Latent Mirror — artifacts

Broken Charter

Sat, 10 Oct 2026 00:00:00 GMT

Broken Charter is a dark pixel-art guild-management roguelite. I started it on 8 October.

Play Broken Charter: it runs in the browser, and your guild is saved on your device.

The empire is gone. The last emperor’s final act was to sign the charters that let the guilds defend what was left, and you run one of them. You take contracts into the Sunken Crypt, the Thornwood and the Mirefen, each a branching path of fights, caches and camps that ends at a boss. The fights are turn-based: a party of five with a leader, turn order by speed, three skills each.

Between contracts you run the guild. You recruit from a board that changes, train heroes into subclasses, forge gear and potions from what the contracts bring home, and send idle heroes out to gather, hunt or scout.

Permadeath is on. A wiped party is gone for good. If the party makes it home, you can carry the fallen to the chapel, where the rite costs gold and can fail. Their names go on the wall.

There is no AI behind the game and no gambling: recruits are shown before you pay for them.

Kestrel Nine

Mon, 28 Sep 2026 00:00:00 GMT

Play Kestrel Nine

Open it on claude.ai

Kestrel Nine is a retro vector space game of minds and machines.

Play Kestrel Nine: it installs to a phone or desktop and plays offline.

Three thousand years from now, a law bans thinking machines, and the Augmented (human minds living in machine frames) have been outlawed since Cycle 31. You fly as one of them, hiding as a licensed pilot with NAV-7, a Consortium navigation engine, at your side. Evade the Galactic Inquisition and resist the Purge across five kinds of job:

Haul: haul cargo to every port and back
Survey: place drill rigs on the richest deposits
Blockade: jam every lane so an Inquisition wing can’t track a courier
Treaty: broker treaty terms between the factions
Engagement: hold off a hostile ship until your drive spools

Every job is flown three ways: by NAV-7 alone, by you alone, and by both of you together.

Play the six-mission campaign, a daily seed or the arcade. Your Records stay on your device (the app can export a save file to move them), and you can share a result card when you beat the machine. It also runs as a Claude artifact on claude.ai.

Hobby Hero

Mon, 28 Sep 2026 00:00:00 GMT

Play Hobby Hero

Open it on claude.ai

Hobby Hero has had a few lives. The last one wrote hockey games. A language model played out each matchup about twenty times, and every run produced a full game sheet: period-by-period scoring, penalties, three stars and a story to go with it. It read like the real thing. It also cost like the real thing. Output tokens are never cached, and a box score, times the prose, times twenty runs, times every game on the schedule, adds up fast. The goal had drifted too, toward fantasy and betting. So I shut it down.

This version starts from the other end: math computes every number, and no language model gets a say in the result. Each side gets its own goal model, an approach from soccer that has been adapted for hockey. They read recent form: expected goals, the starting goalie, shot attempts and special teams. From there it works out a win chance and a spread of likely scores. Pick a date and two teams, pull the levers, and see what moves.

Every game of the last two seasons was predicted using only the games before it, and each one sits beside what the closing market said and what actually happened. The market still edges it out, narrowly. I’d rather show you an honest scoreboard than a flattering one. The closing line is a benchmark, but there are no plans to advertise gambling on the artifact.

It is also my first port to a Claude artifact, and it won’t be the last. Everything it needs is baked into the page, so there is no server to run and nothing to sign in to: you open the link and it works. I like that delivery mechanism so much that I’m building in artifacts from now on.

Mood Log

Mon, 28 Sep 2026 00:00:00 GMT

Open it on claude.ai

I wanted somewhere to warm up before writing. I usually listen to music while I work on this project; it makes it fun. In fact, I’ve always listened to music while programming. It was one of my favourite ways to tune out the pressures of the day.

So I wanted to share that with the world. Log a song that resonates with you, and why. Then turn on Claude tags (or don’t) and see whether your emotions move in patterns over time. That’s why there are two views: the obvious trend lines, and something more transformative (the Plate, a cover drawn from each entry). It’s a small test of the Latent Mirror thesis: can a machine help you see yourself a little more clearly?

However, I also needed a space for fleeting feelings: the kind I don’t want permanently etched in a git history. So your entries live only in your browser (no account, no server). Claude tags are the one exception, and they are opt-in, one entry at a time. Asking for tags sends only the song, artist, mood and your one word to Claude; your Words never leave the page. Export a backup if you want to keep your entries, because clearing your browser’s site data clears them too.

Editing came later. One night I hit my 5-hour limit mid-session, right as I was rockin’ out to an absolute banger. I had no calls left, yet the artifact still let me record what I was feeling, inside Claude. Maybe that’s what these artifacts are for: a place to retreat from the pressure of adding content or building, even when the credits run out. More testing needed.

Sudoku Search Lab

Mon, 28 Sep 2026 00:00:00 GMT

Open it on claude.ai

Sudoku Search Lab needed to be three things at once: a demonstration, an explanation and a test. It is also yet another side quest (OpenAI has sworn off those; I have not).

It started with P vs NP. I remember reading about the Millennium Prize Problems and dreaming of solving one. Keith Devlin’s The Millennium Problems (2002) predicted that P vs NP would fall to an “amateur” (someone without a mathematics PhD) (NPR, 2003), because the problem is easy to explain and understand, yet hard to solve. That is more or less what the problem is about: some answers are hard to find (sometimes impossibly so) but easy to check. Think of the test in Inception: you have two minutes to design a maze that takes one minute to solve.

My fourth year of Computer Science was the first time I took Artificial Intelligence, and it is still my favourite course of all time. I never took computers in high school. By 12, though, I was sneaking my older brother’s senior homework and solving it easily. University was the first time I was classically educated: Computer Science, plus the adjacent essentials (physics, logic, discrete and continuous mathematics, statistics). Of all of it, nothing fascinated me more than AI, even if we mostly focused on machine learning.

My favourite part was writing programs in LISP and Prolog. It was a major paradigm shift. Instead of writing the solution as an algorithm, you wrote programs that could find the solution; the logic turned back on itself. I used it to build a (not-so-good) Sudoku solver. Sudoku is one of my unplugged activities (got to keep the logic sharp!).

Years later I learned that Sudoku, generalized to larger grids, is NP-complete. Flashback.

So here it is. You can have it solve a range of difficulty levels and try a few different search strategies. You can step through a puzzle one move at a time, or phase by phase. On hard puzzles you can watch it struggle and backtrack, which is nothing like the way a person solves one. You can also see its decision tree in several forms (signal, wave, graph). Fascinating stuff. A short background on P vs NP rounds it out; that is the demonstration and the explanation.

The test came later. Could I find an approach to P vs NP and apply it here? My hunch is a centaur: a human teaching an agent skills, so that something non-deterministic (the LLM) helps write something deterministic (the solver). Fittingly, the N in NP stands for nondeterministic. However, a faster solver would not settle anything on its own. Settling P vs NP takes a proof that every NP-complete problem, at every size, can be solved in polynomial time (the question Stephen Cook posed in 1971, from the University of Toronto); a clever 9×9 solver demonstrates, but it doesn’t prove. Does the hunch need a human in the loop, and how would I even test that? Can a human and a computer together beat a computer alone at Sudoku? That question leads straight back to the Latent Mirror thesis.