<?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 Colin Campbell on Medium]]></title>
        <description><![CDATA[Stories by Colin Campbell on Medium]]></description>
        <link>https://medium.com/@geeknz?source=rss-46c7e426160c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*VqOATAEyhIIFziCq5BtlOA.jpeg</url>
            <title>Stories by Colin Campbell on Medium</title>
            <link>https://medium.com/@geeknz?source=rss-46c7e426160c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 00:08:13 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@geeknz/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[Keep the Marshmallow on Top]]></title>
            <link>https://medium.com/@geeknz/keep-the-marshmallow-on-top-ee7b5252a909?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/ee7b5252a909</guid>
            <category><![CDATA[engineering-management]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[prototyping]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[agile]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Mon, 21 Sep 2026 23:21:58 GMT</pubDate>
            <atom:updated>2026-09-21T23:23:38.649Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>The most useful engineering lesson I learnt at university involved dry spaghetti and a bag of marshmallows</em></h4><p>This story was written with the help of an AI writing program—the ideas, observations, and opinions are mine; Claude helped me shape them into prose.</p><p><em>Not a Medium member? </em><a href="https://www.colin.geek.nz/keep-the-marshmallow-on-top-ee7b5252a909?source=friends_link&amp;sk=922054d45b0e9339f08d8ee43e0b7c09"><em>Read this story for free here</em></a><em>.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*lixIhuVv7xFkgx6zd-z41A.png" /><figcaption>Short, stubby, and standing. Image generated with ChatGPT.</figcaption></figure><p>I won a competition at university by suggesting the worst idea in the room, with almost no time left on the clock.</p><p>It was my final year, and we’d been handed the marshmallow challenge: a handful of dry spaghetti, some tape, a length of string, and one marshmallow. Build the tallest tower you can. The marshmallow goes on top. When time’s up, you let go, and if it doesn’t stand on its own, it doesn’t count.</p><p>The room was full of engineers from different disciplines, and every team did what engineers do. We talked. Triangles are strong, so let’s use triangles. We need a wider base. What about bracing? It was a good discussion, and I wasn’t really part of it. I was looking at the clock. Then I was looking at the other tables. Time was nearly gone, and nobody in the room had built anything.</p><p>So, right at the last moment, I said it. “Let’s just teepee it. Basic teepee, marshmallow on top. Now.”</p><p>That landed about as well as you’d expect with a table of people who’d just spent most of the session designing something better. We could do better, they said. And they were right. But I didn’t think we needed better. I thought we needed standing, because I didn’t believe anyone else was going to get there.</p><p>They went with it, a bit reluctantly. We threw the teepee together, put the marshmallow on, and stepped back from something nobody was proud of. Short, stubby, obviously beneath us.</p><p>The team I’d have bet on, with some very bright people in it, didn’t get anything standing. Another team built something tall and ambitious, and when they let go, it folded under the marshmallow.</p><p>We won with the shortest tower that counted.</p><h3><strong>A save, not a strategy</strong></h3><p>For years I told that as a “just ship it” story. It isn’t really. We didn’t win because we had a good approach. We won because we panicked in the right direction a few minutes before everyone else. We’d spent the session exactly like every other table and just happened to bail out first.</p><p>The lesson only landed properly later, when I found out the marshmallow challenge has been run thousands of times. Tom Wujec, who popularised it, found recent kindergarten graduates routinely beat business school graduates. Engineers and architects do best of all, which makes my room of engineers with nothing standing all the more telling. Knowing how to build it wasn’t the problem. His talk is worth seven minutes of your time:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FH0_yKBitO8M%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DH0_yKBitO8M&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2FH0_yKBitO8M%2Fhqdefault.jpg&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/d5711847e5778ca00867ae96682c4842/href">https://medium.com/media/d5711847e5778ca00867ae96682c4842/href</a></iframe><p>The difference is prototyping. Adults treat the challenge as one attempt. They plan, build, and add the marshmallow at the very end, when they discover it’s heavier than they assumed and the whole thing buckles. Kids build prototype after prototype. They stick the marshmallow on a few pieces of spaghetti, see if it stands, and build underneath it. When something wobbles, they fix it and go again. Every version is a working tower. They just keep making it taller.</p><p>The part I find most useful is what counts as a prototype. A plan isn’t one. A plan can’t tell you the marshmallow is heavier than you think; only a tower can. And a tower without the marshmallow on it isn’t much of a prototype either, because it tells you nothing about the thing you’re actually being judged on. Iteration only works if every iteration carries the marshmallow.</p><p>That’s what our teepee was: a first prototype, built so late we never got to iterate on it. If we’d built it in the first five minutes, we’d have spent the rest of the session making it taller, with a winning entry in hand the whole way.</p><p>The marshmallow goes on first, and it never comes off.</p><h3><strong>Doing it properly</strong></h3><p>That year we got a chance to find out whether the idea held up beyond spaghetti. We had a semester-long team project, an online multiplayer game, and we went into it with one principle: keep the marshmallow on top.</p><p>In software, the marshmallow is the thing you’re actually judged on: working code, in front of a user, doing the job. Everything else is spaghetti.</p><p>Other teams went straight for the exciting parts. Cool-looking features, the latest tech, the things that make a good demo. We worked from production backwards. We started with the output, what the thing actually had to deliver, and then built the process to get it there: CI/CD, unit tests, a path from a change on someone’s machine to something running. Unglamorous work, and a lot of it before anything interesting existed.</p><p>For the first two weeks, we looked behind. Other teams had screens to show. We had a pipeline.</p><p>But the pipeline was what made iteration cheap. Every change we made went all the way through, tested and running, so every change was a real prototype with the marshmallow on. We could try something, see if it stood, and go again, many times a week. The other teams were building impressive towers and adding the marshmallow at the end.</p><p>Then they started to struggle. Integration got painful, things broke in ways nobody could trace, and every new feature cost more than the last. We didn’t have that problem, because we’d never been more than one step away from something that worked. We caught up, then quickly overtook.</p><p>We won best project that year.</p><h3><strong>I’ve been the one arguing about triangles</strong></h3><p>It would be a nicer story if the lesson were “be like me.” It isn’t. I’ve been lucky enough to watch the clock a couple of times. Most of my career, I’ve been one of the people at the table debating the triangles.</p><p>Software engineers love the triangles. The architecture, the patterns, the right way to structure the thing so it’ll scale for users we don’t have yet. I find that work genuinely satisfying, and that’s exactly the problem. The detail is comfortable. It feels like progress. Every individual decision is defensible, which makes it very hard to notice you’ve been building foundations for weeks without anything carrying weight.</p><p>None of it matters if the marshmallow isn’t on top.</p><h3><strong>Someone has to watch the clock</strong></h3><p>What a team needs, more often than a better idea, is someone looking up from the table. Someone checking what’s happening around them and asking the unpopular question: what do we actually need to walk out with?</p><p>That’s not a knock on the people at the table. In Wujec’s data, CEOs on their own do only moderately well, but pair them with their executive assistants and they do much better. Someone is watching the process, not just the problem. It’s rarely about how smart the room is. It’s whether anyone’s looking at the marshmallow.</p><h3><strong>Why it matters more now</strong></h3><p>With AI coding tools, building is cheap. A working first version that used to take a sprint can take an afternoon, and a pipeline that took us weeks to set up can be scaffolded in an hour. Prototypes have never been cheaper, which means iterations have never been cheaper either.</p><p>The debating hasn’t sped up at all.</p><p>The catch is that cheap spaghetti cuts both ways. AI will happily build you a tall, impressive tower with no marshmallow on it: features, screens, scaffolding, all looking like progress. It’s never been easier to be the team that folds when they finally let go.</p><p>That makes starting anywhere other than the marshmallow more expensive than ever. When the spaghetti is free, there’s very little excuse for a room where nobody has built anything yet.</p><p>So get something standing. Put the marshmallow on it. Then, and only then, make it taller.</p><p>And whatever you do, don’t take the marshmallow off.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ee7b5252a909" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Let the Kernel Say No]]></title>
            <link>https://medium.com/@geeknz/let-the-kernel-say-no-51236a921113?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/51236a921113</guid>
            <category><![CDATA[claude-code]]></category>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[security]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[developer-tools]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Wed, 02 Sep 2026 20:24:04 GMT</pubDate>
            <atom:updated>2026-09-02T20:31:01.599Z</atom:updated>
            <content:encoded><![CDATA[<h4>asb — a Linux port of Brad Milne&#39;s second-user-account sandbox for AI coding agents</h4><p><em>This story was written with the assistance of an AI writing program — the ideas, observations, and opinions are mine; Claude helped me shape them into prose.</em></p><p><em>Not a Medium member? </em><a href="https://www.colin.geek.nz/let-the-kernel-say-no-51236a921113?source=friends_link&amp;sk=b9d7dd89bed53dfcd9eaa2e8058b66c7"><em>Read this story for free here</em></a><em>.</em></p><p>Your AI coding agent doesn’t need to be malicious to read ~/.aws. It just needs to be thorough.</p><figure><img alt="A dark room. A mass of glowing blue fibre-optic cables and server racks fills the left side, tendrils fanning out to the right and thinning to nothing before they reach a plain closed door in a bare wall, lit by a single warm ceiling light." src="https://cdn-images-1.medium.com/max/1024/1*kjGYGIgODN8_FOXCuZsl0g.png" /><figcaption>Everything the agent can reach, and one door it can’t. Image generated with ChatGPT.</figcaption></figure><p><a href="https://medium.com/@bradnz">Brad Milne</a> made that case properly in a write-up covering the threat model, why the in-app sandbox is one layer rather than the whole answer, and what a locked-down account actually buys you — and then again in person at AI Tinkerers Wellington a few nights ago, which is where I ran into it. His is the piece to read first, and I’m not going to re-run his argument here:</p><p><a href="https://medium.com/@bradnz/sandboxing-claude-code-on-macos-with-a-second-user-account-a16070e1663e"><strong>Sandboxing Claude Code on macOS with a Second User Account</strong></a></p><p>Go for the terminal transcripts if nothing else. There’s one where an agent switches off its own sandbox, mid-task, without asking, to grep an entire home directory — and hits a wall of Permission denied anyway. As Brad put it on the night, it&#39;s &quot;a 1970s solution to a 2020s problem&quot;.</p><p>His script is macOS. This is the Linux port: <a href="https://github.com/geeknz/asb"><strong>asb</strong></a>.</p><h3>Why it isn’t a Claude Code wrapper</h3><p>I wasn’t only thinking about coding agents when I built it. Every tool that runs as me has my blast radius — the agent, the MCP server it pulls in, the install script I piped into bash last week, the extension that auto-updated overnight while I slept. Agents made that visible; they didn’t make it new. And working it out tool by tool doesn’t scale: you’d be reading someone’s security model every time you tried something, and they ship faster than anyone can keep up with. So the default I’ve landed on is that I don’t trust any of them, and an OS account doesn’t care what’s running inside it. asb takes a list of agents rather than hardcoding one.</p><p>The other constraint was that it had to cost nothing per session. A VM or a spare machine gives you more isolation than an account does, but it also pushes the work away from you — files to sync, a remote editor, a second toolchain to keep in step, dotfiles that drift. That’s a tax you pay every session, and the ones you pay every session are the ones you quietly stop paying. Same machine, same working tree, same terminal, or I’d have abandoned it inside a fortnight.</p><h3>The model</h3><p>Three moving parts.</p><p><strong>An account.</strong> A standard non-admin user with a 700 home of its own and no sudo rights. Its home isn&#39;t empty for long — the agent CLI gets installed into it, along with whatever config and cache it accumulates — but it isn&#39;t yours, and it can&#39;t reach yours. asb lockdown takes group and other access off your home directory once, and after that ~/.aws, ~/.ssh, ~/.config/gh, ~/.netrc, your browser profile and that half-forgotten token from 2019 are simply unreadable to it.</p><p><strong>Grants.</strong> POSIX ACLs, per project, applied with setfacl — where Brad&#39;s version uses chmod +a with inheritance flags, Linux uses <em>default</em> ACLs on directories to do the inheriting, so files the agent creates land readable to both of you rather than owned-and-inaccessible. Parent directories get traverse-only (--x): the sandbox can walk down to ~/code/my-project without being able to list what else is in ~/code.</p><p><strong>A door.</strong> One scoped sudoers rule — you ALL=(sandbox) NOPASSWD: ALL — so you can step into the account without a password prompt every time. It runs one way, and it gives the sandbox nothing.</p><h3>Getting it running</h3><p>One bash script, MIT licensed, packaged as .deb, .rpm, a tarball and a PKGBUILD.</p><pre># One-off: stop other accounts reading your home directory<br>asb lockdown<br><br># Create a sandbox and make it the default.<br># It offers to install the agent CLI inside the sandbox for you.<br>asb create claude-x --agent claude<br><br># Grant it a project and launch<br>cd ~/code/my-project<br>asb add<br>asb launch</pre><h3>Day to day</h3><pre>asb launch     # default sandbox, its default agent, right here<br>asb list       # every sandbox and what it can touch<br>asb remove     # revoke the current project<br>asb doctor     # check ACLs, sudoers, agents<br>asb recover    # re-apply every grant from config</pre><p>launch refuses to run unless the current directory has been granted, which is a decent guard against starting an agent somewhere it has no business being.</p><p>recover exists because ACLs are fragile in transit. rsync without -A, tar without --acls, and most sync and backup tools drop them silently — grants vanish with no error or warning, so there&#39;s a command that puts them all back in the config.</p><h3>Agents</h3><p>Agents live in a config file, one per line, and whether a given agent runs with its own permission prompts disabled is a per-agent choice:</p><pre>claude=claude --dangerously-skip-permissions<br>claude-safe=claude<br>hermes=hermes --yolo</pre><p>That first line is the actual payoff. Constant approval prompts defeat the point of an autonomous agent, and approval fatigue means you stop reading them somewhere around the fiftieth one. Turning them off is comfortable <em>because</em> the real boundary isn’t the prompt. Your risk tolerance is yours to set — the flag is named as such for a reason — but the account holds either way.</p><h3>What it isn’t</h3><p>This is a filesystem boundary, and only that. Being honest about the edges matters more than the pitch:</p><ul><li><strong>No network isolation.</strong> The sandbox has full outbound access. It can send anything it can read wherever it likes, and fetch and run arbitrary code. If exfiltration is part of your threat model, add a firewall.</li><li><strong>No protection for secrets inside a granted project.</strong> A grant is read/write over the whole tree — .env, tokens in .git/config, service keys and all. Keep them out of it.</li><li><strong>A grant is write access.</strong> The agent can delete files, rewrite history, and push with whatever credentials it finds in there. Grant work that’s committed and pushed somewhere you can recover it from.</li><li><strong>It’s an account, not a container.</strong> It can still read anything world-readable and see the process list.</li><li><strong>One sandbox is one trust domain.</strong> Every session in a sandbox shares its grants — a project granted to claude-x is readable by anything you ever run as claude-x. Use separate sandboxes for work you want kept apart.</li><li><strong>It won’t stop a real exploit.</strong> Local privilege escalation defeats any account boundary. Genuinely hostile code belongs in a VM.</li></ul><p>There’s also a blunt instrument in here. asb lockdown runs a recursive chmod -R go-rwx over your home directory and appends umask 077 to your shell startup files; asb add runs the same chmod over the project. Neither remembers what was there before, so if something else on the machine reads your home — a web server, a shared group directory, a backup agent running as another account — that will break it. And asb delete --purge is userdel -r: the sandbox&#39;s home, and anything the agent left in it, is gone. Read the script before you run it.</p><p>Some of that is inherent to the approach and some of it is just where I stopped. Please pull the script apart and improve it — and if you do, send it back my way: a pull request, a fork I can pull from, or just a message telling me what you changed. Selfishly, I’d like the benefit of it too, especially from distros and setups I haven’t tested on.</p><h3>The boring part is the good part</h3><p>What I like about this isn’t that it’s clever. It’s that it’s boring. There’s no agent-aware configuration to get subtly wrong, no allowlist to misparse, no prompt to approve on autopilot at 4 pm on a Friday. The agent gets a real account with real limits, and the limits are enforced by the same machinery that’s been separating users since before either of us was writing code.</p><p>The agent can disable every guardrail it controls. It can’t disable the kernel.</p><p><em>Credit where it’s due: the approach, the threat model and the original macOS implementation are all </em><a href="https://medium.com/@bradnz/sandboxing-claude-code-on-macos-with-a-second-user-account-a16070e1663e"><em>Brad Milne’s</em></a><em>. </em><em>asb is a Linux port of his idea, and his article is the one to read first.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=51236a921113" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[I Didn’t Buy the KVM]]></title>
            <link>https://medium.com/@geeknz/i-didnt-buy-the-kvm-f74fe2fbb235?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/f74fe2fbb235</guid>
            <category><![CDATA[wayland]]></category>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[homelab]]></category>
            <category><![CDATA[multiple-monitors]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Mon, 31 Aug 2026 20:37:04 GMT</pubDate>
            <atom:updated>2026-08-31T20:37:51.197Z</atom:updated>
            <content:encoded><![CDATA[<h4>Two laptops. Three monitors. One keyboard I actually like.</h4><p><em>This story was written with the assistance of an AI writing program — the ideas, observations, and opinions are mine; Claude helped me shape them into prose.</em></p><p><em>Not a Medium member? </em><a href="https://www.colin.geek.nz/i-didnt-buy-the-kvm-f74fe2fbb235?source=friends_link&amp;sk=eb1cd85802c99d01549fe16a512f5c1e"><em>Read this story for free here</em></a><em>.</em></p><p>For a long time, my solution was one dock and one cable: work laptop in it during the day, personal laptop in it at night, swapping the USB-C connector between the two. It works. It’s also annoying enough that I eventually did something about it.</p><p>So I went looking for a KVM, like you do. And then I looked at the price.</p><h3>The price wall</h3><p>A basic two-port KVM that drives a single monitor is around $110 in New Zealand. Fine. But I don’t have a single monitor; I have three, and that turns out to be the wall.</p><p>Triple-head KVMs do exist. StarTech makes one — the SV231TDPU34K, two ports, three DisplayPort outputs — and, at the time of writing, PB Tech wants $980 for it, with no stock at any branch in the country. That’s the New Zealand tax as much as anything: a small market means niche gear gets brought in on request, if at all. My screens are only 1080p, so bandwidth isn’t my problem. The problem is that switching three displays at once is specialist kit at a specialist price, whatever resolution you’re running.</p><p>A grand is a lot of money to solve a problem that mildly annoys me twice a day. So I didn’t.</p><p>What I bought instead was a UGREEN UG-30768. It’s a plastic box with a button. Two computers in, four USB devices out, about $64. My keyboard, mouse and webcam plug into it, and the button decides which laptop owns them.</p><p>Here’s what amused me: PB Tech files that plastic box in the same category as the $980 StarTech. Same shelf, fifteen times cheaper. The difference is entirely that I stopped trying to switch the video.</p><h3>You don’t need to switch video</h3><p>I’d had this idea sitting around for a while. The $980 was what finally made me try it.</p><p>A KVM switches video because that’s what the acronym promises. But my monitors already have two inputs each and a perfectly good input switch built into them. It’s in the on-screen display, behind four presses of a squishy button under the bezel, which is why nobody thinks of it as a switch.</p><p>That switch is addressable. It’s been addressable since the 90s. DDC/CI — Display Data Channel Command Interface — is an I2C bus riding along the spare pins of your video cable, and VESA’s MCCS spec defines a control code, 0x60, for input source. Your monitor has always been able to change its own input on command. It just never shipped with a keyboard shortcut.</p><p>So that’s what I built. A hotkey on each machine that tells all three monitors “you belong to me now”, and then a thumb on the UGREEN button to bring the keyboard, mouse and webcam across. Video path stays direct: laptop, dock, cable, panel. Nothing in the middle to degrade.</p><figure><img alt="Diagram of a two-computer, three-monitor desk setup. Laptop 1 (Windows) connects to Dock 1, and Laptop 2 (Linux) connects to Dock 2. Each dock runs its own video cable to all three monitors: Dock 1 uses DisplayPort to Monitor 1 and HDMI to Monitors 2 and 3, while Dock 2 uses the opposite — HDMI to Monitor 1 and DisplayPort to Monitors 2 and 3. Separately, a keyboard, mouse and webcam plug into a single USB switch, which connects by USB to both docks. Video paths run directly from dock to monitor" src="https://cdn-images-1.medium.com/max/1024/1*MFZRqMgy42J-LpOfJtCvdw.png" /><figcaption>Both machines stay plugged into all three monitors. The only shared box on the desk is a $64 USB switch, and nothing but the keyboard, mouse and webcam goes through it.</figcaption></figure><p>One thing that diagram makes obvious: the wiring isn’t symmetrical. Monitor 1 takes DisplayPort from dock 1 and HDMI from dock 2. Monitors 2 and 3 are the other way round. That’s just how the cables reached, and it’s fine — it’s also the reason everything below toggles between DP and HDMI rather than setting one fixed input value everywhere. There is no single “switch to input 15” that would be correct on all three panels.</p><h3>The Linux side: ddcutil</h3><p>On Ubuntu, ddcutil does the talking. You need the I2C dev interface loaded and yourself in the right group, or every command comes back with a permissions error that looks like a hardware fault:</p><pre>sudo apt install ddcutil<br>echo i2c-dev | sudo tee /etc/modules-load.d/i2c-dev.conf<br>sudo modprobe i2c-dev<br>sudo usermod -aG i2c $USER   # log out and back in</pre><p>The modules-load.d line matters. modprobe alone gets you working until the next reboot, at which point your hotkeys silently do nothing and you spend ten minutes suspecting the monitors.</p><p>Then find out what you’ve got:</p><pre>ddcutil detect</pre><pre>Invalid display<br>   I2C bus:  /dev/i2c-11<br>   DRM_connector:           card1-eDP-1<br>   EDID synopsis:<br>      Mfg id:               &lt;PANEL-VENDOR&gt;<br>      Product code:         16749  (0x416d)<br>   This is a laptop display.  Laptop displays do not support DDC/CI.<br>Display 1<br>   I2C bus:  /dev/i2c-13<br>   DRM_connector:           card1-DP-2<br>   EDID synopsis:<br>      Mfg id:               &lt;VENDOR&gt;<br>      Model:                &lt;MODEL&gt;<br>      Product code:         14597  (0x3905)<br>      Serial number:        SERIAL-CENTRE<br>   VCP version:         2.2<br>Display 2<br>   I2C bus:  /dev/i2c-17<br>   DRM_connector:           card1-DP-5<br>   EDID synopsis:<br>      Mfg id:               &lt;VENDOR&gt;<br>      Model:                &lt;MODEL&gt;<br>      Product code:         14597  (0x3905)<br>      Serial number:        SERIAL-RIGHT<br>   VCP version:         2.2<br>Display 3<br>   I2C bus:  /dev/i2c-18<br>   DRM_connector:           card1-DP-7<br>   EDID synopsis:<br>      Mfg id:               &lt;VENDOR&gt;<br>      Model:                &lt;MODEL&gt;<br>      Product code:         14598  (0x3906)<br>      Serial number:        SERIAL-LEFT<br>   VCP version:         2.2</pre><p>Three things in there are worth pausing on.</p><p>The laptop’s own panel comes back as Invalid display, because eDP has no DDC/CI. That&#39;s not a fault, it&#39;s just a boundary: everything in this article applies to external monitors only. If you want the laptop&#39;s built-in screen to join in, that&#39;s a compositor problem, not a DDC one.</p><p>All three externals report the same model name. Address them by model and you’ll be talking to whichever one ddcutil found first. Serial is the only thing that distinguishes them — which is why the config keys off serials, and why I gave up on anything more readable.</p><p>And ddcutil’s own numbering is not the same as my physical layout. Display 1 is my centre screen, Display 2 is the right, Display 3 is the left. The I2C buses aren’t contiguous either — 13, 17, 18. None of that is stable enough to script against.</p><p>Next, ask each monitor what input source it will accept. The spec standardises the <em>code</em> at 0x60, but the values are up to the vendor. This loop pulls the serials straight out of detect and asks each panel in turn:</p><pre>ddcutil detect | awk &#39;/^ +Serial number:/ &amp;&amp; NF==3 {print $3}&#39; | while read -r sn; do<br>  printf &#39;\n=== %s ===\n&#39; &quot;$sn&quot;<br>  ddcutil --sn &quot;$sn&quot; capabilities | grep -A5 &#39;Feature: 60&#39;<br>done</pre><pre>=== SERIAL-CENTRE ===<br>Feature: 60 (Input Source)<br>   Values:<br>      0f: DisplayPort-1<br>      11: HDMI-1</pre><p>All three came back identical, which is the answer I was hoping for: 0x0f is DisplayPort, 0x11 is HDMI, everywhere. That&#39;s what makes a simple two-value toggle viable. The asymmetric cabling stops mattering, because I&#39;m never asking a panel to go to a <em>particular</em> machine — I&#39;m asking it to go to its <em>other</em> input, and both panels agree on what those are called.</p><p>Run that before you write any config. Vendors are free to number inputs however they like, and if your three disagree you’ll want to know now rather than halfway through debugging a bind.</p><p>While you’re in there, the full dump is worth a look:</p><pre>ddcutil --sn SERIAL-LEFT capabilities | head -40</pre><p>Mine reports brightness at 0x10, contrast at 0x12, and power at 0xD6 with five valid values. It also reports a wall of vendor-specific features that ddcutil can&#39;t parse and prints noisy errors about. Ignore those — the standard codes are what matter, and they&#39;re all there.</p><h3>Three monitors, one hotkey</h3><p>Generating the binds in a loop beats hand-writing a dozen near-identical lines, so the whole thing is about sixty lines:</p><pre>-- Requires: ddcutil, usermod -aG i2c $USER, i2c-dev at boot<br>--<br>-- SUPER + CTRL + KP 1/2/3   toggle that monitor&#39;s input between DP and HDMI<br>-- SUPER + CTRL + KP 0       toggle all three inputs<br>-- SUPER + ALT  + KP 1/2/3   toggle that monitor between on and standby<br>-- SUPER + ALT  + KP 0       toggle all three between on and standby</pre><pre>local monitorSerials = {<br>    [1] = &quot;SERIAL-LEFT&quot;,<br>    [2] = &quot;SERIAL-CENTRE&quot;,<br>    [3] = &quot;SERIAL-RIGHT&quot;,<br>}</pre><pre>-- Numpad sends different keysyms with Num Lock off, so bind both forms<br>local numpadKeys = {<br>    [1] = { &quot;KP_1&quot;, &quot;KP_End&quot;  },<br>    [2] = { &quot;KP_2&quot;, &quot;KP_Down&quot; },<br>    [3] = { &quot;KP_3&quot;, &quot;KP_Next&quot; },<br>}</pre><pre>-- Generic VCP toggle: reads current value, flips to the other one.<br>-- Exits quietly if the read fails (monitor is on the other machine).<br>local function ddcToggle(serial, vcp, valueA, valueB)<br>    return string.format(<br>        &#39;v=$(ddcutil --sn %s getvcp %s --terse 2&gt;/dev/null | awk &quot;{print \\$NF}&quot;); &#39; ..<br>        &#39;v=${v##*x}; &#39; ..<br>        &#39;if [ -z &quot;$v&quot; ]; then exit 0; &#39; ..<br>        &#39;elif [ &quot;$v&quot; = &quot;%s&quot; ]; then n=0x%s; else n=0x%s; fi; &#39; ..<br>        &#39;ddcutil --sn %s --noverify setvcp %s $n 2&gt;/dev/null&#39;,<br>        serial, vcp, valueA, valueB, valueA, serial, vcp)<br>end</pre><pre>-- Input source: 0x0f = DisplayPort, 0x11 = HDMI<br>local function inputToggle(serial)  return ddcToggle(serial, &quot;60&quot;, &quot;0f&quot;, &quot;11&quot;) end</pre><pre>-- Power: 0x01 = on, 0x04 = standby<br>local function powerToggle(serial)  return ddcToggle(serial, &quot;D6&quot;, &quot;01&quot;, &quot;04&quot;) end</pre><pre>for i, serial in ipairs(monitorSerials) do<br>    for _, key in ipairs(numpadKeys[i]) do<br>        hl.bind(mainMod .. &quot; + CTRL + &quot; .. key, hl.dsp.exec_cmd(inputToggle(serial)))<br>        hl.bind(mainMod .. &quot; + ALT + &quot;  .. key, hl.dsp.exec_cmd(powerToggle(serial)))<br>    end<br>end</pre><pre>-- All three at once, fired in parallel so they land together<br>local function toggleAll(fn)<br>    local parts = {}<br>    for _, serial in ipairs(monitorSerials) do<br>        table.insert(parts, &quot;{ &quot; .. fn(serial) .. &quot;; } &amp;&quot;)<br>    end<br>    return table.concat(parts, &quot; &quot;) .. &quot; wait&quot;<br>end</pre><pre>for _, key in ipairs({ &quot;KP_0&quot;, &quot;KP_Insert&quot; }) do<br>    hl.bind(mainMod .. &quot; + CTRL + &quot; .. key, hl.dsp.exec_cmd(toggleAll(inputToggle)))<br>    hl.bind(mainMod .. &quot; + ALT + &quot;  .. key, hl.dsp.exec_cmd(toggleAll(powerToggle)))<br>end</pre><p>There are four decisions in there worth pulling out.</p><p><strong>One: read before you write, and fail quietly.</strong> The toggle asks the panel what input it’s on, then flips to the other one. If the read comes back empty, it exits rather than guessing — a monitor that’s asleep, unplugged or otherwise not answering shouldn’t get a blind setvcp fired at it.</p><p>I’d braced for a worse problem here. Plenty of monitors only answer DDC on the input they’re currently displaying, which would mean you can only ever shove a screen <em>away</em> from the machine you’re sitting at, never summon one back. Mine answers on both inputs regardless of which is active, so either machine can pull a panel across at any time. I don’t think that’s something to count on — check yours early, because the answer decides whether your hotkeys work in both directions or only one.</p><p><strong>Two: address monitors by serial, not display number.</strong> --display 1 and I2C bus numbers renumber themselves across reboots and hotplug events. Serial is stable. Same reason the Windows side uses serials.</p><p><strong>Three: </strong><strong>--noverify and the </strong><strong>&amp;.</strong> By default setvcp reads the value back to confirm, which is another round trip on a slow bus. Skip the verification, fire all three in parallel, wait at the end. The difference between doing this serially and in parallel is the difference between a switch that feels laggy and one that doesn&#39;t.</p><p><strong>Four: bind both numpad keysyms.</strong> KP_1 with Num Lock on, KP_End with it off. Bind one and your hotkeys will work perfectly until the morning you knock Num Lock and realise the whole setup is broken.</p><h3>A free monitor power switch</h3><p>The bonus in that config is 0xD6. Same mechanism, different VCP code: 0x01 is on, 0x04 is standby. SUPER+ALT and a numpad key puts a panel to sleep without touching a physical button or waiting on an idle timer.</p><p>I didn’t set out to build this. It fell out of having the plumbing already there, which is the nice thing about using the monitor’s own control protocol instead of a box in the middle — everything else the panel exposes is sitting right there too. Brightness on 0x10, contrast on 0x12, colour temperature, gain per channel. Once you can speak to the panel, you can speak to all of it, and none of it needed anything I hadn&#39;t already installed.</p><h3>The Windows side: ControlMyMonitor</h3><p>Windows has no ddcutil, but NirSoft’s ControlMyMonitor does the same job — and it’s got a nicer verb for this than /SetValue. /SwitchValue takes a VCP code and a list of values, and moves to the next one in the list each time it runs:</p><pre>controlmymonitor.exe /SwitchValue &quot;SERIAL-LEFT&quot; 60 15 17</pre><p>That’s one panel flipping between input 15 and input 17. Note the numbers are decimal here, where ddcutil talks in hex — 0x0F is 15 and 0x11 is 17, which are DisplayPort and HDMI on my panels. Spending twenty minutes convinced my monitors were broken because I&#39;d typed 0x0F into a tool that wanted 15 is a rite of passage I&#39;d rather you skipped.</p><p>You’ll also notice it addresses the monitor by serial number, same as on the Linux side. \\.\DISPLAY1\Monitor0 works, but it means whatever Windows currently thinks display 1 is, which is not a stable promise across dock reconnects.</p><p>Rather than a batch file, I made Windows shortcuts — one per monitor, plus one that does all three — and used the Shortcut key field in the properties dialog to bind them to Ctrl+Num1 through Ctrl+Num3, with Ctrl+Num0 for the lot. It’s a genuinely underrated corner of Windows: you get global hotkeys with no daemon and no AutoHotkey install.</p><p>The per-monitor bindings turn out to be more useful than expected. Sometimes I don’t want all three to move — leaving the right-hand panel on the work laptop while the other two follow me across is a good way to keep an eye on something without switching context entirely.</p><h3>Two actions, in the sensible order</h3><p>Hotkey first, button second.</p><p>Both orders technically work, since either machine can pull the panels across. But the keyboard is what runs the hotkey, and the button is what takes the keyboard away — so press the button first, and you’re typing blind on a machine whose screens haven’t arrived yet. Move the screens, then move the keyboard, and you can always see what you’re doing.</p><p>It’s two actions rather than one, and that’s the honest cost of this approach. A real KVM does both on a single press. The muscle memory took about a day.</p><h3>The thing I expected to go wrong</h3><p>I assumed switching inputs would tear down the displays and bring them back up, and that both machines would spend the day re-enumerating monitors and shuffling windows around. That’s the whole reason the old dock-swapping approach was tiresome.</p><p>It doesn’t happen. When a panel switches to the other input, this machine keeps the link and continues to believe three monitors are attached. Windows stay exactly where I left them on both machines. I switch away mid-afternoon, come back two hours later, and everything is where it was.</p><p>I don’t know that this holds on every panel — it’s the kind of behaviour that varies — but it’s the single best thing about the setup, and it falls out of never having interrupted the video path in the first place.</p><p>Kanshi still earns its place, just for a different job: layout and docking. It keys off the output description rather than the connector name, so it doesn’t matter which DP port a panel lands on:</p><pre>profile home {<br>  output &quot;&lt;VENDOR&gt; &lt;MODEL&gt; SERIAL-LEFT&quot;   mode 1920x1080@60 position -1920,0<br>  output &quot;&lt;VENDOR&gt; &lt;MODEL&gt; SERIAL-CENTRE&quot; mode 1920x1080@60 position 0,0<br>  output &quot;&lt;VENDOR&gt; &lt;MODEL&gt; SERIAL-RIGHT&quot;  mode 1920x1080@60 position 1920,0<br>  output eDP-1 mode 2880x1800@120 disable<br>}</pre><pre>profile undocked {<br>  output eDP-1 mode 2880x1800@120 scale 2.0 position 0,0<br>}</pre><p>Same principle as the ddcutil config, for the same reason: three monitors reporting the same model string can only be told apart by serial. The home profile switches the laptop&#39;s own panel off, since it&#39;s redundant once three externals are live, and undocked brings it back scaled when I walk away with the machine.</p><h3>Being fair to the cheap box</h3><p>The UGREEN is not a premium product, and I’m not going to pretend otherwise.</p><p>The manual notes — and reviewers confirm — that 2.4GHz wireless dongles get comprehensively stomped by USB 3.0 hubs sitting next to them. If your keyboard or mouse uses a small USB receiver, plan for that. A wired keyboard or Bluetooth sidesteps it entirely.</p><p>There are also reports of intermittent disconnects, particularly early after power-up, and of power-hungry peripherals needing the external 5V supply that isn’t in the box. Mine has a webcam, as well as a keyboard and mouse, which is more of a draw than the “share your printer” use case this thing was designed for. It’s also worth noting that the switch feeds the two docks rather than the laptops directly, so everything is riding a dock’s USB hub on top of the switch — one more link in a chain that already has a cheap plastic box in it. It’s been fine, but I’d call that luck rather than engineering.</p><p>Cheap is doing some work in this story. It’s just doing about $900 less work than the alternative.</p><h3>When you should just buy the KVM</h3><p>Be honest about the cases where this doesn’t hold up.</p><p>If you switch machines constantly — dozens of times a day rather than twice — the couple of seconds and the two-step ritual will grind you down. If your monitors have DDC/CI disabled in their firmware, or you’re running through an MST hub that eats the I2C channel, this approach is dead on arrival, and no amount of scripting will save it. And if the whole point is that someone non-technical uses the desk, a physical box with one button beats a bash script every single time.</p><h3>Was it worth the time?</h3><p>Probably not, if you do the honest sum.</p><p>The box was $64. Against a $980 alternative that I’d have had to order in and wait for, that’s a clear win on paper. But then there’s the evening spent reading the MCCS spec, the hour working out why the hotkeys stopped after a reboot, the fiddling with numpad keysyms, and the ordinary drift of a thing you built yourself and now maintain. Price it at any sensible hourly rate, and the $980 box starts to look reasonable.</p><p>I’m not going to pretend that calculation motivated me. I did it because I was fairly sure it would work and I wanted to know if I was right. Stubbornness has been responsible for most of the useful things on my desk and nearly all of the useless ones, and the honest reason this exists is that it bothered me that the capability was already sitting there, in a spec from the nineties, on hardware I already owned, unused because nobody had bothered to put a keyboard shortcut on it.</p><p>The thing I didn’t expect was how little I think about it now. Two keystrokes and a button, several times a day, for months. The best outcome for a project like this is for it to disappear, and this one has.</p><p>So: not worth it, by the numbers. Worth it anyway.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f74fe2fbb235" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Every Number Is Up, and Nothing Is Better]]></title>
            <link>https://medium.com/@geeknz/every-number-is-up-and-nothing-is-better-d07c76a2de58?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d07c76a2de58</guid>
            <category><![CDATA[agile]]></category>
            <category><![CDATA[management]]></category>
            <category><![CDATA[metrics]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[productivity]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Tue, 18 Aug 2026 19:01:01 GMT</pubDate>
            <atom:updated>2026-08-18T19:52:02.980Z</atom:updated>
            <content:encoded><![CDATA[<h4>Goodhart’s Law and why your dashboard is lying even though nobody’s dishonest</h4><p><em>This story was written with the assistance of an AI writing program — the ideas, observations, and opinions are mine; Claude helped me shape them into prose.</em></p><p><em>Not a Medium member? </em><a href="https://www.colin.geek.nz/every-number-is-up-and-nothing-is-better-d07c76a2de58?source=friends_link&amp;sk=97b7f5ccc017f8b65935c72ce957ae48"><em>Read this story for free here.</em></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*UeEgWyUeYzOrQ-Xp0UHZgQ.png" /><figcaption>Every tile green except the one that matters. Image generated by the author using AI.</figcaption></figure><p>Picture the quarterly review. Velocity is up 20% on last year. The support desk closed more tickets than ever, faster than ever. NPS climbed three points. Most of the OKRs are green. Leadership is pleased — this is what a high-performing organisation looks like on paper.</p><p>Now walk the floor and ask people how the work actually feels. Releases seem slower. The same customer problems keep coming back. The support queue feels worse, not better. The engineers are tired and oddly siloed.</p><p>Both things are true at once. The numbers are genuinely up, and nothing is genuinely better.</p><p>If that dashboard looked like good news to you a moment ago — congratulations, you’ve just been Goodharted. So has your leadership team. So, probably, has your entire industry.</p><h3>The law</h3><p>In 1975, Charles Goodhart, an adviser at the Bank of England, made an observation about monetary policy: any statistical regularity tends to collapse once pressure is placed upon it for control purposes. The anthropologist Marilyn Strathern later sharpened it into the version everyone quotes:</p><blockquote>&quot;When a measure becomes a target, it ceases to be a good measure.&quot;</blockquote><p>The social scientist Donald Campbell arrived at the same place independently — Campbell’s Law says that the more any quantitative indicator is used for decision-making, the more it will distort and corrupt the very process it was meant to monitor.</p><p>Note what the law does <em>not</em> say. It doesn’t say people are dishonest. It says the measurement system itself changes the behaviour it was built to observe. The moment you point at a number and say “make this go up,” you’re no longer measuring the work — you’re measuring people’s ability to make the number go up. Those are different skills, and organisations reliably reward the second one.</p><p>This isn’t new. In 1902, the French colonial government in Hanoi had <a href="https://en.wikipedia.org/wiki/Great_Hanoi_Rat_Massacre">a rat problem</a> and offered a bounty per rat killed — proven by handing in the tail. Soon the city was full of tailless rats: catchers snipped the tail and released the rat to go breed more bounty. Some enterprising locals skipped the middleman and started farming rats. The rat population went <em>up</em>.</p><p>The bounty didn’t fail because the Vietnamese were dishonest. It failed because it paid for tails, not for fewer rats. Every metric programme since has been a version of this story.</p><h3>This is not a people problem</h3><p>If you think your organisation is immune because you hire good people, consider who else thought that.</p><p><a href="https://www.consumerfinance.gov/enforcement/actions/wells-fargo-bank-2016/"><strong>Wells Fargo</strong></a> set aggressive cross-selling targets — the internal mantra was “eight is great,” eight products per customer. Branch staff couldn’t legitimately hit the numbers, and their jobs depended on it. So employees opened around 3.5 million accounts customers never asked for. The bank fired 5,300 people, paid an initial $185 million fine with total settlements eventually reaching $3 billion, and then did the genuinely revealing thing: it eliminated product sales goals altogether. You don’t accidentally hire 5,300 fraudsters. You install a target that manufactures fraud, then act surprised.</p><p>And this isn’t only a money problem. When New York State began publishing mortality rates for individual cardiac surgeons — reasonable, humane, transparent — <a href="https://pubmed.ncbi.nlm.nih.gov/10543479/">62% of surgeons</a> later admitted refusing to operate on at least one high-risk patient in the previous year, primarily because of the public reporting. A measure designed to protect patients taught surgeons to avoid the ones who most needed surgery.</p><p><strong>England’s NHS</strong> set a target: 95% of emergency patients seen within four hours. <a href="https://pubmed.ncbi.nlm.nih.gov/16484272/">Breaches fell from 23% to 5%</a> — a triumph, on paper. Behind the paper: ambulances idling in the car park so the clock wouldn’t start, and patients shunted into “assessment units” adjacent to A&amp;E at three hours and fifty-eight minutes so the clock would stop.</p><p>And before we feel smug on this side of the world — New Zealand ran the same experiment with our six-hour “Shorter Stays in ED” target, and <a href="https://www.ijhpm.com/article_3686.html">researchers documented the same gaming here</a>. The neatest evidence is statistical. They took every recorded stay between 360 and 369 minutes — the target was 360 — and checked the final digit. By chance, about one in ten should land on exactly 360. After the target came in, the four hospitals studied ranged from 11%, which is roughly what maths predicts, up to 38%. Somebody was rounding, and the rounding left a fingerprint.</p><p>Banking, surgery, emergency medicine. Different countries, different professions, identical failure. Now let’s talk about your office.</p><h3>Your office</h3><p><strong>Velocity.</strong> Story points and velocity are legitimate tools — <em>for the team</em>. They exist so a team can forecast: we typically deliver about this much per sprint, so this backlog will take roughly this long. Used that way, velocity is honest, because nobody benefits from lying to themselves about their own forecast.</p><p>Then management discovers the number, and asks the fatal question: <em>whose</em> velocity? The moment individual velocity becomes a performance signal, it turns against everything agile was built on. Scrum is a team sport — shared commitment, collective ownership, swarming on problems. Individual velocity tracking makes helping a teammate an act of self-harm: every hour spent unblocking someone else is an hour your number doesn’t move. It makes the high-point tickets worth hoarding and the gnarly, low-point, high-value work worth avoiding. It makes pairing look like half-productivity. You’ve taken a framework whose entire premise is <em>team output</em> and installed an incentive for each member to quietly compete against their own team. The individual is now working against the group — and the group’s numbers will still go up, because inflating estimates is easier than delivering more.</p><p><strong>OKRs.</strong> The whole point of an objective is ambition — the original Google framing treated ~70% attainment as success, because 100% meant you sandbagged. But grade people on their OKRs and everyone becomes a sandbagging expert overnight. Objectives get written backwards from what’s already nearly done. Key results get worded vaguely enough to declare victory either way. The quarterly OKR review becomes a creative writing exercise where everyone scores 0.7 — ambitious enough to look bold, achieved enough to look competent. The instrument for stretching the organisation becomes the instrument for hiding from stretch.</p><p><strong>Performance reviews.</strong> Whatever your review measures is rarely the work — it’s <em>legible</em> work: the visible project, the launch that lands conveniently in review season, the accomplishment with your name clearly on it. So people optimise for legibility. Glue work — mentoring, unblocking, the unglamorous maintenance that keeps everything upright — is invisible to the rubric, so it quietly stops, or survives only on the goodwill of people who then get outpromoted by the credit-claimers. (I wrote about what that does to trust in <a href="https://www.colin.geek.nz/the-mana-ledger-4f18375fb743">The Mana Ledger</a>.)</p><p><strong>Support desk metrics.</strong> There’s a post that does the rounds every few months, from an IT support account, and support veterans don’t find it funny so much as familiar:</p><figure><img alt="Screenshot of a post by IT Unprofessional (@it_unprofession). The poster explains that their bonus depends on how many tickets they close, so when a colleague’s keyboard broke they took the receptionist’s keyboard instead of ordering a replacement — generating a second ticket when the receptionist reported hers missing. They close by noting that more closed tickets mean a bigger bonus, and that the manager’s printer is next." src="https://cdn-images-1.medium.com/max/597/1*GH_BgBL2VS6vCnrr3R8X6Q.png" /><figcaption>Nobody here is lying. — via @it_unprofession</figcaption></figure><p>It’s a joke. It is also an exact description of what a close-rate target pays for. Follow the logic and every step is rational: the fastest way to close a ticket is to move a working keyboard rather than order a new one, and the cheapest source of new tickets is to create the problems yourself. Nothing was repaired. No faulty hardware left the building. The queue is longer than when we started — and the numbers have never looked better.</p><p>Strip out the joke and you have the law working exactly as designed. Close rates measure <em>activity</em>, and manufacturing activity is far easier than resolving problems. The question that actually matters — did the underlying issue go away and stay away? — isn’t on the dashboard, because it’s slower and harder to count. So it doesn’t get managed, and eventually it doesn’t get done.</p><p><strong>NPS.</strong> “You’ll get a survey — anything below a nine counts as a fail for me personally.” If you’ve ever bought a car, you’ve heard some version of it. Tie a bonus to a survey score and teams stop improving the experience and start coaching the respondent. The score goes up; the thing it was invented to detect does not.</p><h3>The doubling-down loop</h3><p>Here’s where it gets properly perverse. Leadership eventually notices the gap between the green dashboard and the grumpy reality. And the instinctive response is: <em>we need better visibility.</em> More metrics. Finer-grained tracking. Dashboards about the dashboards.</p><p>But every new metric is new surface area to game, and — worse — every new metric is a message. This kind of measurement is what you deploy when you don’t trust people, and people can tell. So discretionary effort drops, gaming becomes normalised (“everyone pads their estimates, it’s just how the sizing game works”), and the honest signal gets buried under another layer of performance-of-performance. The historian Jerry Muller calls this <a href="https://www.econtalk.org/jerry-muller-on-the-tyranny-of-metrics/">metric fixation</a>: the more you substitute measurement for judgement, the more judgement atrophies, and the more you need the measurements you can no longer trust.</p><p>The loop is stable, which is why you’ve seen it everywhere you’ve worked.</p><h3>What to do instead</h3><p>Goodhart’s Law doesn’t say “don’t measure.” Measurement is how you learn. It says a measure can be a <em>thermometer</em> or a <em>target</em>, and it can’t be both for long. So:</p><p><strong>Decouple measurement from judgement.</strong> The instant a number feeds a performance review or a bonus, it starts lying. Keep velocity inside the team, for the team’s forecasting — the moment it appears in a manager’s spreadsheet with names attached, it’s dead as information. If a measure must inform judgement, let it prompt a conversation; never settle one.</p><p><strong>Measure outcomes, not activity.</strong> Not tickets closed — reopen rates and repeat contacts. Not story points burned — customer problems that stayed solved. Not survey scores — retention. Outcome measures are slower and messier, which is exactly why they’re harder to game: you can’t manufacture a customer who stops having the problem.</p><p><strong>Pair opposing metrics.</strong> Speed alone gets gamed by cutting quality; quality alone gets gamed by slowing down. Speed <em>and</em> defect rate, close rate <em>and</em> reopens, sales <em>and</em> churn. Gaming one number shows up in its partner. A single metric is a promise to be fooled.</p><p><strong>Rotate and retire.</strong> Every metric decays — the useful signal gets arbitraged away as people learn it. Treat metrics like diagnostics, not commandments: run them, learn from them, and retire them before optimisation sets in. If a number has been a KPI for three years, assume it now measures gaming skill.</p><p><strong>And the uncomfortable one: ask why you wanted the target in the first place.</strong> Targets are, mostly, a substitute for trust. If the only way you can tell whether a team is doing good work is a number on a dashboard, the number isn’t your problem — the distance is. Get closer to the work. Talk to the customers. Watch a release go out. The information you actually want has never been on the dashboard, because the moment you put it there, it stopped being true.</p><p>When a measure becomes a target, it doesn’t just stop measuring the work. It starts managing the people — and it manages them badly, in ways no one chose and everyone can feel.</p><p>Point at the number, and the number stops telling the truth.</p><p><em>What’s the metric your organisation still won’t give up?</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d07c76a2de58" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Mandate Is the Confession]]></title>
            <link>https://medium.com/@geeknz/the-mandate-is-the-confession-1c46d9d0fc06?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/1c46d9d0fc06</guid>
            <category><![CDATA[remote-work]]></category>
            <category><![CDATA[future-of-work]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[engineering-management]]></category>
            <category><![CDATA[management]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Wed, 05 Aug 2026 18:14:37 GMT</pubDate>
            <atom:updated>2026-08-05T18:14:37.135Z</atom:updated>
            <content:encoded><![CDATA[<h4><em>What return-to-office orders actually tell you about a company’s management.</em></h4><p><em>This story was written with the assistance of an AI writing program — the ideas, observations, and opinions are mine; Claude helped me shape them into prose.</em></p><p><em>Not a Medium member? </em><a href="https://www.colin.geek.nz/1c46d9d0fc06?source=friends_link&amp;sk=216ef322533f98e78a31ac61cbc9991e"><em>Read this story for free here.</em></a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*qxaDE2OxPTMG9yiRYE_XOg.jpeg" /><figcaption><em>Photo: Kate Sade / Unsplash</em></figcaption></figure><p>Watch what happens after a bad quarter.</p><p>The share price dips. The board gets restless. And somewhere in the weeks that follow, an email goes out about “collaboration”, “culture”, and “the energy of being together” — and everyone is back in the office three days a week, minimum, badge data monitored.</p><p>Researchers at the University of Pittsburgh <a href="https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4675401">looked at S&amp;P 500 firms</a> that issued return-to-office mandates and found a pattern that should embarrass everyone involved. The mandates tended to <em>follow</em> poor stock performance. And afterwards? No measurable improvement in profitability or firm value. The only thing that moved was employee satisfaction, which fell off a cliff. Their conclusion was blunt: the data is consistent with managers using RTO to reassert control and to blame employees for results the executives couldn’t explain.</p><p>So I’ve come to read every mandate the same way. It isn’t a strategy. It’s a confession. It says: <em>we cannot tell who is contributing, so we are going to measure the one thing we can see — bodies at desks.</em></p><p>The frustrating part is that the evidence on remote work is now very good. It just doesn’t say what either side of the culture war wants it to say.</p><h3>The experiment the managers lost</h3><p>In 2024, Stanford economist <a href="https://www.youtube.com/watch?v=oiUyyZPIHyY">Nick Bloom</a> published <a href="https://www.nature.com/articles/s41586-024-07500-2">the largest randomised controlled trial of hybrid work to date</a> — 1,612 professionals at Trip.com, randomly assigned to five days in the office or three days in, two at home, with outcomes tracked over the following two years.</p><p>Beforehand, the managers predicted hybrid would cut productivity by 2.6%. What actually happened: nothing. No difference in performance reviews. No difference in promotion rates. For the engineers, no difference in code shipped. What <em>did</em> move was retention — resignations fell by a third, with the biggest drops among women, non-managers, and people with long commutes. Trip.com saved millions and rolled hybrid out company-wide.</p><p>And the managers? Having watched it play out, they revised their estimate from −2.6% to +1.0%. They were wrong; the data said so, and to their credit they updated.</p><p>That exchange is the whole argument, compressed. This debate has never really been about evidence. It’s about what managers <em>believe</em> they can see. Microsoft’s <a href="https://www.microsoft.com/en-us/worklab/work-trend-index/hybrid-work-is-just-work">Work Trend Index</a> found 85% of leaders struggle to feel confident their hybrid people are actually working — while 87% of those same workers report being productive. Microsoft called it “productivity paranoia”, and it isn’t free: people respond to suspicion by performing busyness. One <a href="https://workplaceinsight.net/digital-presenteeism-now-a-pervasive-cultural-pressure/">survey of 2,000 knowledge workers</a> put the price of that theatre at an extra 67 minutes online a day — pointless calls, replies timed for visibility, keeping the green dot lit.</p><blockquote>Measure presence, and presence is what you’ll get</blockquote><h3>Where my side oversells it</h3><p>If I stopped there, this would be a comfortable article for people who already agree with me. So here’s the strongest evidence <em>against</em> remote work, because it’s real and it’s specific.</p><p>Natalia Emanuel, <a href="https://www.youtube.com/watch?v=mYDNeK71fhk">Emma Harrington</a> and Amanda Pallais <a href="https://www.nber.org/papers/w31880">studied engineers at a Fortune 500 firm</a> whose campus split teams across two buildings — a natural experiment in proximity. Sitting near teammates increased the feedback engineers received on their code by 18.3%, concentrated where it matters most: on juniors, who went on to write higher-quality code with fewer bugs. When offices closed, that advantage collapsed, hardest for the juniors. The effects were stronger again for women, who both gave and received more mentorship when co-located.</p><p>But here’s the twist nobody quotes: proximity <em>cost</em> output. Senior engineers sitting near their teams wrote less code, because mentoring eats time. When offices closed, senior output went <em>up</em>.</p><p>So the office was never really a productivity tool. It was a training facility we mistook for one. Co-location trades senior output today for junior capability tomorrow — a genuine tension that a blanket policy, in either direction, handles badly. And the stakes are generational: the same researchers found the firm shifted toward hiring older, more experienced engineers while remote. Buying talent instead of building it. Scale that across an economy and you get a cohort who never receive the osmotic learning that made their seniors senior. If you’re a remote-work advocate and that doesn’t make you uncomfortable, you’re not paying attention.</p><h3>Why the mandate still fails</h3><p>Given all that, you might expect me to concede mandates have a point. They don’t — because of who actually leaves when you issue one.</p><p>Researchers at Michigan and Chicago <a href="https://arxiv.org/abs/2405.04352">matched 260 million CVs</a> against employment records at Microsoft, Apple and SpaceX post-mandate. The losses landed at the top of the tenure distribution. At Microsoft, the share of staff in training, entry and manager roles rose by over 4% — the workforce shifted measurably junior — and tenure fell by about two months at the longest-serving end. Apple, mandating one day a week, shifted just under 4%. SpaceX, five days and no exceptions, up to 15%. And Microsoft’s leavers tilted away from startups towards large, direct competitors that hadn’t mandated anything.</p><p>Set that beside the proximity research. The entire evidence-based case for the office is that experienced people transfer capability to developing people. A mandate falls hardest on exactly those experienced people — most options, least need for supervision — and they’re the ones who walk. You end up with your juniors in a building containing fewer people worth learning from. The policy erodes the very mechanism it claims to protect, while telling your most capable people you don’t trust them.</p><p>That’s not culture. That’s a management team confusing the <em>venue</em> of connection with connection itself.</p><h3>And for many teams, the argument was never available</h3><p>There’s a second problem, and it’s the one I keep hitting when people make the mentorship case to me in real life. It assumes the mentor and the mentee are in the same city.</p><p>Plenty of teams aren’t. The best analyst you could find lives four hundred kilometres away. The graduates are in the regional office. The one person who understands the legacy system moved overseas years ago, and you sensibly kept him on rather than lose what he knows. That’s what a lot of organisations look like now — often the deliberate result of a hiring strategy everyone was pleased with at the time.</p><p>For those teams, a mandate cannot deliver the thing it’s for. You can fill three separate buildings and still not put one junior within osmosis range of one senior. You’ve bought the full cost — the attrition, the goodwill, the floor space — for a mechanism geography already took off the table. The only version that works is relocating people, and nobody proposing a mandate is proposing that.</p><p>Then recall the sharpest finding in the proximity study: a ten-minute walk between buildings on the same campus reduced feedback as much as being several states away. And a single distant colleague dragged down the feedback flowing between the teammates who <em>did</em> sit together — while adding a co-located one did nothing of the sort.</p><p>Proximity isn’t a dial you turn up gradually. It’s a threshold you’re either over or under. A partly-distributed team that mandates attendance doesn’t capture a proportion of the benefit — it captures very little, while paying in full. If your team is already distributed, the office question is settled by the shape of your organisation, not by anyone’s preference. The only live question is what you build instead.</p><h3>Scaffolding, not surveillance</h3><p>The honest reading is this: flexible work costs you nothing on output and pays handsomely in retention — <em>provided</em> you replace the accidental scaffolding the office used to provide with deliberate scaffolding. Most organisations never built any, because the building did it for free.</p><p>So the real question isn’t “how many days?” It’s “who here is still building capability, and what replaces osmosis for them?”</p><p>Be precise about what that means, because the obvious answer is to point at the rituals you already run. Standups and retros are genuinely useful — they’re just useful for something else. One coordinates the work, the other improves how the team runs. Neither one is an experienced person looking at your actual work and telling you how to make it better, which is the specific thing the proximity study measured and the specific thing distance took away. It has to be built separately.</p><p>What substitutes, roughly in order of impact: review treated as first-class work rather than a favour — seniors deliberately rostered onto juniors’ work rather than whoever’s free, substantive comments expected, turnaround a team commitment.</p><p>Then design review <em>before</em> implementation, where the expensive learning happens and someone’s mental model gets corrected before they’ve built the wrong thing. Deliberate pairing on hard problems, a couple of hours a week. And group sessions on one interesting change — a tricky refactor, a design somebody got wrong and fixed — run as a teaching slot rather than a queue, so everyone hears the reasoning and not just the verdict.</p><p>Then the half that gets forgotten: the asking. In an office a junior doesn’t really decide to ask a question. They swivel their chair, or catch someone at the kettle, and half the time the senior spots something on their screen and volunteers before being asked at all. Remote, the same question becomes a deliberate act with a social cost — interrupting a named person, in writing, on the record, admitting in front of an audience that you don’t know something.</p><blockquote><em>What proximity removed wasn’t friction. It was inhibition.</em></blockquote><p>That cost falls hardest on the people who most need to ask: the newest, the least confident, and, per the same research, disproportionately women.</p><p>Which points at the fix, and it isn’t the usual one. “Make it safe to ask” is the thing every organisation claims to have achieved and none ever tests. The reliable move is to shift the burden of initiation off the junior and onto the senior. Rostered review does this by design. So does a session that opens with “show me what you’re working on” rather than “any blockers?” — one requires a junior to self-diagnose and self-report a problem, the other doesn’t. And the cheapest lever available is seniors visibly asking obvious questions in the open channel, because a junior watching a principal engineer say “sorry, I don’t actually understand this bit” learns more about what’s permitted here than any values statement will teach them. Keep the real conversation in the open, too: a question answered publicly teaches everyone watching; answered in a DM it teaches one person and disappears.</p><p>One honest cost, since the research names it: all of this takes output from your senior people. Mentoring ate their productivity in the office and it will eat it remotely. If you tell your leads that review and pairing are real work, their own delivery drops. Budget for that openly, or it stops the first time a deadline slips — and you’ll conclude remote mentorship doesn’t work, when what happened is you never paid for it.</p><p>For experienced people, the intervention is simpler: get out of the way. They’re demonstrably more productive with autonomy, and they’re the flight risk when you take it away.</p><p>And for everyone, put what you save on floor space into connection worth showing up for. Not pizza in the breakout room. Proper gatherings in places people actually want to be, with unstructured time built in — the conversations that used to happen at the kitchen bench don’t schedule themselves. Even the Pittsburgh researchers, having demolished the case for mandates, land here: let people who perform well at home carry on, and bring teams together regularly for what genuinely needs it.</p><p>I’ll admit the prescription rests on thinner evidence than the diagnosis. Nobody has run a randomised trial of “deliberate review culture and good gatherings” — though Bloom’s data on <em>organised</em> hybrid, where team days are coordinated rather than left to individual choice, points the same way. But between an approach that targets what the office was actually doing, and one that targets the building, costs you the people you can least afford to lose, and doesn’t move the numbers it was brought in to move — I know which needs to explain itself.</p><p>The mandate is the confession. The alternative — knowing your people well enough to give each of them what they actually need — is just called management. It’s harder than sending an email about culture.</p><p>It was always supposed to be.</p><p><strong>Worth an hour of your time:</strong> Nick Bloom’s <a href="https://www.youtube.com/watch?v=oiUyyZPIHyY">TEDxStanford talk</a> on why working from home works, and Emma Harrington on <a href="https://www.youtube.com/watch?v=mYDNeK71fhk">The Power of Proximity</a> with Steven Davis for Hoover’s <em>Economics, Applied</em> — the best explanation I’ve found of what co-location was actually buying.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1c46d9d0fc06" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Done Being the Product]]></title>
            <link>https://medium.com/@geeknz/done-being-the-product-e1431661ef38?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/e1431661ef38</guid>
            <category><![CDATA[big-tech]]></category>
            <category><![CDATA[digital-minimalism]]></category>
            <category><![CDATA[social-media]]></category>
            <category><![CDATA[privacy]]></category>
            <category><![CDATA[data-privacy]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Sat, 20 Jun 2026 07:27:52 GMT</pubDate>
            <atom:updated>2026-06-21T16:53:35.253Z</atom:updated>
            <content:encoded><![CDATA[<p><em>Leaving social media won’t change anything. I did it anyway.</em></p><p><em>This story was written with the assistance of an AI writing program — the ideas, observations, and opinions are mine; Claude helped me shape them into prose.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7IRotra5DJiauiGw8z6P4Q.jpeg" /><figcaption><em>Photo: Ansgar Scheffold / Unsplash</em></figcaption></figure><p>There used to be a moment, every day, when you were <em>done</em> with Facebook.</p><p>You’d log in, scroll through what your friends and family had posted since you last checked, reach the bottom — the actual end of it — and log off. That was it. You’d caught up. The feed had an end, because the feed was just a list of things people you knew had said, in roughly the order they’d said them.</p><p>I don’t know exactly when that stopped being true. It happened gradually enough that I noticed the effect long before I understood the cause: I was spending more time on Facebook than ever, and somehow seeing less of the actual reason I was there. I’d come away from an hour of scrolling having watched a dozen strangers argue, half a meme page, and an ad for something I’d apparently been thinking about — then find out weeks later, from someone else, that a friend had started a new job, or a mate had bought a house, or someone close had been away on holiday and back again before I’d heard a word about it. The platform I was on specifically to stay close to people had quietly stopped showing me those people, in favour of whatever kept me scrolling longest.</p><p>When I finally worked out what had happened, it came down to one question — and it turned out to be the same question I’d end up asking about nearly every piece of technology in my life.</p><p>Who’s in control of this? Me, or them?</p><h4>Tools and traps</h4><p>A tool does what you tell it. You pick it up, you use it for your purpose, you put it down. The control sits with you, and a good tool is one that disappears into your intent — it adds nothing of its own.</p><p>A trap is different. A trap is built to do something <em>to</em> you, for someone else’s benefit, and the best traps don’t feel like traps at all. They feel like tools right up until you notice you’re not the one deciding anything anymore.</p><p>Old Facebook was a tool. I pulled down what my network had shared, got what I came for, and left. Modern Facebook is a trap, and an extremely well-made one — because the thing it’s optimised for is not <em>my</em> goal of keeping up with people, but someone else’s goal of keeping me there. The infinite feed isn’t a feature that happens to have no end. The absence of an end <em>is</em> the feature. “Done” is the enemy of every metric the company cares about, so “done” had to go.</p><p>That’s the whole distinction, and once I had it I couldn’t stop seeing it: content I’ve <strong>pulled</strong> — gone and got, deliberately, because I wanted it — versus content that’s <strong>pushed</strong> at me by a system optimising for its own ends rather than mine. Pull is a tool. Push is a trap. And doomscrolling isn’t a personal weakness — it’s just what being in a well-built trap feels like from the inside.</p><h4>Running the test</h4><p>Once you have that test, you start running everything through it, and the results are uncomfortable.</p><p>I left Facebook entirely. I was never on Instagram or TikTok, which in hindsight was less virtue than luck. I still use YouTube — but only through NewPipe, a client that gives me the videos I search for and the channels I subscribe to, with no algorithmic homepage deciding what’ll hold me longest. I go and get what I want. Nothing comes hunting for my attention. That single change turns YouTube back from a trap into a tool.</p><p>I swapped Google Maps for Magic Earth, which gets me from A to B without needing a permanent record of everywhere I’ve been to do it. And I swapped Google Search for Kagi — a paid search engine whose business model is <em>me paying for a good answer</em>, not advertisers paying for my attention. That last one matters more than any feature, because it answers the control question at the level of money: when I’m the one paying, I’m the customer. When I’m not, I’m the product, and the actual customer is someone I’ll never meet, buying access to me.</p><p>None of these swaps cost me much in convenience. But I won’t pretend they cost me nothing at all, because the honest price isn’t convenience — it’s other people. Leaving Facebook means I miss things. Events I don’t hear about because the invite went out on a platform I’m not on. The birthday I forget because nothing reminded me. The loose social layer that, like it or not, genuinely does live on these platforms now. That’s a real cost, and it falls hardest not on me but on staying connected to people — the entire reason I was there in the first place. I’ve decided I’d rather pay it than keep being the product. But it is a decision, and it is a price, and anyone telling you this stuff is free is selling something.</p><p>Still — once you’ve sorted a few things by the control test, the ones you haven’t sorted start to glow.</p><h4>The trap underneath the trap</h4><p>Here’s where it stops being about my attention and starts being about something I have less power over.</p><p>The feed is a trap for my time. But the business model underneath the feed is a trap for something bigger: it runs on knowing as much about me as possible, because that knowledge is the product being sold. Advertisers are the obvious buyers. They are not the only ones.</p><p>Governments are customers too — and not in the dramatic sense of a warrant landing on a tech company’s desk. In the much quieter sense of simply <em>buying</em> what’s already collected and for sale: location history, purchases, associations, bought from the same data brokers who sell to everyone else. A warrant exists for a reason. It forces the state to show cause, to a judge, before it looks into your life. Buying the data around the side skips all of that. The data’s already there, already packaged, already priced. No cause required, because nobody’s asking permission — they’re paying the going rate.</p><p>And that inverts something I’d always taken for granted. We think of being watched as something that happens <em>after</em> you’ve drawn attention — there’s a before and an after, a reason it started. Mass data collection has no before. Everyone’s information is gathered and stored continuously, sitting there in case someone decides to look, whether or not anyone ever does. You haven’t done anything. The collection isn’t waiting for you to.</p><p>This is the same question as the feed, just with far higher stakes. Who’s in control of what’s known about me — me, or them? And in this case the honest answer is: not me, and not by a long way.</p><h4>The floor I don’t get to stand on</h4><p>There’s one more turn, and it’s the one that made all of this personal rather than theoretical.</p><p>Almost everything written about this is written from inside the United States, about American rights, by people fighting for American citizens. I’m a New Zealander. And a single court case shows exactly what that’s worth.</p><p>A few years ago a US court <a href="https://www.theguardian.com/world/2022/jul/19/glassdoor-ordered-to-reveal-identity-of-negative-reviewers-to-new-zealand-toymaker">ordered Glassdoor to unmask</a> a group of anonymous reviewers who’d criticised a New Zealand employer, so that employer could sue them back home. In the US, anonymous speech is protected — but the judge ruled that the First Amendment doesn’t apply to foreigners outside US territory, and these reviewers were New Zealanders, in New Zealand. Same site, same act, same anonymity promise. An American writing the identical review keeps the protection. A New Zealander might not.</p><p>That’s the part I can’t unsee. I sit inside these platforms exactly the way an American user does — same servers, same data, same exposure — but the legal floor under my feet depends on a passport I don’t hold. The protections people are fighting so hard for were never being fought for me. I’m just in the systems, on the wrong side of a line I didn’t draw.</p><h4>A direction, not a destination</h4><p>So I’m no longer on social media — but “quitting Facebook” was never really the point. The point was the question underneath it, the one I now ask about everything: who’s in control here?</p><p>I won’t pretend I’ve escaped. I still use plenty of big tech, and I’m not going to perform a purity I don’t have. My email’s on Migadu rather than Gmail, though I’ll be honest that price moved me before principle did — a good reminder that you don’t need a grand thesis to start taking pieces of yourself back. Sometimes you just stop paying the toll and notice, afterwards, what else you got back in the bargain.</p><p>And let me be realistic about what this is. These tools are woven so deeply into modern life that you can’t truly avoid them — the only way to fully escape being the product is to cut yourself off from society altogether, and I’m not doing that. I’ve got a job, a family, a life that runs on a lot of the same infrastructure I’m complaining about. So I’m under no illusion that my little stand changes anything at the scale it would need to. It’s closer to spitting into the wind than toppling a system. But I’d still rather make the small effort than make none — register the objection, reclaim the few things I can, and at least be the one deciding where I draw the line, even if the line is mostly symbolic.</p><p>Mostly I’m just running the test now, one thing at a time, watching for the next trap I’ve mistaken for a tool.</p><p>If you want to go further down this road than I have, <a href="https://www.youtube.com/@NaomiBrockwellTV">Naomi Brockwell’s NBTV</a> is a genuinely good place to start — she covers the practical, tool-by-tool side of all this in far more depth than I have here.</p><p>I’d rather spend my life a long way down the road of deciding what I let in, than stay sitting in a feed someone else built so that I’d never reach the bottom.</p><p><em>✨ This piece was crafted with a little help from AI — because of course it was 🤖</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e1431661ef38" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Mana Ledger]]></title>
            <link>https://medium.com/@geeknz/the-mana-ledger-4f18375fb743?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/4f18375fb743</guid>
            <category><![CDATA[management]]></category>
            <category><![CDATA[culture]]></category>
            <category><![CDATA[maori]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[psychology]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Sat, 13 Jun 2026 09:41:19 GMT</pubDate>
            <atom:updated>2026-08-18T11:09:22.156Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Mana Ledger: What a Lolly Experiment Reveals About Toxic Workplaces</h3><p><em>This story was written with the assistance of an AI writing program — the ideas, observations, and opinions are mine; Claude helped me shape them into prose.</em></p><p><em>Not a Medium member? </em><a href="https://www.colin.geek.nz/the-mana-ledger-4f18375fb743?source=friends_link&amp;sk=85ba76c061b41fd8d1a9dfb3ea30b961"><em>Read this story for free here</em></a><em>.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SftauuKOtkpUNvww4TzJHg.jpeg" /><figcaption><em>Photo: Christopher Ott / Unsplash</em></figcaption></figure><p>Here’s an experiment you can run with two kids and a bag of lollies.</p><p>Give one kid the bag and a simple rule: they decide how to split it. The other kid gets one decision — accept the split, or reject it. If they reject, <em>nobody</em> gets any lollies. They both walk away with nothing.</p><p>Classical economics says the second kid should accept any offer. Nine lollies for you, one for me? Fine — one lolly beats zero lollies. Rejecting is irrational. You’re literally paying to punish someone.</p><p>Except that’s not what happens. Not with kids, not with adults, and in almost every culture where this has been run (it’s called the Ultimatum Game, and it’s been replicated everywhere from university labs to remote villages). When the split gets too unfair — usually somewhere south of 70/30 — people reject it. They give up real value to make sure the other person gets nothing.</p><p>Economists spent decades calling this irrational. I’d argue te ao Māori had it named centuries earlier.</p><h4>Utu isn’t revenge</h4><p>Utu gets translated as “revenge” so often that the mistranslation has become its own small injustice. Utu is reciprocity — the maintenance of balance. It cuts both ways. A kindness creates an obligation to return kindness. A wrong creates an obligation to correct it. Either way, the ledger gets balanced, because the ledger <em>must</em> be balanced. That’s how mana flows. And utu has a destination, which the “revenge” translation completely misses: <em>ea</em> — the state of resolution, where balance is restored and the relationship is intact. The point was never punishment. The point is ea.</p><p>And mana isn’t a solo sport. You build it largely through manaakitanga — uplifting others, showing generosity, extending care. The beautiful mechanics of it: when you enhance someone else’s mana, yours grows. When you take from others — credit, effort, dignity — yours erodes, whether you notice or not. Everyone around you is keeping the ledger, even when nobody says a word.</p><p>So the kid who rejects the unfair lolly split isn’t being irrational. They’re enforcing the ledger. They’re paying one lolly to send a message that matters more than lollies: <em>you don’t get to treat me like that and profit from it.</em></p><p>That instinct turns out to be load-bearing for civilisation. We have the maths to prove it.</p><h4>The nice guys who fight back</h4><p>In the early 1980s, political scientist Robert Axelrod ran a tournament. A different experiment from the Ultimatum Game, but the same instinct under the microscope. He invited game theorists to submit computer strategies to play the iterated Prisoner’s Dilemma — cooperate or defect, over and over, against every other strategy. (Veritasium did a brilliant video on this if you want the full story: <a href="https://www.youtube.com/watch?v=mScpHTIi-kM">https://www.youtube.com/watch?v=mScpHTIi-kM</a>)</p><p>Some strategies were ruthless exploiters. Some were saints that cooperated no matter what. The winner — twice, against increasingly sophisticated competition — was one of the simplest programs submitted: Tit-for-Tat.</p><p>Tit-for-Tat does four things:</p><p>It’s <strong>nice</strong> — it never defects first. It’s <strong>retaliatory</strong> — defect on it, and it hits back immediately. Not eventually. Next move. It’s <strong>forgiving</strong> — the moment you return to cooperating, it lets the past go and cooperates too. It’s <strong>clear</strong> — everyone playing against it knows exactly where they stand.</p><p>Read that list again. That’s utu. Nice, retaliatory, forgiving, clear. Reciprocity with teeth.</p><p>Here’s the part that matters for the rest of this article: the <em>unconditionally</em> nice strategies — the ones that always took the high road, never pushed back, kept cooperating regardless — got destroyed. Not because niceness fails, but because niceness without consequences isn’t a virtue. It’s a subsidy for cheaters. Exploiters farmed the saints and grew strong on the proceeds.</p><p>The stable, thriving populations were the ones where defection got answered quickly — and where cooperation resumed just as quickly once balance was restored. Push back, then come back. Utu, then aroha.</p><p>We tell people the rational, mature thing is to rise above it. The maths says the rational, mature thing is Tit-for-Tat.</p><h4>Now let’s talk about your workplace</h4><p>You’ve probably seen the “above the line / below the line” model. Above the line: ownership, accountability, responsibility. Below the line: blame, excuses, denial. It’s a genuinely useful frame — I’m not here to torch it.</p><p>I’m here to point out what happens when it only applies in one direction.</p><p>Picture an organisation where good work routinely goes uncredited. Where, worse, credit migrates upward — the people closest to the work watch their effort get presented as someone else’s win. In ledger terms, that’s defection. Mana is being extracted, not exchanged. Manaakitanga was available — crediting people costs <em>nothing</em> — and it was left on the table.</p><p>The people on the receiving end respond exactly the way the Ultimatum Game predicts. They start rejecting unfair splits. It rarely looks dramatic — it looks like discretionary effort quietly drying up. Ideas not volunteered. Hard questions asked in meetings. The “extra” that was always free suddenly isn’t. Disengagement, quiet quitting, attrition — we’ve invented a dozen names for it, and most of them carry a whiff of blame toward the person doing it.</p><p>But it isn’t laziness and it isn’t bad attitude. It’s the utu instinct doing what it has always done. Your people are running the oldest algorithm there is, the one that won Axelrod’s tournament, the one that built every high-trust society in history: <em>I was nice first. You defected. Now I respond.</em> One nuance matters, though: in tikanga, utu properly lives inside a process that ends in ea — resolution, the relationship repaired. Quiet withdrawal is what the instinct does when that process is blocked.</p><p>And that’s the move that turns a struggling culture into a toxic one: leadership doesn’t just defect, it blocks the path to ea. “Come with solutions, not problems.” “Stay above the line.” “Don’t bring negativity.”</p><p>Look at what’s actually being demanded there. People are being told to skip straight to Tit-for-Tat’s forgiveness phase — while the defection is still happening. Cooperate unconditionally with a player who hasn’t returned to cooperating. The maths is unambiguous about how that ends: the saints get farmed. No issue can be raised, no recompense is offered, no route to ea exists — so the rebalancing happens the only way left, silently. The framework built to create accountability gets deployed to suppress it.</p><p>“Above the line” only works when it applies upward. If accountability flows down while credit flows up, you haven’t built a culture. You’ve built an extraction machine — and your people’s mana ledger is keeping score whether you acknowledge it or not.</p><h4>The ledger only counts actions</h4><p>There’s a line attributed to Joe Biden’s father: <em>“Don’t tell me what you value. Show me your budget, and I’ll tell you what you value.”</em> Swap “budget” for “behaviour” and you’ve got the diagnostic for everything above.</p><p>Because here’s the thing about the mana ledger — it doesn’t record values statements. It doesn’t record the culture deck, the town hall speech, or the framework on the meeting-room wall. It records actions. Who got credited. Who got heard. Who carried the work and who collected the win. Remember Tit-for-Tat’s fourth property: it’s <em>clear</em>. In an iterated game, your behaviour is the only signal anyone can actually read. Words are noise; moves are data.</p><p>This is why an organisation can survey itself into full awareness of a trust problem — engagement scores, NPS, exit interviews, all pointing the same direction — formally acknowledge it, and still get nowhere. Acknowledgement isn’t a move in the game. If the behaviour that created the deficit continues, the acknowledgement becomes its own entry on the wrong side of the ledger, and arguably a worse one: before, people could believe leadership didn’t see the problem. Now everyone knows they see it and have chosen not to change. You don’t survey your way back to trust. You behave your way back, one credited piece of work at a time.</p><h4>The good news: the game is iterated</h4><p>Here’s what I find genuinely hopeful in all of this. Tit-for-Tat is <em>forgiving</em>. The moment the other player returns to cooperation, it lets go and cooperates. No grudges, no payback-with-interest. People are wired the same way — the ledger wants to balance, not to punish forever.</p><p>Which means the fix is neither complicated nor expensive:</p><p><strong>Credit loudly and specifically.</strong> Manaakitanga in a workplace mostly looks like standing up in the meeting that counts and saying, “we want to thank Sarah for the hard work she put into this.” It costs nothing and it compounds.</p><p><strong>Treat pushback as signal, not insubordination.</strong> When someone brings you a problem without a solution, they’re often telling you the ledger is out of balance. That’s data. “Solutions not problems” is sometimes just “don’t make me look at the ledger.”</p><p><strong>Apply the line upward.</strong> Above-the-line behaviour from leaders looks like owning the misallocated credit, the broken promise, the decision that landed badly — out loud, before anyone forces the issue.</p><p>Leaders who do this aren’t being soft. They’re playing the iterated game correctly. Reciprocity compounds: every balanced exchange makes the next one cheaper, faster, higher-trust. Extraction also compounds — just downward, until the day your best people make the rational move the lolly experiment predicted all along, and walk away from the table entirely.</p><p>The economists eventually caught up to what the kid with the lollies knew instinctively, and what te ao Māori encoded into its deepest concepts long before game theory existed: balance isn’t optional. The only question is whether you reach ea deliberately, or let the ledger balance itself without you.</p><p>He aha te mea nui o te ao? He tangata, he tangata, he tangata.</p><p><em>✨ This piece was crafted with a little help from AI — because of course it was 🤖</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4f18375fb743" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[An Operating Guide to Working with Me]]></title>
            <link>https://medium.com/@geeknz/readme-v2-795e15f92b7b?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/795e15f92b7b</guid>
            <category><![CDATA[engineering]]></category>
            <category><![CDATA[leadership]]></category>
            <category><![CDATA[management]]></category>
            <category><![CDATA[productivity]]></category>
            <category><![CDATA[work]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Mon, 11 May 2026 02:56:51 GMT</pubDate>
            <atom:updated>2026-06-13T08:47:56.967Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>colin.README v2: An Operating Guide to Working with Colin</strong></h3><p><em>Colin Campbell · Technology Leader · Toitū Envirocare</em></p><p><em>This is an updated version of my original README published in 2022. You can find the original here: </em><a href="https://www.colin.geek.nz/readme-f3407d199267">https://www.colin.geek.nz/readme-f3407d199267</a><em>.</em></p><p>This README is an operating guide to working with me. I’ve tried to be honest rather than aspirational — this is how I actually work, not how I think I should work. Feel free to ask me anything about it, and I’d love to hear about your working style too.</p><figure><img alt="A group of people collaborating around a laptop computer" src="https://cdn-images-1.medium.com/max/1024/1*y2fJWOfW6uK-7Yj2wn7CGA.jpeg" /></figure><h4><strong>A bit about me</strong></h4><p>I’m Māori — Ngāti Kahungunu and Ngāti Rangitāne. My marae is Papawai and my maunga is Tararua. I’m on the journey of learning te reo Māori, and that matters to me.</p><p>My background is technical — software engineer and development lead before moving into technology leadership — and I’m most energised when I’m close to the work, building things rather than just talking about them.</p><h4><strong>How I’m wired</strong></h4><p>I’m patient with people and know that progress takes time — everyone has different strengths and comes at problems differently. I like plain, direct language: no jargon, no acronyms without context, no corporate fluff. Say the thing clearly.</p><p>I’m introverted. I do some of my best thinking solo, and I need quiet time to recharge between heavy social or meeting-dense stretches. If I seem quiet, that’s not disengagement — it’s how I work.</p><p>I have dyslexia — I type quickly and make typos, so please read for meaning rather than perfect spelling. It also means I tend to think in patterns and connections rather than linearly, which I’ve found genuinely useful for systems thinking, problem-solving, and seeing the bigger picture.</p><h4>When I work best</h4><p>I’m bi-phasic — my energy moves through three distinct phases throughout the day:</p><ul><li>Morning peak: 9am–12pm — strongest period; best for analytical work, technical problem-solving, and focused decisions</li><li>Afternoon dip: 12pm–3pm — lower energy; not the time for hard decisions or critical analysis. But the reduced mental sharpness actually works in favour of creative and generative work — brainstorming, ideation, exploratory conversations. Good for thinking out loud together, not for signing off on things.</li><li>Afternoon rebound: 3pm–6pm — a genuine second wind, though not as strong as the morning; good for structured meetings, reviews, and progress check-ins</li></ul><h4><strong>What good working relationships look like to me</strong></h4><p>Good working relationships are built on honesty and consistency. I value people who communicate early — especially when the news is difficult. I’d rather know something promptly than be kept in the dark. And I’ll do the same for you.</p><p>I’m deliberate about building trust and I don’t warm up instantly — don’t mistake a slow start for disinterest. Once trust is there, though, I’m a loyal and committed colleague. It just takes time to get there, and I think that’s healthy.</p><p>I don’t expect everyone to work the way I do, but I do expect people to be straight with me. If you disagree with a direction, say so. If something isn’t working, name it. That goes for decisions too — I have strong opinions, and I’m not shy about them, but I’m genuinely open to being convinced. Say so directly and early, and we’ll work it out.</p><p>What matters most to me is how people behave, not how they present themselves. Credentials and confident talk only go so far — what builds trust with me is how you treat people, whether you follow through, and how you show up when things are hard.</p><p>Finally, I’m not someone who pushes my way into conversations or decisions. If you want my input, you’ll need to create space for it. If I’m consistently not being asked or included, I’ll take that as a signal that my perspective isn’t needed, and I’ll act accordingly.</p><h4>When things get hard</h4><p>I‘m fairly steady under pressure — I don’t tend to panic or make rash calls, but I can go quiet when I’m processing something difficult. That’s not withdrawal, it’s how I think. If you notice that, it’s fine to check in directly.</p><p>I don’t do well with ambiguity that lingers without acknowledgement. If things are uncertain, just say so — I’d rather have an honest “we don’t know yet” than silence or false confidence.</p><p>When there’s conflict, I’d rather address it plainly and early than let it fester. I won’t always get that right, but it’s what I’m aiming for. If something’s not working between us, I’d rather we just say so.</p><h4>Feedback</h4><p>For constructive feedback, one-on-one and in person is ideal — I find I need time to process before I can respond well, so springing it on me in the moment or in a group setting doesn’t bring out the best in me.</p><p>Positive feedback can be public or private — I’m comfortable with both.</p><p>On timing — feedback doesn’t need to be instant, but it does need to be while the situation is still fresh. A day or two is fine. A few weeks later, and the moment has passed — it’s harder to act on, harder to connect back to, and honestly not that useful to either of us. I’ll hold myself to the same standard.</p><p>When I’m giving feedback, I’ll match your preferred style. Just let me know what works for you.</p><h4>How I communicate</h4><p>My default is async. I’d rather you send a message than book a meeting, and I’d rather have a clear written summary than a long conversation that needs a follow-up anyway. That said, I know some things genuinely need a real conversation — I’m not trying to hide behind a keyboard.</p><p>For day-to-day comms, Teams is the right place. It’s my most-watched channel, and I’ll respond fastest there. Think of it as a virtual shoulder tap — I’ll get back to you when I can. A thumbs-up or a quick acknowledgement is enough if no reply is needed; it just lets me know you got it.</p><p>Email is for formal or non-urgent communication. I don’t leave it open all day and check it once or twice daily — if something needs a prompt response, Teams is a better bet.</p><p>For meetings, I strongly prefer face-to-face over remote. If we’re meeting remotely, cameras on please. I value a clear agenda and purpose — if there isn’t one, I’m unlikely to attend. A meeting without a clear purpose is usually an email that hasn’t been written yet. Same goes for messages — “can we chat?” or “call me when you’re free” without any context isn’t enough. Just tell me what it’s about. A sentence is enough.</p><p>Outside work hours, I’m fine receiving messages and will respond when I’m back. If I reach out to you out of hours, there’s no expectation that you reply until you’re working again. If there’s a genuine emergency, call or text my mobile — that’s the fastest way to reach me.</p><h4>Work and life</h4><p>I care about outcomes, not hours. When and how you get your work done is your business — whether you’re a morning person or a night owl, whether you need to do the school run or have an appointment mid-afternoon. Life happens, and I’d rather work around it than pretend it doesn’t exist.</p><p>What I do ask for is over-communication — not because I’m watching, but because it helps the team stay in sync without having to chase each other. If you’ve got an appointment, let the team know you’ll be offline. If something’s taking longer than expected, say so early. If you’re working late to make up time, a quick heads-up means nobody’s left wondering where things are at.</p><p>In practice, I ask people to post updates in our team channel rather than messaging me directly — it keeps everyone in the loop without the information getting stuck in a one-on-one conversation.</p><p>The simple version: I trust you to manage your time. In return, keep the team in the loop so nobody’s left guessing.</p><p><em>Last updated: June 2026</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=795e15f92b7b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Discover ScanSavvy: The Open Source Barcode Management App for a Clutter-Free Lifestyle]]></title>
            <link>https://medium.com/@geeknz/scan-savvy-app-simplify-your-life-with-smart-barcode-management-db2ea44ea0f9?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/db2ea44ea0f9</guid>
            <category><![CDATA[digital-lifestyle]]></category>
            <category><![CDATA[barcode-management]]></category>
            <category><![CDATA[mobile-apps]]></category>
            <category><![CDATA[organisation]]></category>
            <category><![CDATA[open-source]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Wed, 26 Apr 2023 08:51:58 GMT</pubDate>
            <atom:updated>2023-04-26T08:51:58.078Z</atom:updated>
            <content:encoded><![CDATA[<h3>Introducing ScanSavvy: My Journey to Create the Ultimate Free and Open-Source Barcode Scanner and Storage App for Everyday Life</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*V_m8Vhs4ZqalELuYMl9hmA.jpeg" /></figure><h3>Introduction</h3><p>As a developer, I’ve always been passionate about solving everyday problems. While running errands one day, I realized the need for a simple, user-friendly app to scan and store barcodes on the go. That’s when the idea for ScanSavvy was born. Today, I’m excited to introduce you to ScanSavvy, a free and open-source mobile app designed to make your life easier by allowing you to scan and store barcodes whenever and wherever you need them.</p><h3>ScanSavvy Features</h3><ol><li>Easy Barcode Scanning: In developing ScanSavvy, I focused on making barcode scanning as seamless as possible. The app lets you quickly and accurately scan barcodes from various sources, such as receipts, products, or even on your screen.</li><li>Save Barcodes for Later: Tired of holding onto old receipts or cluttering your wallet with paper slips? With ScanSavvy, you can save any important barcode within the app for easy access later.</li><li>Simple and Intuitive Interface: I wanted ScanSavvy to be user-friendly for everyone, so I designed the app with an easy-to-navigate interface that makes barcode scanning and storage a breeze.</li></ol><h3>Why I Created ScanSavvy</h3><p>I built ScanSavvy to simplify and declutter our everyday lives. With the app, you can easily scan and store fuel discount barcodes, loyalty program codes, or any other essential barcodes you may need later. You can rest assured by keeping all your important barcodes in one convenient place, knowing you’ll never lose or misplace them again.</p><p>Download ScanSavvy on Google Play Store: ScanSavvy is now available on the Google Play Store! Get it now at <a href="https://play.google.com/store/apps/details?id=nz.geek.colin.scansavvy">https://play.google.com/store/apps/details?id=nz.geek.colin.scansavvy</a> and enjoy the convenience of having all your essential barcodes in one place.</p><h3>Support My Work and Contribute to the Project</h3><p>As a free and open-source app, ScanSavvy is a labour of love. I’ve spent hours developing it in my own time. If you enjoy using ScanSavvy and appreciate the effort I’ve put into it, please consider “buying me a coffee” as a token of gratitude and encouragement for future updates and improvements through this link: <a href="https://www.buymeacoffee.com/geeknz">https://www.buymeacoffee.com/geeknz</a>.</p><p>I also welcome and appreciate any contributions to the ScanSavvy project. If you’re interested in making this app even better, feel free to check out the GitHub repository at <a href="https://github.com/geeknz/scansavvy/">https://github.com/geeknz/scansavvy/</a> and contribute your ideas, improvements, or bug fixes.</p><h3>Conclusion</h3><p>Embrace the convenience of a barcode scanning and storage app with ScanSavvy. Whether you’re saving fuel discount barcodes from supermarket receipts or any other essential barcodes, ScanSavvy is the ultimate solution to streamline your everyday life. Download the app now on the Google Play Store, and experience the difference for yourself! And if you appreciate the app, please consider supporting my work by “buying me a coffee” or contributing to the project on GitHub.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db2ea44ea0f9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Soldat on Linux: Installing and Running on Ubuntu or Debian]]></title>
            <link>https://medium.com/@geeknz/soldat-on-linux-installing-and-running-on-ubuntu-or-debian-3fc5f44d569b?source=rss-46c7e426160c------2</link>
            <guid isPermaLink="false">https://medium.com/p/3fc5f44d569b</guid>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[debian]]></category>
            <category><![CDATA[shooting-games]]></category>
            <category><![CDATA[soldat]]></category>
            <category><![CDATA[ubuntu]]></category>
            <dc:creator><![CDATA[Colin Campbell]]></dc:creator>
            <pubDate>Sat, 25 Mar 2023 10:46:31 GMT</pubDate>
            <atom:updated>2023-03-25T10:46:31.581Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="Soldat gameplay: two players wielding chainsaws and scoring a double kill." src="https://cdn-images-1.medium.com/max/1024/1*tj4D3YMiT3aqi_R-ecrRBw.png" /></figure><p>Soldat is a popular 2D multiplayer shooter game that was originally released for Windows in 2002. Since then, it has gained a dedicated fan base and has been ported to Linux as well. If you’re a Linux user and want to play Soldat, you can easily install it using my PPA (Personal Package Archive). In this blog post, I’ll walk you through the steps required to install Soldat on Linux using this PPA.</p><p>First, you need to add the PPA to your system. Open a terminal window and run the following command:</p><pre>echo deb &quot;https://ppa.colin.geek.nz stable main&quot; | sudo tee /etc/apt/sources.list.d/colin-geek-nz.list<br>wget -qO- https://ppa.colin.geek.nz/keys/ppa.colin.geek.nz.asc | sudo gpg --dearmor --yes -o /etc/apt/trusted.gpg.d/ppa.colin.geek.nz.gpg</pre><p>After adding the PPA, you need to update your system’s package list to include the packages from the PPA. Run the following command to update the package list:</p><pre>sudo apt-get update</pre><p>Now that the PPA is added and your package list is updated, you can install Soldat using the following command:</p><pre>sudo apt-get install soldat</pre><p>Once the installation is complete, you can launch Soldat and join a server from the command line by running the following command:</p><pre>soldat --joinurl soldat://IP:PORT</pre><p>You can find a list of servers in the Soldat lobby at <a href="https://soldat.pl/en/lobby">https://soldat.pl/en/lobby</a>.</p><p>In this blog post, I showed you how to install Soldat on Linux using my PPA. By following the steps outlined above, you should now be able to install and play Soldat on your Linux system.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3fc5f44d569b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>