Back to the consoleDocumentation
Documentation
The reference library behind the console. Every page here describes something the product does — the format it signs, the runs it measures, the rules it operates under — and every number on them is traceable to the run that produced it.
Start here
What this is and how the pieces fit together.
The evidence format
The signed document at the centre of the product — its schema, its provenance, and the proofs that make a score checkable by someone who was not in the room.
Evidence PassportLIVEThe signed evidence format itself — every field, what it means, and the twenty-five published passports you can check against it.SpecificationsLIVEThe normative documents and JSON Schemas: the passport schema, the policy and envelope schemas, and the submission contract.Trust GraphLIVEEvery published number wired to the run, the dataset and the code that produced it, as something you can walk.Score proofLIVEA held-out score proven correct without publishing the held-out set or the answer key.
What we measure, and how
The benchmarks and the stress tests behind every number the console shows you.
BenchmarksLIVEThe scored runs, by domain — what was measured, on which models, against which datasets.The instrument fleetLIVETwelve generators, each carrying the physics of one instrument, each proved against an invariant a model without that physics cannot satisfy.Synthetic variationPROTOTYPEWhat happens to a model's accuracy when the input drifts away from the data it was trained on, across 2,772 measurements.Shift labPROTOTYPEThe same phenomenon at arm's length: one model, one slider, and the accuracy floor you hit by moving it.Wearable signalLIVEThe raw-fidelity layer — what is lost between a sensor and the summary statistic most platforms store instead.Live evaluationPROTOTYPEHow a benchmark stays honest after publication: a rotating held-out set, sealed and scored on a cycle.
Operating the standard
The rules the standard runs under, and what has happened when it has been used.
Trust RuntimePROTOTYPEThe evaluator behind the deployment gate: how a policy is written, and how a proposed use is tested against it.GovernanceDESIGNEDWho can change the standard, how a dispute is settled, and what stops the people who publish the scores marking their own work.Case studiesLIVEWhere this has been applied, what it caught, and what it cost to run.