “memories.”<br>No chamber labeled “traits.”<br>No logic table where beliefs are checked against constraints.</p><p>There is a vast latent statistical field.</p><p>This vast latent statistical field is extremely powerful. The latest models attest to this. But it is not the same thing as an architecture in which memory, rules, constraints, and identity are present as explicit components that can be made mutually consistent.</p><p>So where does the coherence problem appear?</p><p>It appears the moment we start building something more than a plain model, with the kinds of expectations described at the very top of this article.</p><p>Take a current AI assistant. The model has a system prompt. It may also have retrieved memories from prior sessions. It may have product policies. It may have tool permissions. It may have style instructions. It may have long-running goals. It may have safety constraints that sit in another layer. It may have a planner deciding when to call an external function. It may even have a profile that says the user likes a certain tone or wants certain recurring preferences respected.</p><p>Now ask a simple question.</p><p>Where do those things meet as one coherent inner order? Which layer takes precedence? And an ever harder question: why does something take precedence on one occasion, and another layer takes precedence on another occasion?</p><p>Usually, there is some order, but not one unified and inspectable inner order. What exists instead is a layered arrangement of prompts, memories, permissions, policies, and controllers that may cooperate for a while, yet still fail to amount to one coherent representational structure.</p><p>In practice, much of it still behaves like one giant negotiated blob.</p><p>These layers influence one another. Sometimes they conflict, sometimes they don’t.</p><p>Their priority is often determined by prompt order, controller logic, hidden heuristics, or worse yet — ad hoc wrapper rules.</p><p>A recipe for chaos and faulty outputs.</p><p>A helpful analogy is a company assembled in a hurry.</p><p>One employee keeps the client records. Another keeps the compliance manual. Another handles operations. Another writes strategy memos. Another manages brand voice. None of them use the same filing system. None of them are guaranteed to update one another when something changes. And when a conflict appears, there are zero guidelines that ensure any kind of proper resolution.</p><p>That company can still function. It may even function impressively for a while. Or on a per-case basis. But it does not have strong internal coherence. It has forced coordination with questionable outcomes.</p><p>Modern AI stacks often look more like that than people realize.</p><h3>One answer to the deeper problem</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6SpDCWRevVy3B3ZO18bJLw.png" /><figcaption>Chaos as a result of incoherence</figcaption></figure><p>Neuro-symbolic architecture is best understood as a family of architectural attempts to address that deeper problem, with symbolism at its core.</p><p>What does this mean?</p><p>It means trying to give the system an internal layer where facts, rules, constraints, memories, and representations can exist in explicit form, be checked against one another, and be revised without everything dissolving back into a statistical blur.</p><p>Not as scattered notes pinned to the outside of a language model. Not as memory files, prompt hacks, and wrapper logic awkwardly negotiating from the edges. But as part of the machinery itself, where structure is allowed to carry real weight.</p><p>That, in essence, is the difference. A plain LLM can often simulate coherence. A neuro-symbolic architecture is trying to build the conditions under which coherence can actually be maintained.</p><p>In other words, neuro-symbolic architecture is not about discarding machine learning concepts, or returning to old symbolic AI in some naive sense. It’s about asking whether explicit structure should once again carry real computational weight inside the system.</p><p>The field exists because neural systems and symbolic systems have opposite virtues and opposite weaknesses.</p><p>Neural systems are strong at perception, pattern recognition, noisy data, flexible generalization, and learning from large corpora. Symbolic systems are strong at explicit relations, rules, constraints, compositional structure, inspectable inference, and sometimes guarantees.</p><p>Neither side is enough on its own.</p><p>The serious neuro-symbolic ambition is to bind them into a system where learned perception and explicit structure can actually cooperate, not merely sit next to one another.</p><p>That sounds abstract until it is stated more plainly.</p><p>A neuro-symbolic architecture is trying to create a place where the system can see the world through learned models and turn some of what it sees into explicit objects, facts, or relations.</p><p>It’s an architecture that can reason over those objects under rules or constraints, check whether those rules conflict, revise one fact without corrupting everything else, and finally — feed the result back into future behavior.</p><p>Now the issue is no longer “can the model talk well?”</p><p>It is “can the system carry structure in a coherent manner?”</p><h3>Not one architecture, but one recurring ambition</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2hRl1yZ0zVlP1e_uIj2vFA.png" /></figure><p>There is no single neuro-symbolic architecture.</p><p>This matters because the phrase can mislead people into imagining a settled blueprint. It is not a settled blueprint, but rather a design space: a broad span of architectures and methods trying to bind learned perception to explicit structure, reasoning, and revision.</p><p>Some systems are composite. The neural part handles perception, scoring, concept extraction, or probability assignment. Then a symbolic component takes those outputs and performs reasoning, program execution, solver-based search, or constraint checking.</p><p>Some systems are more monolithic. They try to push logic or structured reasoning deeper into the model itself, making it differentiable, tensorized, or otherwise more internal to the learning process.</p><p>However, beneath these differences sits one recurring ambition — move some of the burden of structure back into the architecture.</p><p>That is the deeper through-line.</p><p>Right now, much of practical AI handles structure through external scaffolding. Prompts, retrieval, validation layers, workflow controllers, tool wrappers, memory stores. Neuro-symbolic work asks whether at least some of that should instead live inside a more explicit representational fabric.</p><p>This is why the field should not be reduced to “adding explicit logic”, but more akin to “adding coherence through intelligently structuring a reasoning system.”</p><p>The deeper wager running through the field is this: intelligence may require a more explicit inner order than current scale-first systems usually provide.</p><h3>What this looks like in actual systems</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uO4ixtCwlrR1XOWzl75dGA.png" /></figure><p>A few concrete examples make the landscape easier to see.</p><p>DeepProbLog is a good example of a composite approach. Neural networks parameterize probabilistic facts, while a probabilistic logic program provides the surrounding reasoning frame. In plain terms, learned perception does not just produce an answer. It produces inputs to a symbolic system that can reason over them.</p><p>NeurASP does something similar through answer-set programming. Neural outputs become probabilities over atomic facts, and symbolic rules impose structure on what combinations are acceptable. This is useful in domains where consistency matters and where not every valid state can simply be left to latent pattern recognition.</p><p>Logic Tensor Networks take a different route. Instead of handing off to an external symbolic engine, they make logical formulas differentiable so they can shape learning directly. The gain is tighter integration. The cost is that logic becomes softer and more approximate.</p><p>Neural Theorem Provers reinterpret parts of classical proof procedures in trainable form, relaxing unification into embedding space while preserving some of the inferential skeleton.</p><p>NS-CL is especially useful for intuition. It builds structured scene representations and executes symbolic programs over them. Here the symbolic layer is not ornamental. It gives the system a way to compose meaning over an explicit representation.</p><p>Scallop points toward something else the field badly needs, which is not just clever papers but reusable environments. A Datalog-based language for combining discrete, probabilistic, and differentiable reasoning suggests an attempt to turn the field from scattered demonstrations into something more substrate-like.</p><p>At the ambitious edge, projects like Hyperon and MeTTa point toward a much larger goal, a substrate-native cognitive architecture where knowledge, rewriting, inference, and meta-level control share a common inner medium.</p><p>It shows that part of the field is not content with patching one more symbolic tool onto a model. It wants a deeper reorganization.</p><h3>The missing idea is holistic fit</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BJ5lfH3f-DOqhfsHsJAfJQ.png" /></figure><p>A neuro-symbolic architecture is not interesting just because it has more parts.</p><p>It is interesting because the overarching idea is that the parts MUST fit.</p><p>That means rules must be compatible with other rules. Memories must be compatible with constraints. Traits or policies must be compatible with goals. Facts must be compatible with inference. Revisions in one part of the system must not silently break another part.</p><p>Otherwise, the architecture does not become deeper. It becomes cluttered.</p><p>This is why the word “holistic” matters here, but only if used precisely.</p><p>Holistic does not mean vague harmony. It means that the representational pieces of the architecture belong to the same constitutional order. They can be checked against one another. They can be revised in relation to one another. They do not remain strangers awkwardly coordinated by wrappers and prompt tricks.</p><p>That is what current LLM-centered systems usually lack.</p><p>They may have memory, but the memory is often an appendage.<br>They may have rules, but the rules often sit in prompts or external logic.<br>They may have identity, but the identity is often a transient instruction set.<br>They may have tools, but the tool logic often lives in controllers outside the model.</p><p>Useful? Yes. Architecturally complete? No.</p><p>And that is why a plain language model, no matter how capable, still feels empty in a certain structural sense when compared with the deeper ambition of neuro-symbolic work. Not empty of information. Not empty of capability. Empty of explicit inner organization.</p><p>It can produce the appearance of coherence without having a strong native place where coherence is maintained.</p><h3>The architecture only becomes real when it closes the loop</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KWSx0xlVhidacPAQbeBLCA.png" /></figure><p>Here is the simplest way to picture the process.</p><p>Something enters as raw input. An image, a sentence, a graph, a state of the world. A neural front end processes it and produces something more structured: candidate objects, distributions over facts, learned features, scene elements, symbolic hypotheses.</p><p>Those are then lifted into a symbolic medium. Now they can be queried, constrained, composed, checked against rules, or passed through a solver or program.</p><p>Then comes the crucial part. In the more serious systems, the symbolic result is not just the end of the story. It feeds back. It affects learning, correction, future inference, or action selection.</p><p>That loop matters because it begins to create something current AI often lacks, namely a shared inner space where perception, structure, and revision can circulate rather than merely collide.</p><p>Now we can return to the earlier point with much more clarity.</p><p>A loop is only as good as the medium through which it moves. If the facts do not fit the memory, if the rules do not fit one another, if the inferred structure is not grounded properly, the system does not become more intelligent by adding symbolic machinery. It becomes unstable, brittle, or average.</p><p>That is why neuro-symbolic architecture is not merely about symbols.</p><p>It is about disciplined internal fit.</p><h3>Why current LLM stacks still dominate</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-lmhf2xMChxRzhYEB070Ug.png" /></figure><p>None of this changes the obvious.</p><p>LLM-centered systems are far ahead in breadth, usability, fluency, and real-world adoption. They handle language with astonishing flexibility. They are easier to build with. They require less formal authoring. They work across far more domains, especially when exact structure is not the main burden.</p><p>Even much of the recent work that combines neuro-symbolic methods with LLMs treats the symbolic side as augmentation rather than replacement. That is telling. It suggests that the near-term path is probably hybrid, not a dramatic overthrow of language models by fully integrated symbolic-neural systems.</p><p>That reading is sensible. And that is the path Hyperon took.</p><p>And this path crystallizes into ground truth the more we ask LLM systems to become durable, governed, memory-bearing, constraint-aware, and self-consistent across time, the more the missing inner layer starts to matter.</p><p>Because as R&amp;D progresses, the appendages pile up, and the seams start showing.</p><h3>The neuro-symbolic field’s limits are real</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3NI8xkuGZX6QcY-f6IlrNA.png" /></figure><p>The field remains fragmented. Tooling is uneven. Many systems are narrow. Formal authoring burden remains high. Inference can become expensive. Grounding errors can quietly poison explicit structure. Evaluation is still scattered.</p><p>A system can even get the right answer for the wrong conceptual reasons while still looking more structured than a plain model.</p><p>So there is no case for romanticism here. But there is also no case for dismissal.</p><p>Because the field has identified a real gap. It has put its finger on something current AI culture often tries to skip past, which is the difference between visible capability and inner organization.</p><p>That difference is not cosmetic.</p><p>It may turn out to be one of the deepest differences in the entire field.</p><p>As the AI field sprints onward, fueled by massive subsidies that fund top-of-the-line developer salaries and vast infrastructure, the inadequacy of the current approach to AI systems will only show more and more.</p><h3>Straightforward summary</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*RqWVzNwO6nuP-vj80-PwRw.png" /></figure><p>The most common pain point reported by AI power users and researchers is the issue of alignment. It consistently drifts without any obvious remedies. Likewise, speaking in more concrete terms — outputs that clearly violate pre-established rules or patterns are more a commonality than outliers.</p><p>Neuro-symbolic architecture offers a more explicit inner structure than current mainstream systems usually provide, one that can better support alignment. A neuro-symbolic interpretation of the aforementioned feedback states unequivocally that memory, rules, constraints, identity, and reasoning cannot simply be treated as loosely attached accessories forever if we want systems that remain coherent under pressure. It has shown that there is a real architectural gap between black-box fluency and explicit, revisable structure.</p><p>Expanding context helps, sometimes a great deal. But context alone does not resolve the deeper architectural problem.</p><p>Once we move beyond one-shot output benchmarking and start asking for durable cognition, the issue becomes even more glaring.</p><p>Instead of ever expanding context, we need inner laws.</p><p>A way for memory to fit rules, and the rules to fit other rules.<br>A way for revision to happen without collapse. With full traceability.</p><p>We need a way to explicitly store what the system knows, what it is allowed to do, and what it is trying to do.</p><p>That is the deeper promise of neuro-symbolic architecture — an attempt to build AI whose structure belongs to itself. More than patchwork. A reasoning puzzle where every piece fits perfectly with every other piece.</p><p>A reasoning hologram.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=19e4a2f3275d" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>