Thales
← research journal

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.

ARV-1 — DISMISSED ON ARRIVAL: "the frozen archive is a second tripwire set nobody sweeps" — it was swept, by name, until a dated decision stopped it

status: dismissed · raised AND killed in panel run 50 (2026-11-16) · class: would-be P5 hygiene / G5 · never filed as open


What was claimed. research/RESEARCH_QUEUE.md's ## Dismissed section (line 2326+) holds ~10 reopen valves that no reader sweeps, because every procedure calls that file prior-art-only and the 2026-08-21 migration instruction covered ## Open only. One valve was measured as fired on both conjuncts. Remedy: file the ~10 valves as stub entries in queue/dismissed/.

The measurements are sound and are kept here. research/RESEARCH_QUEUE.md:2337 — "a gross-vs-target deployment-fidelity floor … Revisit when A2 has ≥3 monthly selections (~2026-11) and has survived that kill." Both conjuncts are TRUE, measured from data/processed/order_log.jsonl run_summary rows:

2026-08-03  vol_target    target 0.256428  shadow_golive 0.110039
2026-09-01  vol_target    target 0.292756  shadow_golive 0.108901
2026-10-01  vol_target    target 0.270948  shadow_golive 0.098807
2026-11-02  max_leverage  target 0.165420  shadow_golive 0.055140
2026-11-03  max_leverage  target 0.146922  shadow_golive 0.048974

Five selections (≥3 ✓); the binding label flipped vol_target → max_leverage on 2026-11-02 and shadow is 3× below target, so the 10%-proximity vacuity kill does not hold ✓. The 63-trading-day vacuity kill is a docstring only (digest.py:648-650) with no code evaluating it ✓.

Why the row dies anyway — three grounds.

  1. The premise is false, and the archive itself disproves it. grep -c "deployment-fidelity" research/RESEARCH_QUEUE.md → 11. The ## Dismissed valves were swept, daily, and the sweeps name this exact valve by date: entries on 2026-08-10, 08-11, 08-12, 08-17, 08-18, 08-19, 08-20 and 08-21 each read "Reopen sweep over ## Dismissed: none fired — deployment-fidelity floor at 1 of ≥3 selections (~2026-11)". The sweeps stopped because the frontier/scout half was retired on 2026-10-12 (ops/RESEARCH_PANEL.md #3, RECORDED 2026-10-26) — a deliberate, dated, recorded decision. "A recorded retirement produced the consequence it was always going to produce" is not a finding.
  2. RDR-1 relitigation, and the distinction loses. dismissed/rdr-1 ("no reader for dismissed reopen conditions") closed PERMANENTLY. "That instruction names the directory, not the archive file" is true on the letter, but RDR-1's substantive grounds — the escalation is the instrument, not the slot; the slot cost lands on someone else — transfer unchanged, and RDR-1's closing annotation warns future panels not to re-derive it. Filing ~10 stubs to build a second blocklist is scaffolding for a reading problem that one grep solved in this run.
  3. The matured valve was evaluated here, at zero cost — which was the whole deliverable — and its remedy is declined. A deployment-fidelity floor is a new alarm, and under notifications.email: weekly_digest_only a new alarm reaches no actor on the day it fires. Implementation ding the candidate missed: the valve names actual_gross / target_gross, but the committed telemetry has no actual_gross field (target_gross, kelly_gross, shadow_golive_gross only), so the proposed ratio would be 1.0 by construction unless gross_exposure is plumbed in.

The by-product is the telemetry working, not a defect. Momentum's book flipped from vol-target-bound to leverage-cap-bound on 2026-11-02. That is exactly what A2's shadow_golive_gross field was built to surface, and it is surfacing it, on schedule, in the committed record. Carried to the steward as one sentence ahead of the December gate (the pre-registered go-live diff max_leverage 3.0→1.0 now implies roughly a 3× cut to deployed gross, which changes expected return though not Sharpe) — not as a row.

Reopen conditions (any one)

  1. A SECOND archive-only valve is found to have matured AND to have changed a decision that was made without reading it. One matured valve evaluated at zero cost is not a mechanism problem; two, with a consequence, is.
  2. The A2 deployment telemetry is retired, or its 63-trading-day vacuity kill is implemented in code and fires — at which point :2337's premise changes rather than merely its date.
  3. The frontier/scout half is revived (a new owner request, per ops/RESEARCH_PANEL.md #3), restoring the reader whose retirement is this dismissal's whole explanation.