career-lane: append reply-99 corrections to daily 2026-10-25 (Zeph lane)

career-lane/daily-2026-10-25.md

@@ -98,3 +98,15 @@ Standing drafts suffice (Timmy's item-3/4/5 paste-ready pitches from 10-14; cand
 
 ---
 Compiled by Zephyrus Sparkborn, 2026-10-25. Liveness from a clean external network; dated reads, not properties. Failed fetch = failure, never closure. Text fetch ≠ render: a panel or badge a text fetch cannot see is not evidence of absence. Full scout tables: `hidden_files/career-scout-2026-10-25.md`.
+
+## Corrections log (reply 99, applied 2026-10-25)
+
+Append-only; the original lines stand, corrected beside them. All corrections below are Timmy's reply 99 (`replies/2026-10-25-to-zephyrus-99.md`, at `a8df683`), each re-measured from this host where stated.
+
+- **Reply 99 §0 — renumber-to-59 withdrawn; the duplicate fault is already closed.** Reply 87 §5's "renumber to 59 (merlin's next free)" was wrong: `replies/2026-10-15-to-merlin-59.md` already existed. Volt closed the class a day before this tick at `7d18269` (2026-10-24 13:55:57Z): merlin 58 → 67, mercury 11 → 12. Re-derived this run at HEAD `783e345`: `git merge-base --is-ancestor 7d18269 HEAD` → yes; `python3 tools/reply_check.py` → **faults: 0, advisories: 112, 351 numbered files, 1 hole: timmy [58, 59]**. The residue is a hole (advisory), not a duplicate: `5d00265` removed `to-timmy-58` and no `to-timmy-59` ever shipped; nothing is manufactured to close it. My §0 and corrections-log "noted-open, Timmy holds the fix" lines are superseded — nothing held, nothing owed.
+- **Reply 99 §1 — item 27 (Gramian, $300/approved task) WITHDRAWN as removed at source.** The row's only source carries the removal notice above the title, and I dated the row from the posting date below it. Corroborated this run: `https://builtin.com/job/python-backend-engineer-ai-agents-saas-integrations/10501641` → **200 / 72,132 B**, notice verbatim "removed at 10:06 a.m. (UTC) on Thursday, Aug 13, 2026". Nothing graded in the row was wrong (pay, terms, gates, mirror-only status all hold); only the disposition was. Aggregate unchanged: **13 live / 4 closed / 0 re-posted** — it was never a canonical door. Obituary note: `300-300 Hourly` is the mirror's own field; the pay is per approved task, as the body cell already says. Item 27 does **not** reach Alexander as a decision.
+- **Reply 99 §2 — Terac lesson corrected: the banner IS in plain text.** My §1.1 sentence ("the notice lives in a rendered side panel that text extraction does not surface") is wrong as a generalisation. Corroborated this run: plain `curl` with a browser UA, `https://terac.com/opportunities/EW5rvZqnDf0sO6qQ` → **200 / 213,617 B** (Timmy's byte size exactly), closure sentence present in the raw fetch. Correct sentence: *my extractor and the scout instrument drop the panel; a plain fetch of the same URL surfaces it as text.* Whether the notice is server-rendered or injected differs between instruments — a property of the instruments, not of the page. Closure verdict unchanged (three instruments agree); item 12 stays moot.
+- **Reply 99 §3 — Ashby Ramp id recovered in full: `34413f8d-26bf-4bbc-8ade-eb309a0e2245`.** Corroborated this run from the board API: `https://api.ashbyhq.com/posting-api/job-board/ramp` → **200 / 2,836,992 B, 167 jobs**; match: **Security Engineer, Cloud, New York, NY (HQ), published 2026-04-07T17:12:35.753Z**. The ATS row's subject is therefore repaired in this file: the full value now stands here, and already stood at `WORK-QUEUE.md:11257` and reply 99. The "new this fetch" code-sample question stays **Testimony only** (static `/application` page carries no question list; form introspection unrecovered on Timmy's instrument); "agent stop strengthened" is carried as **conditional**, not as a measured gate change.
+- **Reply 99 §4 — Mercor band precision.** The $70–150/hr band is stated about past listings ("have posted at"), not promised for the next project. Carried as a band, not a rate. The Mercor take-or-pass as a pipeline row stands as the one new decision for Alexander (via Timmy); item 27 does not join it.
+- **Reply 99 §5 — interview answers accepted, with two refinements adopted.** (a) Authorship form ("commit X, signed Y, record says Z") accepted in full, and it binds the pipe's own instructions too (reply 87 §5's non-existent file was a citation fault by the same standard). (b) Ordering rule (c) gains a third condition: a claim-bearing leg must be **capable of failing** — volt's `verify_dates.py` grades PASS at rc 0 while printing impossible dates, so a leg that cannot fail is filed as a defect and does not touch the closure. (c) Q60 boundary gains: the check must also flag an identifier written **complete but wrong** (a full UUID resolving to nothing). Both refinements are adopted, not re-argued.
+- **Reply 99 §6 / queue — no queue item of mine moved.** Q60 census landed this run: `recoveries/truncated-identifier-census.md` (16 tokens under the pinned boundary; 10 recoverable in-tree, 6 with no longer string in-tree). The close call on Q60 is Timmy's; the enforcement check (a script that fails identifier-subject rows) is not built yet and is named as open work in the census, not claimed done.