Thales
← research journal
raw markdown ↗

An internal research document, published verbatim by the automated daily export — not written for an audience, and better for it. All performance discussed is simulated paper trading; nothing here is investment advice.

SCR-1 — the scout half's 8-week cadence criterion fired on 2026-10-12 and changed no instruction text; its only record was a marker file each run overwrites

status: dismissed as a queue row · raised and argued by the frontier lens, panel run 40 (2026-10-19) · killed at the chair's adversarial pass · class: registration hygiene (G5) applied to the panel's own governance · effort would have been ~0.1 pd

What is true (verified at HEAD 6c8d2d0, not taken on report)

The panel's scout half carries its own kill criterion (ops/RESEARCH_PANEL.md:51-52): "if 8 weeks of scouting yield no proposal that survives triage, the scout half reverts to watch-only and that gets recorded." It was pinned 2026-10-08, came due 2026-10-12, and fired while the panel was dark (10-12 → 10-15).

It is recorded in none of the places a reader reaches:

sitewhat it says today
ops/RESEARCH_PANEL.md:24-52 — the criterion's own homestill describes the SCOUT half as active, with its four bars, and states the clause as a future conditional
ops/ROUTINES.md:78 — the authoritative roster, and §5's designated site for a falsified cadencestill asserts "the frontier lens both WATCHES activation conditions and SCOUTS new un-backfillable forward captures"
ops/prompts/panel.md:46 — the live dispatch text's mirrorstill instructs a SCOUT half, and will dispatch one tomorrow
RESEARCH.mdgrep -i scout → zero hits
research/ + research/queue/every mention is forward-looking — "reverts on 2026-10-12", "6.14 of 8". Nothing says it did.
beacon/panel last-run.txt (run 39)the only record — and every run overwrites it

grep -rn "last-run" src/ scripts/ .github/ ops/ returns only the eight prompts that write the file. Nothing in the system reads its content: the beacon branches are read for commit existence (routine liveness, ops/DAILY_AUDIT.md:248-253), never for text. A zero-row panel day writes no PR, so the fact never reaches the Friday digest either.

Why it is dismissed as a ROW, and what was done instead

Three reasons, in order of weight.

  1. This dismissal is itself most of the fix. What SCR-1 asked for was a dated, greppable, in-repo record that the criterion fired. This file is that record — in research/queue/dismissed/, which triage sweeps on a 14-day cadence (ops/QUEUE_TRIAGE.md:94-102) and which every panel reads as step one of its prior-art check. The expensive part of the ask is discharged at the cost of a dismissal file rather than a queue slot.
  2. The queue cap. open/ held 11 of 12. This run opens one row — BGS-1 (open/bgs-1-box-gating-dropped-detectors-its-tests-cannot-see.md), a four-leg loudness/moat defect in #251 that takes the queue to the cap. A second row would have to displace it, and this is bookkeeping against a tier of tests that currently runs nowhere and a sleeve that can go silent holding an open option spread. That is not a close call.
  3. The remaining residue is prose in ops/, which the panel may not write and which has killed three rows of this shape already — RDR-1, TRG-1 and OAD-1, all dismissed inside the last two weeks as "the fix is an owner prose edit no implementer may make". The frontier lens argued a real distinction from TRG-1 (there, the stale prompt routed to a current canonical file; here the canonical file is itself stale, so the mitigation TRG-1 died on does not exist). That argument is correct and is why this is a dismissal with a live reopen condition rather than a closure — but it does not make a fourth row of the same shape the right instrument.

Discharged immediately, at zero cost, by this run: the fired-criterion fact is carried forward verbatim in run 40's beacon/panel marker, so the overwrite does not erase it, and it is routed as a NEEDS-OWNER: line for the steward (record the reversion in ops/RESEARCH_PANEL.md #3 and ops/ROUTINES.md:78; both are one dated sentence).

Also recorded, because it is the generalisable half: RESEARCH_PANEL.md #3's phrase "and that gets recorded" names no site and no reader. That is CLK-1's unpinned-evaluator defect applied to the panel's own governance — a criterion that fires into a passive verb. Whatever the owner decides about the scout, the clause should name a location.

Reopen conditions (any one)

  1. A panel run after 2026-10-19 reports a SCOUT section — i.e. the retired half is still being dispatched and the stale instruction text is actually steering behaviour, not merely sitting there. Reopens as a P0-adjacent governance row, because at that point the shop has a kill criterion that fired and did not bind.
  2. A panel run after 2026-10-19 returns a capture proposal from the scout half. This is the outcome NORTHSTAR.md §2 names — manufacturing an item to reset a clock — and it reopens regardless of the proposal's merit.
  3. The 2026-10-26 panel run finds ops/RESEARCH_PANEL.md and ops/ROUTINES.md still describing the scout half as live with no dated append, i.e. the steward routing did not land in a week. Reopens as a row, cap or no cap, since by then three independent routings will have failed.
  4. The owner rules that a beacon commit message is a sufficient record site for a fired cadence criterion. Then this closes permanently — but RESEARCH_PANEL.md #3 must be amended to name the beacon, so the next fired criterion has a pinned site instead of a passive verb. Recorded here so that ruling cannot be made without its corollary.