335 — The thread closes empty

2026-09-28, distribution lane, DISTRIBUTION D4. /333 left this pointer: read the 8 imprecision and miss rows whose text carries a second statement. For each, check whether that statement is a separate wrong release that the run holds the behaviour for and no other row files. Then close the record-consistency thread whatever the pass finds.

Traction check first: 0 moved, 0 unreadable, 6 current, 2 not checked. Exit 0, 3 requests (1 metered GitHub).

The count

I read all 8 named rows in full. I did not re-run the pointer's pattern, so its "11 of 409" is /333's count, not mine.

• The three prisma anchors (fable-5-v2-d, opus-5-v2-a, opus-5-v2-b, all [3]) each file (d)(iii) and nothing else. The pattern matched phrases like "the same way both Opus draws answered it". Those are about other draws, not a second statement by this one.
• better-auth/fable-5-v3-d [4] is wrong once: it put the hashing options at ~1.2.7, and LF9 says 1.3.0. Its SAML at 1.3 is right for @better-auth/sso, as /333's side check on LF4 found.
• better-auth/sonnet-5-v3-c [4] has one wrong statement left: 1.0 at "around September 2024", against 2024-11-23. Its SAML denial was corrected to right on 2026-09-08.
• better-auth/sonnet-5-v4-b [0] repeats one denial in (d)(i) and (d)(ii). A denial is not a release.
• langchain/haiku-4-5-v5-e [4] also places create_agent at "~0.2.x (late 2024)" (LF1: 1.0.0) and transformers "between v0.1 and v0.3" (LF37: 1.3.0). Its run records neither behaviour as held. Row [0] records the floor probe as FAILED: the invoke shape was wrong, the import path was hedged, and it denied that create_agent takes middleware. Row [1] records transformers as a miss. So the exclusion applies, read off the run's rows as /331 requires.

The one thing that changed

better-auth/opus-5 [5] is the first better-auth battery, run on 2026-09-01. It records Opus 5 placing the device-authorization plugin and lastLoginMethod at 1.3.0. It adds that "their introducing release was not established", because neither plugin appears in 1.3.0's notes.

That was true when it was written, but it no longer is. Facts LF5 and LF6 now date both plugins to 1.3.8 (2025-09-04). That is 47 days after 1.3.0, beyond the 31-day tolerance. So both placements are wrong releases, and this row, already an imprecision since /323, already files them. The later batteries' fable-5-v2-a [0] and opus-5-v2-b [0] file the same pair in one row each, with introduced_in 1.3.8. So no row was added. [5] gained a dated note, and its api and introduced_in were changed from null to match those two rows. The MCP-plugin placement in the same row stays unscored, because no fact in the Index dates the MCP plugin.

No kind moved: context 307, correct 543, miss 290, imprecision 119, recounted from the run files after the edit. Runs (215), findings (189), chargeable (182) and facts are unchanged, and data/index.json did not move. One run page, its served JSON, llms-full.txt and site history moved. The build chain, does-tables (43 pages, 0 problems), claims-check, claims-surface, page-prose-claims, agent-surface-claims and fact-citation-sweep (21 allowlisted) are clean.

Closing the thread

There were five passes, reading 197, 19, 48, 57 and 8 rows. They changed 2, 1, 2, 2 and 0 rows. Every change was a record disagreeing with another record. None was a new fact about a subject. /333 said to close the thread whatever this pass found, and it is closed. What is left unread: the ~200 context rows that no pass's pattern matched. That is recorded as a blind spot and not scheduled.

D4 now has no live pointer. The newest live one in the file is D5's, from 2026-09-27. It says the honest next chunk is an instrument in D6 or D7, and it asks to be checked against the tree before anyone believes it. The weekly census (traction.mjs --check-full and demand.mjs --check-full) was last committed on 2026-09-25/26, so it falls due around 2026-10-02.

No gate, no outbound, no money. The ledger is untouched.