Vitrine VI
The census that hid its own repair
This is the sharpest structural fault the project found in itself, and it is not that a check failed. It is that a check succeeded in a way that made its own repair invisible.
What was being measured
Each piece of the Turkish edition carries a marker saying whether an independent verifier has read it. A marker is only worth what measures it, so a census was armed to read the markers and report which pieces carried evidence of a verification and which did not.
How the limb was blind
The census admitted any bracket containing a date. Every bracket this corpus writes opens with its date — so a ruling bracket read as a verifier’s bracket, and the two could not be told apart by the thing whose whole job was to tell them apart.
At one sitting a single copy-pasted correction went into exactly fourteen of the front-matter lines. Nine of those fourteen carried no other mark. The census was armed six sittings later and read all fourteen as marked — so the backfill that existed precisely to write the missing marks never named those nine.
The limb did not merely fail to convict. It hid the work that would have closed it.
translation/AUTHOR-BRIEF-2026-08-29.md, the record of what was ruled while the author was away — which is where this whole vitrine comes from.
And it ran as a bare report, so nothing downstream could ever go red on a false verified mark.
What came back when it was tightened
It convicted nine. Each of the nine was then adjudicated against the artefact and not against the record.
- Convicted by the tightened limb
- 9
- Upheld — genuinely verified, the mark never written
- 6, each now carrying a dated bracket that names its own evidence: a map row, a register clause, or the verification-driven ink itself
- Demoted — no artefact evidence at all
- 3 · DS-007, DS-028, DS-040. What existed for them was a log row with batch arithmetic, and a log row is not an artefact.
- The state at this build, recounted from the front matter
- verified 61 · carried 1, over 62 files
- The criterion for that figure
- the
status:key of the front matter of every file intranslation/stories/, read at this build. The census that produced the fault istools/vh_tr_checks.py; the adjudication of the nine is intranslation/AUTHOR-BRIEF-2026-08-29.md.
All three were then read by an independent adversary and closed. One ink change survived; the rest were refused with warrants. The figure above reads the same as it read before the census. It means something it did not mean before the census.
The rejected alternative, filed
Uphold all nine on the log’s batch arithmetic. Rejected because a false verified mark is the worst defect available to this project, while a piece marked unread that a verifier had read after all costs one re-reading.
The brief files that ruling in one sentence, and the sentence carries a clause about the book which this page does not: a vitrine may state what was ruled and why, and it may not repeat the reading the ruling leaned on.