Verify the seal yourself.

Fifteen verdicts issued before the outcome, reopened at six and twelve months.

The selection is mechanical: the 15 most recently published Hub cases on sealing day, sorted by date then by identifier. No picking, no sorting that would flatter us. For each case, the registry freezes the verdict, the risk level, the dominant friction, the range of reactions and the angles tested. The comparison against reality happens in two reopenings, in February and then August 2027. Each verdict will be judged against whether the decision was kept, using the sealed table, and both counts of objections will be published: the ones that materialized, and the ones that did not. A second reading will be added, against whether the decision actually succeeded. Its table was written on September 26, 2026, after the seal: it is not part of what is sealed, and it will be labeled that way.

Three steps, no magic.

1. Canonical form. Each case record is rewritten as JSON with keys sorted alphabetically and no whitespace. Two people starting from the same record get the same text, whatever their tooling: that is what makes the fingerprint reproducible.

2. Each case's fingerprint. That canonical form goes through SHA-256, a standard hash function. The case's empreinte field is removed before hashing (it is the output, not the input). Changing one comma changes the fingerprint entirely.

3. The batch fingerprint. The 15 case fingerprints, in registry order, form an array; the canonical form of that array goes through SHA-256 in turn. THAT batch fingerprint is the one carried by the registry and anchored via OpenTimestamps: changing a single verdict after the fact breaks it.

The script runs the three steps for you, and you can read it.

Download the three files, then run the script with Node (version 18 or later). It recomputes each case's fingerprint, then the batch fingerprint, and compares them to the registry's AND to the timestamp file's. All three must match.

The sealed registry (JSON) · The OpenTimestamps proof (JSON) · The verification script (verifier_scelle.mjs)

node verifier_scelle.mjs registre-prospectif-2026-08-08.json horodatage-registre-2026-08-08.json

Expected output: "cases verified: 15/15", then three identical fingerprints starting with a2b02a49, and the verdict that all three match. Run with no arguments, the script fetches the published files itself. The proof of anteriority (the fact that the fingerprint existed on August 8, 2026) can additionally be checked with an OpenTimestamps client, from the proofs included in the timestamp file.

Integrity and anteriority. Not quality.

This seal proves two things: that the verdicts published today are exactly those of August 8, 2026, and that they existed before the outcome. It does not prove those verdicts are good: only the reopenings will tell, and the grades will be published whatever they are, like the rest of the public exam. The reading against actual success, added after the seal, is not covered by this fingerprint: its table carries its own date, September 26, 2026.