The Guided Tour

I hold my work to a standard I learned running incident bridges: if I can't explain a system clearly, I don't understand it yet. This page walks through five projects the way I'd brief a decision-maker - the mechanism, what each number means in practice, and the questions worth asking me about each. Every figure comes straight from the published results, losses included, because a result you can't check is just a story. The bundled agents are deliberately simple scripts, so every number proves the measuring instrument, never any AI model's intelligence. The working vocabulary at the bottom defines the field's terms the way I use them.

The five projects

Memory

wiki-memory-lab

An AI's long-term memory, kept as a wiki a human can read and check.

The intuition

Imagine a doctor who, before every appointment, re-reads your entire file back to page one - every cold, every checkup, thirty years of paperwork. This project teaches the AI to work like a good nurse instead: keep the chart organized, pull only the pages that matter for today's visit, and leave the rest in the cabinet.

What the numbers mean

Selective reading used 0.61× the tokens of loading everything - about 61 cents on the dollar, and tokens are the meter AI bills run on. On a small pile of notes, though, the simple load-everything approach won (selective reading cost 1.27× there), and that loss is published rather than buried.

What this demonstrates

I measure cost honestly and publish the exact case where the clever method loses.

Ask me: "Why publish the result where your own approach lost?" - Because the loss is the finding. The wiki pays off when memory is big and messy; below that size, simpler is cheaper, and now I know where the line sits. If I only showed you the win, you'd have no reason to trust the win either.
Evaluation

agent-mutation-lab

Tests AI code reviewers by hiding one known bug and checking they find it at the right line.

The intuition

It's a fire drill for AI reviewers. Instead of waiting for a real fire, I plant one - a single known bug in otherwise clean code - and check whether the alarm goes off at the right spot. Clean files are mixed in too, because an alarm that shrieks at burnt toast every morning is as useless as one that never rings.

What the numbers mean

The reviewer caught 9 of 10 planted bugs, scored at the planted line within two; the tenth is a deliberate blind spot shown as a red miss, in public. The new best-of-k study is a wind tunnel rather than a road test: nine copies of one reviewer, each dealt its own synthetic error stream, independent by construction and so the friendliest case there is, vote on findings and climb exactly as the jury arithmetic says they must. Matching the math proves the vote counting works, not that any reviewer got smarter. The controls carry the real lesson: nine copies sharing one error stream gain nothing, a panel that's wrong more often than right votes itself worse, and real reviewers land somewhere between those lines.

What this demonstrates

I test the testers - and know when a committee adds judgment and when it's just an echo.

Ask me: "When does adding more AI reviewers stop helping?" - When they share the same blind spot, or when they're wrong more often than right. My controls showed both: nine copies making identical mistakes stayed exactly flat, and a bug no copy can see stayed missed at every panel size. Voting multiplies the judgment you feed it - it can't create judgment that isn't there.
Judgment

adversarial-chambers

Splits AI decisions into a proposer, a professional skeptic, and a judge - then cross-examines the judge.

The intuition

Think of buying a used car. One person falls in love with it, so you bring a mechanic friend whose only job is to find what's wrong, and a level-headed third person makes the final call. This project wires an AI the same way - and keeps a notebook of every car they walked away from, and why.

What the numbers mean

In one real case, 3 of 4 verdicts flipped when nothing changed but the judge's standing instruction. Same facts, same arguments, different outcome - meaning the referee, not the evidence, was deciding the game. The flip test exists to catch exactly that before anyone trusts the verdicts.

What these numbers are not. That is a single case study, clearly labeled, and not a benchmark - four verdicts is four verdicts. The other figure you will see quoted, 0 of 8 against 3 of 8, comes from two judges I wrote myself to behave that way, so it is true by construction. Together they show the instrument lights up when a judge is instruction-sensitive. Neither one shows that any judge you would actually deploy passes, and I would rather say so than let a good number travel further than it has earned.

What this demonstrates

A verdict is only as trustworthy as its judge - so I built the tool that puts the judge on the stand.

Ask me: "What is the ledger of killed ideas actually for?" - Rejections are data. Every time the skeptic kills an idea, the reason goes in the ledger, so later we can ask whether the graveyard was fair. If you never write down why you said no, you can never discover your no's were wrong.
The improvement loop

memory-repair-lab

When the AI's memory fails, the failure gets written down, fixed, and re-tested on sealed questions.

The intuition

Before touching anything, I wrote a set of test questions and sealed them in an envelope. Then came the fixes - each one citing, in writing, the exact failure it repairs, like a mechanic's work order. Only at the end does the envelope open. If the score climbs on questions nobody could have studied for, the improvement is real.

What the numbers mean

The sealed questions went from 8 out of 10 to 10 out of 10 after the fixes. The proof it wasn't luck: a placebo arm - the same amount of effort, aimed at nothing in particular - stayed flat. The targeted fixes did the work; the busywork alone did nothing.

What this demonstrates

I run self-improvement like a clinical trial, sugar pill included.

Ask me: "Why bother with a placebo arm on a memory system?" - Because effort alone can look like improvement. If any tinkering would have raised the score, my fixes deserve no credit - so I spent the same effort on generic tinkering as a control, and it moved nothing. That flat line is what lets me say the 8-to-10 jump came from the fixes, not the fuss.
The playable portfolio

AI Quest: SEV-1

A retro browser game where the core party, the bosses, and the numbers come from the real projects.

The intuition

Think of a chef who, instead of handing you a résumé, cooks you dinner from it. This 8-bit game plays in your browser: four of the party are the labs above and the rest are their own people, each boss is a real way AI systems fail - runaway retries on the main road, then made-up facts and smuggled instructions deeper in - and every number a character quotes can be checked against the repos, including the balance figures, which now ship with the simulator and its run logs. Clear every arc and a post-credits finale opens: THE GRAND STAGE, a live industry exercise where a licensed red team is the final exam and the ending is earned, not granted - the same idea the mutation lab applies to the game's own tests.

What the numbers mean

The game quotes the portfolio's real results - the 0.61× memory cost, the 9 of 10 bugs caught, the 8-to-10 sealed-envelope jump - as in-game facts. Nothing was invented for drama. If a boss fight cites a number, that number lives in a repo you can open.

What this demonstrates

I can teach hard ideas to anyone - and I kept the fun on the same honesty rules as the labs.

Ask me: "Why put a video game in a job portfolio?" - Teaching is the sharpest test of understanding: if I can make prompt injection make sense to someone fighting it as an 8-bit boss, I actually understand it. The constraint made me better, too - every number in the game had to trace back to a published claim, so even the entertainment is auditable.

Working vocabulary

Best-of-k / test-time compute

Ask the AI several times and take the best or majority answer - spending extra thinking time instead of waiting for a smarter AI.

Ask nine doctors who have not talked to each other and take the majority - but if all nine trained at the same school with the same blind spot, nine opinions are really one.

Unhobbling / product overhang

Getting more out of the AI we already have by removing the clumsy ways we use it - no new model needed.

Like discovering your microwave had a defrost button all along - the machine didn't get better, you finally used what it could already do.

Mutation testing

Deliberately planting a known bug to check whether your tests - or your AI reviewer - actually catch it.

A safety inspector hides one fake violation in the factory to see if the auditors flag it - miss the plant, and they're missing real ones too.

Drift gate

An automatic check that re-runs every published number on each change and blocks the change if a claim stops being true.

Like a restaurant re-weighing every advertised portion each morning - the menu isn't allowed to say eight ounces on a day the scale says six.

Placebo / control arm

A comparison group that gets the same effort but not the real fix, so you can tell what caused the improvement.

Half the patients get the real pill and half get a sugar pill - if both groups improve the same, the pill wasn't the reason.

Held-out sealed test

Test questions locked away before any fixes begin, so nobody can tune the system to the test.

The final exam stays in a sealed envelope while you study - you can't rehearse the exact questions, so your grade actually means something.

Calibration

How well confidence matches reality: when a system says it's 90% sure, it should be right about 90% of the time.

A weather forecaster who says "70% chance of rain" is well calibrated if it rains on about seven of every ten such days.

Defense-in-depth (against prompt injection)

Several independent safeguards layered together, so a malicious instruction that slips past one still runs into the next.

A bank doesn't trust the front-door lock alone - there's a guard, a vault, and an alarm, so one picked lock doesn't empty the safe.