Appendix A · Dossier Method and reproducibility ← Back to the findings

Five controls, one piece. Verified, then sealed.

A matte aluminium balance with two pans at exactly the same height; the right pan is the tool's polished onyx.

This page shows what the Dossier is made of: which reports it reads, which five controls it runs on each, what its own public measure covers, and where it refuses to guess. The Dossier is the keystone of the suite: it proves the chain, or it names what is missing.

Five controls, run in order, refusals named

Every report the Dossier reads passes the same five controls: present (a readable report from a named tool), sealed (the fingerprint it carries is the fingerprint of its content), signed (the public record it cites carries the house signature), fresh (measured within the declared rhythm; due under two rhythms, stale beyond), consistent (one dataset, one seal; an unknown tool version is named, not guessed). The registry derives the list from code: an absent control is computed, never recited.

Each verdict carries its reason in words. A report that cannot state its measurement day is not called stale: it is named as unjudgeable, which is a different and worse fact, and the Dossier says it.

Verify the controls

The suite exercises every control in both directions, refusals included.

run it yourself

npm testthe five controls and their refusals, with the rest of the suite

Where it livessrc/controles/index.ts:15 · src/controles/index.ts:24 · src/controles/fresh.ts:5

Its own public measure: the suite, judged by its keystone

The Dossier's public record is the suite judging itself: it runs on the public records of the other tools and prints coverage, seals, signatures and freshness as it finds them on the day of the measure. On day one that meant three questions of four covered with the fourth named missing, two seals recomputed, and zero verifiable signatures, printed rather than waved through. A keystone that flattered its own house would be worthless to a reviewer.

Verify the record

The public record is sealed; recompute the fingerprint with the repository's own tool.

run it yourself

npm run sceller -- releve-public.jsonsays: already sealed, and the seal matches

Where it livessrc/measure.ts:93 · src/sceller.ts:42

The rhythm is declared, and age is counted against it

Freshness only means something against a rhythm, and the rhythm is yours: ninety days by default, named in the record as a declaration, changed with a flag an auditor can see. The record photographs the settings it was measured under, so two Dossiers can be compared knowing whether the rule moved between them.

Verify the settings

The record carries its settings block, rhythm included.

run it yourself

python3 -c "import json; r=json.load(open('releve-public.json'))['reglages']; print(r['rythmeJours'], 'days, stale beyond', r['staleApres'], 'rhythms')"

prints the declared rhythm the verdicts were judged under

Where it livessrc/controles/fresh.ts:22 · src/measure.ts:98