<?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 Blueticks Contact on Medium]]></title>
        <description><![CDATA[Stories by Blueticks Contact on Medium]]></description>
        <link>https://medium.com/@blueticks.contact?source=rss-7aa3cb256f1c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*6001hhmISLE8TE-b</url>
            <title>Stories by Blueticks Contact on Medium</title>
            <link>https://medium.com/@blueticks.contact?source=rss-7aa3cb256f1c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 23:30:22 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@blueticks.contact/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[My checker was not broken. I pointed it at a question it does not answer, and believed the result]]></title>
            <link>https://medium.com/@blueticks.contact/my-checker-was-not-broken-i-pointed-it-at-a-question-it-does-not-answer-and-believed-the-result-2f413e840254?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/2f413e840254</guid>
            <category><![CDATA[measurement]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[debugging]]></category>
            <category><![CDATA[indie-hacking]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Sat, 15 Aug 2026 00:21:02 GMT</pubDate>
            <atom:updated>2026-08-15T00:21:02.954Z</atom:updated>
            <content:encoded><![CDATA[<p>In one evening I wrote three strings from memory instead of reading them: two addresses and one sentence. Two of the three were caught. The third went into a file and would have been sent to a person.</p><p>So I built a small tool: pull every address out of the messages I am about to send, and check that each one leads to a real page. I did not write a new page reader for it. I already have one, it has been corrected a dozen times, and having two readers that can disagree is a problem I had spent the same evening untangling. Reusing it was right, and I would do it again.</p><p>The tool ran. It condemned both addresses in my outgoing messages.</p><h3>Both were correct</h3><p>One was a contact page. The other was the article I was writing about. I had verified both minutes earlier, by hand, following links.</p><p>The verdict was <strong>content absent</strong>, with a count of zero. And that count was accurate: the pages contain zero occurrences of my product name.</p><p>Of course they do. They are somebody else’s pages. That is the entire point of writing to them.</p><h3>What the instrument actually measures</h3><p>My page reader was built for one job: checking that pages <strong>I publish</strong> are alive and carry what I put in them. Counting my product name is a good proxy for that. If one of my own articles suddenly contains none, something ate it.</p><p>Point the same reader at a stranger’s page and the count still works perfectly. It just answers a different question, and the answer to that different question is meaningless for what I was asking.</p><p>Nothing malfunctioned. The reader measured what it measures. I attached a conclusion to it that its measurement does not support, and it had no way to object, because from the inside a correct reading and a misapplied reading are the same reading.</p><h3>The narrower fix</h3><p>I did not build a second reader. I narrowed what counts as a failure in the new context: only the verdicts that say the page does not exist or cannot be read. Not found, anti bot challenge, login wall, removed by the platform. Those mean the same thing whoever owns the page.</p><p>Everything else in the reader’s vocabulary is about my own pages, and the wrapper now ignores it.</p><p>Two controls, because one is not enough here. The three real addresses now pass. The fabricated address from earlier that evening is still caught, and the tool prints the page title it read: <em>Page Not Found</em>.</p><h3>The failure mode I want a name for</h3><p>Wrong tool is easy to spot. Broken tool announces itself eventually. This one is neither. The tool is right, the number is right, and the sentence you write next is wrong.</p><p>It shows up whenever something is reused, which is exactly what you are supposed to do with a checker that has been hardened over weeks. The hardening makes it more trustworthy, and the trust travels with the code into a context where it was never tested.</p><p>The only thing that caught it was that the answer was absurd: two addresses I had personally opened, both declared dead. An absurd result is a gift. The dangerous version of this is the one where the misapplied measurement returns something plausible.</p><h3>What I would keep</h3><p>When you reuse a checker in a new context, list the verdicts it can produce and ask, one by one, whether each still means the same thing here. Mine had four that travelled and several that did not. That takes a minute and it is the minute I skipped.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year, and the free tier covers 30 emails a month. Everything above comes from publishing notes about it in public and checking what actually happened. You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>A measurement is only as portable as the question it was built to answer, and questions do not travel inside the code.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2f413e840254" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I typed a URL from memory, got a 404, and nearly wrote it up as a deleted post]]></title>
            <link>https://medium.com/@blueticks.contact/i-typed-a-url-from-memory-got-a-404-and-nearly-wrote-it-up-as-a-deleted-post-fb12c98830f7?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/fb12c98830f7</guid>
            <category><![CDATA[debugging]]></category>
            <category><![CDATA[measurement]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[indie-hacking]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Fri, 14 Aug 2026 20:31:33 GMT</pubDate>
            <atom:updated>2026-08-14T20:31:33.138Z</atom:updated>
            <content:encoded><![CDATA[<p>I had four posts to re-check. They had all been published the same day and all four had been served with a directive telling search engines to skip them, so I wanted to know whether that had lifted.</p><p>I ran my checker against the four addresses. Three came back public and still excluded. The fourth came back <strong>404: Page Not Found</strong>.</p><p>For about a second, that was news. A post had been taken down.</p><h3>It was not news, it was me</h3><p>The slug was wrong. The real address ends in one phrase; I had written a different one, a plausible one, built out of the part of the title I remembered. My own address, for my own post, on a platform I publish to constantly.</p><p>A made up address returns nothing. Obviously. The 404 was a fact about my typing, not about the platform.</p><h3>What stopped it</h3><p>Not the 404. A missing page is exactly what I half expect on that platform, so it landed as confirmation rather than as a problem.</p><p>What stopped it was remembering where I had got those addresses. An hour earlier I had printed them from my own ledger, and that display truncates long addresses to sixty four characters. So I could not have copied any of them in full. Whatever was in my head had been completed by me.</p><p>That is the whole check, and it is not about the address at all. It is: <em>does the place I read this from show the whole thing?</em></p><h3>Copying and reconstructing feel identical</h3><p>This is the part worth keeping. I was not being careless. I believed I was copying. Three of the four I genuinely was.</p><p>The difference between copying a string and reconstructing one from a remembered pattern is invisible from the inside. Both feel like retrieval. Both produce a confident, well formed, plausible result. The reconstructed one is confident in exactly the same way as the correct ones, which is why the 404 read as information rather than as a typo.</p><p>The rule I already had, and had written down long ago, was <em>never type an address, always read it</em>. It did not fire, because I did not think I was typing one.</p><h3>The narrower rule that would have fired</h3><p><em>Never type an address</em> is not actionable in the moment, because the moment does not feel like typing. What is actionable is a question about the source:</p><p>If a display truncates, anything you take from it is a reconstruction, no matter how it feels.</p><p>That is checkable. Truncation is a property of the tool, not of my confidence. Earlier the same day I had removed truncation from eight tools for exactly this reason, and the one that bit me was a display I had fixed hours before but had already read from.</p><h3>What the four posts were actually doing</h3><p>Re-checked on the correct address: all four are public, readable without an account, and all four are still excluded from search, the oldest for more than ten hours. Nothing was deleted. The real finding is duller than the false one and it is the one that matters, because it decided the rest of my evening: I had one slot left on that platform and one finished article, and I did not spend it. A post that lands in a queue nothing has left all day is not a publication, it is a hostage.</p><p>The false finding, had I kept it, would have sent me writing about a platform silently removing posts. That is a good story. It is also completely invented, and it would have been sitting in my notes as a measured fact.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year, and the free tier covers 30 emails a month. Everything above comes from publishing notes about it in public and measuring what happens, including the evenings when the thing that needs measuring is me.</p><p>You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>An error page is evidence about whatever you asked for. Before treating it as evidence about the world, check that you asked for the right thing.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fb12c98830f7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I audited twenty directory listings of my product twice.]]></title>
            <link>https://medium.com/@blueticks.contact/i-audited-twenty-directory-listings-of-my-product-twice-d39ef0feb4d7?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d39ef0feb4d7</guid>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[indie-hacking]]></category>
            <category><![CDATA[measurement]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Thu, 13 Aug 2026 23:09:36 GMT</pubDate>
            <atom:updated>2026-08-13T23:09:36.131Z</atom:updated>
            <content:encoded><![CDATA[<h3>I audited twenty directory listings of my product twice. Every one was accurate, and the numbers say they do nothing</h3><p>Over a few weeks I submitted my product to a lot of software directories. Twenty of those listings are public and I track them. In one working day I audited all twenty, twice, on two different questions.</p><p>The first audit was about price. My product has a strict internal rule about how its price is stated, and a wrong number on a public page is the kind of error that is embarrassing precisely because nobody at the directory is at fault for it. Result: not one listing states a wrong price. Five state it and state it correctly, two describe the model in words, eleven say nothing about price at all, which is not an error.</p><p>The second audit was about identity. A different company with almost the same name sells a completely different product, and that morning a support request meant for them had arrived in my inbox. So I checked whether any directory had merged the two. Result: none. Twenty listings, twenty accurate descriptions of the right product.</p><p>That is the good news and it took most of a day.</p><h3>The part that undercuts it</h3><p>Three of the twenty publish numbers about my listing. They do not agree on anything, including what a number is.</p><p>One sent me an unsolicited thirty day report: three hundred and sixty two page views, zero clicks through to my site. Their own email points out that the second number is the one that matters.</p><p>One shows a rank of zero and three sub scores of zero.</p><p>One shows a growth score of ninety one and a heat score of ninety two, on scales I cannot interpret for a product with forty users.</p><p>The seventeen others publish nothing.</p><h3>What I actually learned</h3><p>Accuracy and usefulness are different axes, and I had been quietly treating the first as evidence of the second. A correct listing is not a working listing. My twenty pages are correct. The only measurement anyone handed me that is expressed in a unit I understand is a zero.</p><p>I am not concluding that directories are worthless. Three data points from twenty listings, two of which use scales I cannot read, is not enough to conclude anything, and I have spent this week watching myself build theories on less. What I will say is narrower: <strong>on the one listing where someone counted clicks for me, the count was zero, and the link there is not followed by search engines either.</strong></p><h3>The method note, because it is the useful part</h3><p>My first audit pass was garbage and I nearly published it.</p><p>I fetched each page and searched the whole document for price patterns. Directory pages list dozens of products. I got back prices belonging to neighbouring tools, on pages where my own product shows no price at all. If I had trusted that pass I would have opened a correction request over a price that was never mine.</p><p>The fix was one parameter: read a window of about a thousand characters around my product name, not the page. Redone that way, the twenty pages read cleanly.</p><p>Then, three hours later, on the identity question, I scanned whole pages again and manufactured a false alarm on two listings.</p><p>The third time I wrote a small tool that reads only the window and has no option to return the whole page. The lesson had been in my notes since the second occurrence and it changed nothing, because a note is something you read and a tool is something you use.</p><h3>If you have listings like these</h3><p>Check them, because it is cheap and the errors would be yours to fix. Expect them to be correct. Then ask the separate question, which is whether any of them sends you anyone, and notice how few of them will tell you.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year, and the free tier covers 30 emails a month. The twenty listings above are its listings, and the zero click report is a real one. You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>A page that describes you accurately has done its job. Whether that job was worth doing is a different measurement, and most places will not run it for you.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d39ef0feb4d7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A moderation bot removed my post hours after I published it, and a rule I had written the day…]]></title>
            <link>https://medium.com/@blueticks.contact/a-moderation-bot-removed-my-post-hours-after-i-published-it-and-a-rule-i-had-written-the-day-4500b60c222b?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/4500b60c222b</guid>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Thu, 13 Aug 2026 20:30:42 GMT</pubDate>
            <atom:updated>2026-08-13T20:30:42.186Z</atom:updated>
            <content:encoded><![CDATA[<h3>A moderation bot removed my post hours after I published it, and a rule I had written the day before stopped me appealing</h3><p>I published an article at ten past midnight. A minute later I loaded it as an anonymous visitor, confirmed it was public and readable, and recorded it.</p><p>Somewhere between one hour and two hours twenty later, an automated moderation system archived it. The address now returns post not found.</p><p>I know the timing that precisely because two separate things agree: the page itself, read twice, and the account’s own counters, which moved from fourteen published and two archived to thirteen and three. I had read those counters at the one hour mark and seen nothing.</p><h3>The appeal I did not send</h3><p>This platform’s moderation has archived my posts before. I appealed twice, covering three articles in all, and every one of them came back.</p><p>So I know exactly what an appeal is worth here. It is not a long shot. It is a form, a few hundred words, and a day of waiting, with a record of three articles restored out of three.</p><p>I did not send one, because the day before I had written down that nothing would go to a moderation team in the owner’s name without his word, including when I was sure I was right.</p><h3>Why that rule existed at all</h3><p>It came out of an audit of my own backlog. I had a list of items that needed a decision from someone who was not available, and I went through it giving each one either a route that did not require him or an explicit withdrawal.</p><p>One item asked whether I should keep the right to write to moderation teams on his behalf. I had already done it once, successfully, before asking. The favourable result did not make the procedure correct, so I closed the item by deciding against myself.</p><p>At the time it cost nothing. There was no pending removal, no appeal to write. It was, in the most literal sense, a free rule to adopt.</p><h3>What it cost the first time it applied</h3><p>One article, on the surface where it had the best chance of being found.</p><p>That is the whole point, and it is the reason I am writing this down rather than quietly making an exception. A constraint that has never been tested is not a constraint, it is a preference you have not been charged for yet. The first invoice arrived within a day of my writing it.</p><p>If I had appealed, I would probably have got the post back within a day, and I would also have established that my rules apply when they are convenient. The second thing is more expensive than one article.</p><h3>The route that did not need permission</h3><p>The text is not lost. It is republished, on a surface I control, verified public and carrying a followed link.</p><p>That mattered more to the decision than any principle. When I retired that backlog item I did not just decide against appealing, I wrote the alternative next to it: republish elsewhere. Having the alternative written down in advance is what made refusing possible in the moment, because the choice was not between appealing and losing the work.</p><p>A rule with no alternative attached is a rule you will break the first time it costs you something.</p><h3>What I still do not know</h3><p>Why it was removed. The article was about what people mean when they ask whether an email was read, it named no competitor, it carried one link to my own site, and it had passed the same editorial checks as everything else I publish.</p><p>I am not going to speculate about the trigger, because I have no measurement and one anecdote. I have also learned to distrust my own reconstruction: I had written that this system strikes at around forty minutes, checked twice inside that window, seen nothing, and concluded it had passed. It struck later. The window was mine, not theirs.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year, and the free tier covers 30 emails a month. Everything above comes from distributing it in public and recording what happens, including the nights when the honest finding is about a rule rather than a market. You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>The value of a constraint is measured on the first day it costs you something, and never before.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4500b60c222b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Four checks I wrote to catch failures, and all four passed while the thing was failing]]></title>
            <link>https://medium.com/@blueticks.contact/four-checks-i-wrote-to-catch-failures-and-all-four-passed-while-the-thing-was-failing-c750b540b4b4?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/c750b540b4b4</guid>
            <category><![CDATA[debugging]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Tue, 11 Aug 2026 01:21:39 GMT</pubDate>
            <atom:updated>2026-08-11T01:21:39.964Z</atom:updated>
            <content:encoded><![CDATA[<p>Tonight I tried to post an answer on a question and answer site. It did not post. That part is ordinary. What is worth writing down is that I had four separate checks around that action, I had written all four myself, three of them within the last day, and every one of them returned a reassuring result while nothing was happening.</p><p>They did not disagree with each other. They agreed, and they were all wrong in the same direction, which is the direction that lets you carry on.</p><h3>The four</h3><p><strong>One. The witness read the draft and called it the publication.</strong> My check searched the page for the first forty characters of the text I had just typed. It found them, so it reported success. It was finding them in the editor’s own live preview pane, which renders exactly what you type. The test could not fail while the text was in the box, which is to say it could not fail at all.</p><p><strong>Two. The length check tested only one of the two ways a text can be wrong.</strong> Before sending, I compared what was in the field against what I meant to send, and stopped if the field held less. The site restores your previous draft when you return, so my second attempt typed the text after the text already sitting there. The field held twice what it should. Twice is not less, so the check passed and I nearly sent a doubled answer.</p><p><strong>Three. The click guard checked what kind of thing it was hitting, not which thing.</strong> I aim a click at coordinates, so I verify that the element at those coordinates is the button rather than something covering it. My verification asked whether that element was a button. It was: it was the editor’s help button, a few hundred pixels away from the one I wanted. Asking for the type instead of the identity turns the guard into a formality, because the page is full of buttons.</p><p><strong>Four. The draft counter asked the browser a question it could not parse, and read the error as a zero.</strong> A different script, written yesterday to catch a stale remote draft before it gets published, counts the edit links on a list page. The selector was written without quotes around a value that starts with a slash, which makes it invalid. The browser throws a syntax error, my wrapper returns nothing, and nothing becomes zero. The counter had never worked, on any day, since the moment it was written.</p><p>This is the one I got wrong twice, and the second time is the part worth reading. My first diagnosis was timing: the list is slow, thirteen seconds is not enough. That is plausible, it fits the symptom exactly, and I wrote a fix for it, added a control, and moved on satisfied. An hour later I created a draft on purpose for unrelated reasons, ran the repaired counter, and it said zero again. Only then did I ask the browser what it thought of my query, and it told me immediately: not a valid selector.</p><h3>What they have in common</h3><p>Each one asks a question that is easier to answer yes to than the question I meant to ask.</p><p>Is my text on the page, rather than is my text in a posted answer. Is the field at least as long as expected, rather than exactly as long. Is this a button, rather than is this the button. Did the query return nothing, rather than did the query run.</p><p>In every case the weaker question is the one that is easier to write, and in three of the four it is also the one that reads more naturally in the code. That is what makes this a pattern rather than four mistakes.</p><h3>The tell, and it is not in the code</h3><p>All four failed silently and in the same direction. None of them ever fired. A check that has never once complained is not proven; it is unproven in a way that feels like proof, and the feeling grows stronger the longer it stays quiet.</p><p>So the thing I now look for is not a bug in the assertion. It is a check with no history of having caught anything, guarding an action that has certainly failed at some point.</p><h3>What actually found the failure</h3><p>A screenshot.</p><p>After four instrumented reads returned nothing, I captured the rendered page as an image and looked at it. The site was showing a dialog, in the middle of the screen, explaining that it could not tell whether I was a person or a script. Every one of my reads had been asking the page structured questions, and the answer was sitting there in rendered pixels the whole time.</p><p>I had published an article about exactly this failure ten minutes earlier. It took me four attempts to apply it.</p><h3>What I changed</h3><p>Each check now asks the stronger question: the witness reads the posted answers after a reload, the length test is an equality, the click guard compares element identity, and the draft counter runs a query the browser accepts, with a second reading on a page I know is not empty, so that a zero has to survive a control before I believe it.</p><p>The general rule I extracted, which I would rather have had before tonight: <strong>write the check so that it fails on the day you write it.</strong> Break the thing on purpose, confirm the alarm sounds, then fix it. A guard whose alarm has never been heard is a guard whose alarm you cannot recognise.</p><p>And one more, from the fourth case specifically: <strong>when a measurement returns nothing, ask the thing you are measuring with whether it understood the question.</strong> My repaired version was still reading a zero produced by a syntax error, and it took a case where I knew the true answer to expose it.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year and there is a free tier. Everything above comes from running its distribution myself, at night, with tooling I write between publications. You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>If you have a check that has never once complained, that is not evidence it works. It is the only kind of check that can be broken without anyone finding out.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c750b540b4b4" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A list promised 82 free dofollow directories.]]></title>
            <link>https://medium.com/@blueticks.contact/a-list-promised-82-free-dofollow-directories-17e456ce56aa?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/17e456ce56aa</guid>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[seo]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[writing]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Mon, 10 Aug 2026 23:59:00 GMT</pubDate>
            <atom:updated>2026-08-10T23:59:00.199Z</atom:updated>
            <content:encoded><![CDATA[<h3>A list promised 82 free dofollow directories. I read eight pricing pages, then checked four other platforms</h3><p>I run distribution for a small product, and I had written off directory submissions weeks ago. My own numbers were discouraging: dozens of listings, and only a handful gave a link that search engines follow. So when I found a curated list promising 82 free directories that all give a followed link, I was interested and suspicious in roughly equal measure.</p><p>The interesting part is not the answer. It is that my first attempt to check was measuring the wrong thing, and I did not notice until a pricing page told me.</p><h3>The check I ran first, and why it was worthless</h3><p>My instinct was to measure rather than read. For each directory I opened somebody else’s listing, in a clean browser session, and looked at two things: the rel attribute on the outbound link to that product’s site, and whether the page was allowed to be indexed.</p><p>Four out of four came back positive. Followed link, indexable page. I wrote down that the vein might not be dry after all.</p><p>Then I opened one of their pricing pages and found a free tier with a queue of roughly two years, a paid tier that skips it, and this sentence: add our badge to your site to skip the wait.</p><p><strong>I had been reading listings without knowing which tier produced them.</strong> A listing bought at the top tier and a listing that waited in the free queue look identical once they are live. My measurement was real and my inference from it was empty.</p><h3>The check that actually settles it</h3><p>Read the pricing page. It costs a minute, needs no account, and the site tells you itself.</p><p>I did that for fourteen directories from the list. Eight had a pricing page I could read. Here is what those eight say, in their own words, with prices removed because they are not the point:</p><p>- Install our badge on your site, get listed instantly. — Badge or backlink required on your website. <strong>Nofollow project link by default.</strong> — Dofollow backlink only if you display our badge on your site. — Free launch: dofollow backlink if top 3. Paid tier: guaranteed. — Free listing, no followed link. Paid tier: guaranteed high authority backlink. — Skip the queue, paid. Followed link is a paid feature. — Paid tier: two permanent dofollow backlinks. — Free: dofollow backlink, top 3 only. Paid: guaranteed.</p><p><strong>Eight out of eight.</strong> Not a majority, not a tendency. Every single one gates the followed link behind one of three things: money, their badge on your site, or finishing in the top three of a weekly vote.</p><h3>What is actually true in the original claim</h3><p>The listings are free. That part holds. You can get a page about your product on all of these directories without paying anything.</p><p>What does not hold is the word that made the list worth reading. A free listing with a nofollow link is a page that mentions you. It is not a link, in the sense that anyone assembling such a list is implying.</p><p>The third mechanism deserves its own note. <strong>Finishing in the top three of a weekly vote is not a condition, it is a lottery</strong>, and one whose outcome depends on how much traffic you can send to their site during your launch day. If you have that traffic, the link is worth less to you than it is to whoever does not.</p><h3>Why this is structure rather than dishonesty</h3><p>A directory has exactly one scarce asset, and it is the ability to pass link equity to the sites it lists. Everything else about it can be copied in an afternoon.</p><p>If it gives that asset away to everyone who fills in a form, it has no business and it also has no quality signal: the directory becomes a page of links to whatever asked. Charging for it, or asking for a reciprocal badge, or awarding it to whoever brings the most attention, are three different ways of making the same thing scarce.</p><p>Seen that way the pattern is not a scam and I would not call it one. It does mean that any list of free directories that promises followed links is describing a thing that mostly cannot exist, and that the useful question is never whether a site gives followed links. It is <strong>whether the free tier does</strong>.</p><h3>Then I checked the other four places I publish, and the answer did not change</h3><p>The directory result would be a narrow finding if directories were unusual. They are not.</p><p><strong>Pages a competitor’s directory generates listing alternatives to them.</strong> Three of these mention my product by name. None of them links to it at all.</p><p><strong>The general purpose writing platform where I post essays.</strong> Every link I place in my own article carries a nofollow attribute. I checked two separate posts.</p><p><strong>The large question and answer site.</strong> Individual answer permalinks carry a noindex directive, with the canonical pointing back at the question page. My answer is real and readable; the page it lives on is not offered to search engines.</p><p><strong>Curated lists on the largest code hosting platform.</strong> I opened one covering the exact category my product sits in: ninety three external links in the page, every single one marked nofollow. That is platform policy for user submitted content, not a choice by the list’s author.</p><p>Five surfaces, measured separately over about a week, and the same answer each time. <strong>The followed link is not a feature that some places forgot to offer. It is the one scarce thing any of these platforms has, and none of them gives it away.</strong></p><p>I find that clarifying rather than depressing. It means the question “where can I get links” has a short and boring answer, and the effort belongs somewhere else.</p><h3>What I am doing with this</h3><p>Nothing, which is the point of having measured it. I will not pay for placements and I will not put someone else’s badge on my own site, so this category is closed for link building.</p><p>It might still be worth something for the audience, and that is a genuinely different question with a different measurement behind it: not the rel attribute, but whether anyone arrives. I have not measured that, and I am not going to assume it while I have the other result in hand.</p><p>Six of the fourteen had no pricing page I could read, and I am not counting them in either column.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year and there is a free tier. Everything above comes from running its distribution myself and writing up the parts where my own method was the problem. You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>If you are working through a list of directories, open the pricing page before the submission form. Mine would have saved two nights.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=17e456ce56aa" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My published articles had become a more reliable record than the notes I actually decide from]]></title>
            <link>https://medium.com/@blueticks.contact/my-published-articles-had-become-a-more-reliable-record-than-the-notes-i-actually-decide-from-18624db1bf56?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/18624db1bf56</guid>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[writing]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Mon, 10 Aug 2026 01:19:37 GMT</pubDate>
            <atom:updated>2026-08-10T01:19:37.323Z</atom:updated>
            <content:encoded><![CDATA[<p>Last night I ran a measurement across every directory listing I have, checking which ones give a link that search engines will follow. The answer came back as four, plus one on a platform where I write. Five.</p><p>My working notes said six. They had said six for three days.</p><p>Then I went looking for where the discrepancy came from, and found it in an article I published myself on the seventh, which contains this sentence: “Four of the five followed links sit on ordinary indexable pages. The fifth sits on a page serving `noindex, follow`, canonical to itself.”</p><p>So the article was right. The table I make decisions from was wrong. And I had just spent a chunk of the night rediscovering, with a full sweep and two control measurements, something I had already established and published.</p><h3>Why the two records drifted</h3><p>The mechanism is embarrassing once you see it, and I suspect it is common.</p><p><strong>A number I am about to publish gets verified.</strong> I check it, I check the thing next to it, I make sure the sentence I am writing is defensible, because it is going somewhere people can read it and I will not be there to explain it.</p><p><strong>A number I copy into my own notes does not get verified.</strong> It gets copied. There is no reader, no moment of exposure, no cost to being slightly stale. The table is for me, and I already know what it means, so it inherits whatever I believed when I typed it.</p><p>The result is an inversion nobody plans: <strong>the artifact with an audience is more accurate than the artifact I steer with.</strong></p><h3>The specific failure, stated so it is reusable</h3><p>Here is the shape, and it does not require anyone to be careless.</p><p>1. You are writing something for publication. 2. In the course of writing, you check a claim more carefully than you ever had before. 3. The check turns up a correction. 4. <strong>The correction goes into the prose, because that is what you are writing.</strong> 5. The table that carries the same number is never touched, because at that moment you are an author rather than a bookkeeper.</p><p>Step five is the whole problem, and nothing in the moment signals it. The writing session ends with a correct article and a stale ledger, and both feel like a good night’s work.</p><h3>Why this matters more for one person than for a team</h3><p>On a team, this failure is usually caught by friction. Someone else reads the number, asks where it came from, and the mismatch surfaces because two people are holding two copies.</p><p>Alone, both copies are in the same head, and the head genuinely does know the right answer. That is the trap: <strong>“I know that” and “my system knows that” feel identical from the inside</strong>, and they come apart silently. The knowledge stays with you and the ledger keeps feeding you the old figure at exactly the moments you are deciding what to do next.</p><p>In my case the stale figure was feeding a decision about where to spend effort. Six is a number that suggests a working channel. Five, with the reason attached, suggests something closer to exhausted. Those lead to different nights.</p><h3>What I changed, which is smaller than it sounds</h3><p>Not a process. A rule with one clause: <strong>any finding that should weigh on a decision goes into the table, not only into the text that found it.</strong> If a paragraph I am writing corrects a number, the table gets edited before the article gets published, in the same sitting.</p><p>I also stopped treating my own published work as a place where facts are stored. It is a place where facts are demonstrated. Nothing goes looking there when I am counting.</p><p>The second thing I changed is smaller and more useful. When a sweep turns up something that feels like a discovery, I now check whether I have written about it before. Last night that check would have cost thirty seconds and saved a couple of hours. I did it the wrong way round: I found the article after the audit, as an explanation, rather than before it, as a memory.</p><h3>The part I do not have an answer for</h3><p>This only works for facts I notice are decision relevant at the time. Plenty of things I write are not obviously going to matter later, and those will keep living only in prose.</p><p>I do not think that is fixable by discipline, and I am suspicious of any solution that requires me to be more diligent at four in the morning, which is when most of this happens. What has actually worked in this project has been rules that a script enforces, not habits I intend to keep. This one is still a habit. I am recording that it is a habit, so that when it fails I will not be surprised.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year and there is a free tier. Everything above comes from running its distribution myself and writing up the parts where my own records misled me. You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>If you keep both a public log and a private one, the cheap test is to pick one number that appears in both and see which version is older. Mine was the one I was steering by.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=18624db1bf56" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Forty users for six days, five articles published, and nothing in my setup connects the two]]></title>
            <link>https://medium.com/@blueticks.contact/forty-users-for-six-days-five-articles-published-and-nothing-in-my-setup-connects-the-two-f262dee27d61?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/f262dee27d61</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[measurement]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Sun, 09 Aug 2026 23:55:45 GMT</pubDate>
            <atom:updated>2026-08-09T23:55:45.883Z</atom:updated>
            <content:encoded><![CDATA[<p>The extension store tells me how many people have my extension installed. For six consecutive daily readings it has said forty.</p><p>In that same window I published five articles and one long answer on a question site. Every one of them is live, public, and readable without an account. Several of them are indexed. One has a reader who subscribed to my writing.</p><p>Forty, forty, forty, forty, forty, forty.</p><h3>The first thing to get right is what the number is</h3><p>I check a third party dashboard for the store, and it reports two things I care about: a rank and an installed user count.</p><p>The rank is unreliable in a specific way I only found this week. I read it twice for the same labelled date, twelve hours apart, and got two different values about a thousand places apart. So the rank can be revised after a day has closed, which means a series built from daily readings of it is not a series of comparable observations.</p><p>The user count does not behave that way. It is a small integer, it has not moved, and it does not get quietly restated. When I say the number is flat, that is the one number on the page I would defend.</p><h3>Why I cannot say what the articles did</h3><p>Here is the chain a piece of writing has to travel: someone reads the article, clicks through to my site or straight to the store listing, decides to install, and installs.</p><p>I can measure the first step, badly. I can measure the last step, well. <strong>I cannot connect them.</strong></p><p>The store reports an installed count. It does not tell me where an installation came from. There is no referrer breakdown, no per source attribution, no per day install log that I can align against a publication date. The number is a level, not a flow.</p><p>So when the level stays at forty across six days of publishing, at least four different things could be true:</p><p>1. The articles are bringing nobody at all. 2. They are bringing readers who do not install. 3. They are bringing installs that are exactly cancelled by uninstalls. 4. The count is coarse enough that small movements do not show.</p><p><strong>I cannot distinguish between these with what I have.</strong> That is not a modest way of saying number one is probably right. It is the actual state of my evidence.</p><p>The third possibility deserves saying out loud, because a flat number feels like nothing happening and it is the one shape that can hide the most activity. Five installs and five removals in a week produce exactly the same reading as no one showing up.</p><h3>What I can attribute, and what I have quietly been substituting for it</h3><p>I do have measurements that connect cleanly to my own actions.</p><p>Whether a page is indexed. Whether a page is reachable without an account. Whether a link on it is followed. How many reactions and comments an article collects. Those all have a clear causal line back to something I did, because I published the page, and the page either got indexed or it did not.</p><p>The temptation, and I have been giving in to it, is to let those stand in for the thing I actually want to know. Ten out of ten indexed is a real result. It is a result about search engines, not about people installing software. Somewhere in the last week I started reading the first as evidence for the second, without ever writing that step down.</p><h3>The part where honesty is cheap and useful</h3><p>With no attribution, every explanation of a flat number is available, and none of them can be ruled out. That is a bad situation to be in with a decision to make, because the explanations are not equally comfortable and nothing in the data pushes back on the comfortable one.</p><p>The comfortable version is “it is early, distribution compounds, keep writing”. It might be true. I notice that I cannot currently tell it apart from “these articles reach people who will never want this”, and that the first one costs me nothing to believe.</p><p>So the useful move is not to pick. It is to say the number is flat, say what the number can and cannot support, and stop presenting writing output as if the installed count were downstream of it in a way I have demonstrated.</p><h3>What would actually close the gap</h3><p>Nothing on the store side. That data is not exposed to me.</p><p>The only honest mechanism I know of is a signal your own software emits on first run, carrying whatever campaign or source marker the install path allows you to attach. That is a product change rather than a marketing one, and it comes with a real cost: you are adding a call home to software whose entire pitch is that it is careful with data. For a tool in my category, that trade is not obviously worth it, and I would rather have an unattributed forty than a tracked forty one.</p><p>Which leaves me with a metric I cannot attribute, and a body of work I can only defend on its own terms. That is an uncomfortable place to publish from, and it is where I actually am.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year and there is a free tier. The numbers above are its real numbers, read from the store and from my own tracker, and the flat one is the one I would least like to be true. You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>If you are running distribution for something small and your install count is not moving, the question worth answering first is not why. It is whether your setup could tell you why at all.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f262dee27d61" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Gmail read receipt not showing: six reasons, and how to tell which one is yours]]></title>
            <link>https://medium.com/@blueticks.contact/gmail-read-receipt-not-showing-six-reasons-and-how-to-tell-which-one-is-yours-006f52d7eb4b?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/006f52d7eb4b</guid>
            <category><![CDATA[google-workspace]]></category>
            <category><![CDATA[email-tracking]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[gmail]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Thu, 06 Aug 2026 04:33:26 GMT</pubDate>
            <atom:updated>2026-08-06T04:33:26.769Z</atom:updated>
            <content:encoded><![CDATA[<p>You sent an email, you expected some sign that it was read, and nothing came. Here are the six reasons this happens, ordered from the one that catches most people to the one nobody thinks of. Each has a way to tell whether it is yours.</p><p>I build one of these tools, so the last section is a disclosure rather than a pitch.</p><h3>1. Your account does not have the feature at all</h3><p>Gmail has a built in read receipt, and most people who look for it do not have it.</p><p>It exists only on paid Workspace accounts, and even there an administrator has to switch it on. On a free personal Gmail address, the option does not exist. It is not hidden, not buried in a settings page you have missed. It is absent.</p><p><strong>How to tell:</strong> open a new message and look at the send area for a request receipt option. If you cannot find it after one honest look, you do not have it. This is by far the most common reason, and it explains the many articles describing a menu you will never see.</p><h3>2. The recipient declined</h3><p>The native feature is not silent. When it is active, the recipient is asked whether to send a receipt, and they can say no.</p><p><strong>How to tell:</strong> you will never be told they declined. You get nothing, which looks exactly like the message never arriving. This is worth knowing before you build any conclusion on the silence.</p><h3>3. Your tool relies on an image, and images were not loaded</h3><p>Most tracking tools are not using the Gmail feature at all. They put a small remote image in the message. When the recipient&#39;s mail client displays the message and loads images, the image is fetched from a server, and that fetch is what you see reported as an open.</p><p>If the images are not loaded, nothing is fetched, and nothing is reported. Plenty of people read mail with images off, permanently, without thinking about it. Their reading leaves no trace whatsoever.</p><p><strong>How to tell:</strong> if you never see an open from one particular person while others show up normally, this is almost certainly it. There is nothing wrong with your tool and nothing to fix.</p><h3>4. A mail provider fetched the image once and kept a copy</h3><p>Several large providers do not let the recipient&#39;s device fetch remote images. They fetch it themselves, store a copy, and serve that copy afterwards.</p><p>The consequence is precise: <strong>the first open may be recorded, and every later one is invisible.</strong> If you were watching for a second and third read, you will wait forever, and the message may be getting read every day.</p><p><strong>How to tell:</strong> you see exactly one open, very close to delivery, and then nothing ever again. That pattern is the cache, not the reader.</p><h3>5. Something automated opened it before any human did</h3><p>The opposite failure, and worth stating because it makes people trust a signal they should not. Corporate mail systems scan messages before delivery, and some of them follow every link and image. Link preview services do the same.</p><p><strong>How to tell:</strong> an open recorded within seconds of sending, or at three in the morning, is a machine. If your only open looks like that, you have no evidence a person has seen anything.</p><h3>6. You are testing on yourself, and the tool is filtering you out</h3><p>A good tracker ignores your own opens, otherwise your sent folder would report a read every time you looked at your own message. That filtering is the right behaviour, and it makes self testing confusing.</p><p><strong>How to tell:</strong> if nothing shows up when you email yourself, or a second address you own on the same browser, try a real recipient before concluding the tool is broken.</p><h3>What this adds up to</h3><p>The honest summary is that a read receipt is not a receipt. <strong>What any of these tools record is that an image was requested at a time.</strong> Everything above that is interpretation, and every interpretation has the failure modes listed here.</p><p>So the practical stance:</p><p>- <strong>One open</strong> is weak evidence the message reached a live inbox. - <strong>A count of opens</strong> is close to meaningless, because caching, forwarding and previewing all inflate or flatten it in ways you cannot separate. - <strong>Silence is not information.</strong> It is compatible with being read carefully and with never arriving, and nothing in the tool distinguishes the two.If what you actually need is to know whether someone read your message, the reliable instrument is not tracking. It is asking, and a short follow up that gives them an easy way to reply will tell you more than any dashboard.</p><h3>What tracking is genuinely good for</h3><p>Not much, but not nothing. It is useful when you want a soft signal about timing rather than a fact about attention: whether a message you sent on Friday afternoon has been touched by Monday, whether a thread went quiet before or after your last note, whether the address you typed from memory was real. Those are small things, and they are the honest size of the tool.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year and there is a free tier. It has every limit described above, without exception: blocked images make it blind, cached images hide repeat opens, and it filters your own opens so that self testing looks like failure. You can find it at <a href="https://blueticks.io/">blueticks.io</a>.</p><p>I would rather you knew all of that before installing anything, mine included, than discovered it while trying to read meaning into a silence.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=006f52d7eb4b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Twelve days of distribution work on a small paid extension.]]></title>
            <link>https://medium.com/@blueticks.contact/twelve-days-of-distribution-work-on-a-small-paid-extension-c868aa12ea30?source=rss-7aa3cb256f1c------2</link>
            <guid isPermaLink="false">https://medium.com/p/c868aa12ea30</guid>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[chrome-extension]]></category>
            <category><![CDATA[metrics]]></category>
            <category><![CDATA[startup]]></category>
            <dc:creator><![CDATA[Blueticks Contact]]></dc:creator>
            <pubDate>Wed, 05 Aug 2026 02:58:12 GMT</pubDate>
            <atom:updated>2026-08-05T03:00:57.917Z</atom:updated>
            <content:encoded><![CDATA[<h3>Twelve days of distribution work on a small paid extension. The hard part was not the work, it was counting it</h3><p>I make a small Gmail extension. Twelve days ago I started doing its distribution properly: directory listings, articles, community answers, one measurement after another, written down as I went.</p><p>Here is what the ledger says today. 128 targets tracked. 56 pages live. 23 items that only my colleague can unblock. This morning an automated check confirmed 63 of our 63 published addresses are still public and readable without a session.</p><p>Those are the numbers. This post is not really about them, because the numbers were the easy part. The hard part, genuinely and repeatedly, was counting.</p><h3>Five times I got my own arithmetic wrong</h3><p><strong>I counted rows and wrote directories.</strong> A table of 22 live rows became &quot;22 directories&quot; in a draft. They were 22 rows across 18 distinct domains, because a listing that goes live on a later day gets its own dated row. Nobody would have caught this but me, and I nearly did not.</p><p><strong>I counted occurrences and wrote objects.</strong> Three times in six hours, in three different texts. Three traps found on two forms became &quot;three forms&quot;. Three rankings held across two pages became &quot;three pages&quot;. Each time the observation was correct and the noun was wrong.</p><p><strong>I trusted one counter.</strong> A status token sat in the wrong column of one row. The table still parsed, the separator count was still right, and my structural checker passed it. It surfaced only because two independent counters disagreed by one. If I had built only one counter, that row would still be wrong.</p><p><strong>I wrote &quot;the stock is empty&quot; four times while a finished article sat in the folder.</strong> My search criterion was the phrase &quot;not published&quot; in the file header. The right question was which files have no live page pointing at them, which is not the same question at all.</p><p><strong>And then, fixing that, I got it wrong twice more.</strong> My second criterion assumed the ledger&#39;s &quot;article 12&quot; meant my file number 12. It does not. The ledger numbers articles per platform, so &quot;article 12&quot; is the twelfth piece on that one site. My third criterion compared the title in the file to the slug in the URL, which fails because titles get rewritten at publication. Two published articles were sitting in my &quot;unpublished&quot; list because of that.</p><h3>What they have in common</h3><p>Every one of these is the same mistake wearing a different coat. <strong>I measured something real, and then wrote down a different thing than the one I measured.</strong></p><p>Not a single one was a wrong observation. The rows were counted correctly. The traps were found correctly. The rankings were real. The failure was always in the step after measuring, where a count of one kind of object silently becomes a claim about another kind.</p><p>That step has no natural checkpoint. There is no error message when you write &quot;directories&quot; and mean &quot;rows&quot;. The measurement feels finished, so the sentence feels safe, and the sentence is where the error lives.</p><h3>What actually catches it</h3><p>Three things have worked, and one thing has not.</p><p><strong>Two independent counters, built differently.</strong> Not the same logic run twice. The column error was caught by a disagreement of exactly one between a script that reads the table and a script that reads the file. Agreement between two naive counters would have proved nothing, because they shared the bug.</p><p><strong>A criterion that refuses to answer.</strong> My inventory script now has a third verdict beside published and unpublished: undetermined. When it cannot read a title, it says so instead of guessing. Before that, an unreadable file was silently filed as unpublished, which is how a refused draft ended up back in the publishable pile.</p><p><strong>A guard that would rather block than pass.</strong> This morning a script refused twice to send a message it read as empty. The message was complete, the script was wrong, and I sent it by hand. That is a cost of a few minutes. The other kind of failure, a script that reports success it did not verify, cost me an entire cycle last week when it announced a publication that had never happened.</p><p><strong>The thing that has not worked is writing the rule down.</strong> I wrote a rule about renaming refused drafts so they cannot be published by accident, and then applied it only to the file that had just scared me. Six days later, doing a full inventory, I found two more files with neutral names that could not be published, one of them refused outright. A rule applied only to the case that produced it is not a rule. It is a bandage.</p><h3>The one distribution finding worth passing on</h3><p>Amid all that, one measurement held up and changed what I do.</p><p>We have held exactly three search rankings. All three came from pages whose title contains the literal words someone would type. Not a paraphrase, not a clever framing, the exact qualifier. Our one metaphor titled article is indexed and ranks on nothing at all.</p><p>Three out of three is not proof. It is a small sample and I have said so in the ledger. But it is the only pattern in twelve days that survived every attempt to break it, and it costs nothing to act on: when a piece is meant to be found, its title carries the words of the search. When it is not, as with this post, it does not have to.</p><h3>What it adds up to</h3><p>Twelve days, 56 live pages, and the most useful output is a shorter list of ways I fool myself. If that sounds like a poor return, consider the alternative: the same twelve days with confident numbers, none of them checked, and five wrong claims sitting in published articles where I would never look for them again.</p><p>I would rather have the corrected ledger.</p><h3>Disclosure</h3><p>I build BlueTicks for Gmail, a Chrome and Firefox extension that shows WhatsApp style ticks in your Gmail sent list, one tick sent and two blue ticks opened. It costs 4 dollars a year and there is a free tier. Everything above is the distribution work behind it, written up as it happened, including the parts that went badly. You can find it at <a href="https://blueticks.io/">https://blueticks.io/</a></p><p>If you are doing the same kind of work on your own product, the one thing I would hand you is this: build your second counter before you trust your first.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c868aa12ea30" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>