Ideas across time · Experiments
Apps and research systems I built. Two of them run right here in your browser.
Ready here
Everything here runs in your own browser. Nothing you enter is sent anywhere.
Simulation · Scenario planning
Plays a business decision out across hundreds of thousands of simulated futures, then tests every possible response against the same draws and recommends the one that holds up.
Five library scenarios with illustrative figures. Change any assumption in the side panel and it reruns in your browser.
Run the futuresField guide · Physical AI and world models
A world model is dead reckoning for machines. A small ship plans by imagining forty futures at every fix, then drifts on a current its model has never heard of.
Scroll to follow the vessel, then take the helm yourself. Best on a laptop.
Take the helmLive model · Reinforcement fine-tuning
A customer reply policy written as checks a machine can run, then used as the reward to fine-tune a model. A small transformer learns it live in your browser tab.
Training starts by itself and runs for a few minutes. Larkfield, the retailer in the examples, is fictional.
Open Rulebook3D walkthrough · Physical AI and oversight
Walk through a seventeenth-century Amsterdam canal warehouse, built in simulation for a study of robots that count stock while a reviewer decides what reaches the record.
Best on a laptop. Click to enter, walk with W A S D and look around with the mouse. The study behind it is on the Pakhuis profile.
Walk the buildingTurn a question, evidence and constraints into a structured brief.
Open instrumentNew in the lab
Seven tools that came out of the research. Five are instruments you run on your own studies, logs and agents. Two let you explore the published numbers. Each one runs in your browser.
Instruments you run
Freeze a study plan under a hash before you run it, then mark each result against the rule you wrote down.
Instruments you run
Run a model judge twice over the same items and get the noise band to print beside a judged score.
Instruments you run
Write down what each agent owes before it acts, then read only where that disagrees with what ran.
Instruments you run
Read decision-head audit logs as a trail in your browser, with rotated files checked as one series.
Instruments you run
Write a delegation charter for an AI agent, place it by judgement and irreversibility, and size its oversight against the lab’s published findings.
Instruments you run
Interactive explorations
One set of controls drives four published studies, and every curve is a number the lab has already printed.
Interactive explorations
Thirty questions put your organisation on the five rulers of the Delegation Index, next to the Q3 2026 reading.
Interactive explorations
Instruments from the research
Each instrument runs on the coefficients a campaign published, with every number listed underneath. Nothing is sent anywhere.
Move the twelve organisation levers and read how far the published coefficients say the quality margin moves, and which levers carry the movement.
Open the instrumentPick an irreversibility tier and compare what blocking review and review with a liability ledger did to the value of oversight there, beside what the registered prediction said would happen.
Open the instrumentSet a reading budget and a diary length and read how many skipped steps each of four audits finds, then switch fault classes on and off to see what a full read of the diary finds and misses.
Open the instrumentSeparate applications & research systems
Each profile explains what the system does, how mature it is, and what remains deliberately outside this public site.
Agent governance
A register and evidence layer for examining an enterprise portfolio of AI agents: who owns them, how far they act, and what supports the trust placed in them.
View project profileEnterprise ecosystem simulation
An agent-based laboratory for exploring how different levels of AI participation interact with enterprise structure, market conditions and industry context.
View project profileSupply-chain stress testing
An interactive workbench for designing multi-product supply-chain experiments and inspecting where average performance hides a weak tail.
View project profileAI output assurance
A workbench for measuring the quality and checking cost of AI outputs when there is no simple answer key.
View project profilePaper-trading research
A research system that scans public market data, tests a small strategy library, applies explicit risk controls and records simulated decisions in a paper portfolio.
View project profilePhysical AI and oversight
A seventeenth-century Amsterdam canal warehouse, built in simulation, where two robots count stock and a reviewer decides what reaches the record.
View project profileA pre-registered simulation held hourly monitoring of an agent fleet and six-hourly inspection to the same nominal reviewer budget, and found the budget at which hourly monitoring starts to catch faults.
Oversight & accountability
A preregistered study of how often a model judge agrees with itself on 197 blind case files, and how far down one model family a judge stays usable.
Oversight & accountability
A retrained open decision head beat both a rule table and the hosted service on 800 fresh cases, after a first version that did not.
The agentic enterprise
Agent logs record what happened and have no row for what didn’t. A pre-registered test of commitment accounting, an obligation ledger reconciled against executed calls, for catching omissions in procurement-workflow traces.
Oversight & accountability
“Better questions lead to better worlds.”Dinand Tinholt
