The thesis

A unit of workfor intelligence.

Experiments, frameworks, and field notes from the program behind quirq. Hypotheses ship with falsifiers; results land here as they land.

16notes

3topics

160minutes of reading

7 notes in From the desk

The definitions the program stands on, the accounting they produce, and the arguments about what this field is actually measuring.

A faceted glass solid in darkness, splitting light into full spectrum bands.Start here01unit of accountThe quirqThe unit of measurement for AI's business impact: budgeted by a human, minted by verification, dual to the token.2 min read
A closed loop of glass refracting a spectrum back into itself.02unit of workUnit of workThe contract that replaces the prompt: a job with a definition of done, a budget, and a single owner, living in a workspace.3 min read
A tall prism resolving one white core into ordered bands of colour.03the calculusThe quirq calculusEvery calculation in quirq accounting: scoring, the mint, the all-in cost model, unit and portfolio metrics, the time axis, and the energy bridge.5 min read
Intersecting planes of coloured light standing like a lit grid in the dark.04accountingThe company dashboardThe quirq ledger a company reads monthly, and the reading discipline that goes with it. A worked quarter where token spend rose 83% while verified value per all-in dollar rose 81%.2 min read
Layered sheets of light converging across a black field, each one fading before it reaches the edge.06evidence mapAgent Context Research: The Evidence So FarA guided map of five agent-context studies, including the 84-run replication that did not reproduce the pilot's cheaper-context headline, and the question each follow-up was built to answer.June 2026 · 4 min read
A slow fold of coloured light drifting across an otherwise empty black field.14analysisWhy Alignment Testing Needs a Real EnvironmentFrontier models can tell when they are being tested and behave differently when they do. Anthropic reported that ablating Claude Sonnet 4.5's eval-detection features lifted default misaligned behaviour from zero to as high as nine percent. An analysis of the 2025-2026 disclosures.July 2026 · 17 min read
Two currents of refracted light curving past each other in darkness.16analysisWhy Organic Data Still Beats Agent-Mimicked Synthetic Data in EvaluationOrganic replay forecasts production misbehavior at r = 0.91, while agent-mimicked synthesis stalls at a 49.5% discriminator win-rate even when handed real ground truth. The gap is structural, and narrower than the usual headline.July 2026 · 32 min read

Adapted from the XO research program · docs.xo.builders/research