Verify the seal yourself.
What it is for: proving that the conclusions Kapari published were not changed after the fact. On August 8, 2026, we froze those of our 15 most recently published cases, before we knew how each story would end.
Who needs it: anyone who wants to check our results without taking our word for it, such as a client evaluating Kapari, an auditor or a journalist.
How to do it: download three public files and run a small program; if it reports that all three fingerprints match, nothing has moved.
See you in 2027
We will reopen these cases in February and again in August 2027 to compare our conclusions with what actually happened, as we did in the public exam.
For the curious: the technical detail and the step-by-step check
What is sealed
The selection is mechanical: the 15 real cases tested and most recently published on the day of sealing, sorted by date then by identifier. For each case, the registry freezes the conclusion, the risk level, the main source of friction, the range of reactions and the angles tested. In 2027, each conclusion will be compared with whether the decision held, using the sealed table. A second reading will also compare it with how the decision actually turned out; that table, written on September 26, 2026 after the sealing, is not part of the seal.
The exact recipe, in three steps
1. The standard form. Each case is rewritten as JSON, keys sorted alphabetically, with no whitespace at all. Two people starting from the same case get the same text, whatever tool they use.
2. The fingerprint of each case. That text goes through SHA-256, a standard fingerprint function. The case's empreinte (fingerprint) field is removed before computing (it is the result, not the input). Changing a single comma in the case changes the fingerprint completely.
3. The fingerprint of the batch. The 15 fingerprints, in registry order, form a list, which in turn goes through SHA-256. That batch fingerprint is the one written into the registry and dated with OpenTimestamps: changing a single conclusion after the fact breaks it.
Check it with one command
Download the three files, then run the program with Node (version 18 or later). It recomputes each case's fingerprint, then the batch fingerprint, and compares it with the one in the registry and the one in the timestamp file.
The sealed registry (JSON) · The OpenTimestamps timestamp (JSON) · The verification program (verifier_scelle.mjs)
node verifier_scelle.mjs registre-prospectif-2026-08-08.json horodatage-registre-2026-08-08.json
Expected output: a line confirming that 15 of 15 cases were verified, then three identical fingerprints starting with a2b02a49, and "les trois concordent" (the program speaks French: "all three match"). Run with no arguments, the program downloads the published files itself. The date (proof that the fingerprint really existed on August 8, 2026) can also be checked with an OpenTimestamps client, using the proofs included in the timestamp file.