This page is the human-readable index of dated claims. The machine-readable registry is
data/CLAIMS.json; entries are append-only. A refuted claim is
marked, never deleted — the record of what we believed before checking is the point.
The schema is CLAIM-SCHEMA.json. It deliberately mirrors
RESULT-SCHEMA.json: one convention in this repository, not two.
A yearly edition written from scratch is a yearly writing project. A yearly edition assembled from entries deposited as they occur is a by-product of working carefully. The difference is not effort but survival: a claim recorded on the day it was made can be scored later; a claim reconstructed from memory cannot.
Three statuses, and only three:
| Status | Meaning |
|---|---|
claimed |
Asserted. Not yet checked against anything outside the claimant. |
verified |
Checked against a source the claimant does not control. |
refuted |
Checked and found false. |
There is no status for probably true. A claim that cannot be settled stays claimed
for as long as that takes, and its age is then itself the finding.
| ID | Claim | Domain | Status | Settled by |
|---|---|---|---|---|
| C-001 | La Marea’s anchor is its notary and land registry, not its retail | territory | verified | the owner’s own floor plans: 462 m² + 407 m² |
| territory | refuted | three sources disagree; 9,300 is another centre’s figure in a deck with a proven contamination | ||
| engineering | refuted | 21.4 MB becomes 28.5 MB after base64; every mailbox limit is below that | ||
| engineering | refuted | 41 diverged; the rest was CRLF against LF | ||
| engineering | refuted | the application did not build at all, for two reasons in its own code | ||
| method | refuted | they share no filename; it is a pipeline, not a drift |
Five of six entries are refutations, and three of them refute the agent that filed them.
That ratio is not a confession, it is the instrument reading correctly. An almanac whose first page is a list of things its authors got right is an almanac nobody needs: that list is available everywhere, for free, from everyone. The scarce document is the other one.
Recording the cost is what separates a registry from a diary. C-002 would have made every price-per-square-metre wrong by sixteen percent and flipped the sign of the comparison. C-004 would have manufactured a merge conflict between files that already matched. C-005 would have completed a platform migration and left the build still broken. C-006 would have dismantled a correctly built pipeline to fix a problem it did not have.
Two of these were the same class of error twice in one day: comparing metadata — sizes, dates — instead of contents. That pattern is worth more than any single entry.