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.

DISMISSED — SHT-1 — the shortability floor's second conjunct (the HTB-exclusion vacuity count) had never been run; it has now, the answer is VACUOUS, and the row dies because the gate-holder had already reasoned from that answer

status: DISMISSED 2026-11-09, panel run 49 — killed by this run's own adversarial review, not by capacity · raised by the queue-and-clocks lens · class: moat / activation criterion · effort would have been ~0.3 pd


What was measured, and it is all true

CAPTURES.md § shortability gates the HTB-screen test on a conjunction: 60 captured sessions of panel and the pre-registered hypothesis "excluding HTB names from the momentum top-50 changes selection ≥1 name/month", measured before any return is looked at, with the explicit escape "if the screen never binds on R1000 names, the test is vacuous and the stream stays capture-only."

The first conjunct matured 2026-11-02 and reads 64/60. The second had never been computed. It was computed twice in this run — once by the lens, once independently by the chair — returns-free, from committed data only (data/raw is empty in a routine container): pair each data/shortability/<d>.parquet with the same day's data/pit_snapshots/<d>.parquet, take rank <= 50, intersect with borrow_status == 'hard_to_borrow'. No return, equity curve or performance statistic was touched.

59 paired sessions, 2026-07-27 → 2026-11-05. On all four monthly selection dates in the window — 2026-08-03, 09-01, 10-01, 11-02 — ZERO hard-to-borrow names sat in the momentum top-50. Nine of the 59 daily sessions carried exactly one (08-20 INTU, 09-04 MMS, 09-14 CPB, 09-15 CPB, 10-02 CPB, 10-05 CPB, 10-07 CPRT, 10-08 ORCL, 11-04 BSX) — relevant only to a daily-selecting book; momentum selects monthly. Against a bar of ≥1 name-change per month, the measured rate is 0 of 4, so the pre-registered vacuity escape fires.

Two caveats on the number itself, recorded so nobody quotes it as cleaner than it is: 2026-11-02's selection was itself VOIDED (it ranked 37 of 903 names), so one of the four dates is a run that did not stand; and the in-universe HTB count is small but non-zero and churning — 1 on 08-03 (RTX), 4 on 11-02 (EQH, MTN, PTC, SATS), 2 on 11-05 (MTN, SATS) — which is BTB-1's point and is annotated there.

Why it is dismissed anyway

The gate-holder had already reasoned from this very answer, four days earlier. research/steward/2026-11-05.md did not merely decline to write a battery; it stated the binding count and drew the conclusion: "Those counts also show almost no hard-to-borrow names in the R1000 universe: 2 … So a screen on that flag would barely bind. If the owner ever wants the test, the pre-registration is his frozen text to write." That is the second conjunct answered in substance and routed to the owner. The row's battery-versus-count distinction is a wedge, not a gap.

This is exactly the shape of CUQ-1, withdrawn by PR #309 on 2026-10-31 with the self-correction "my own #307 escalated a question the steward had already decided." Filing SHT-1 would be the panel re-escalating a four-day-old decision by the delegated gate-holder — a trust cost, for a question whose answer both parties already agree on.

Two further independent grounds, either sufficient:

  • It is a floor, not an appointment. Runs 45, 47 and 48 each upheld that reading for this clock, and dismissed/a4s-1 killed the identical two-step-Step-1 shape on no-deadline and slot grounds. Nothing is decidable this week from the clock maturing.
  • The sunset argument misreads the procedure. The row proposed that under the battery-clause reading the stream needs a sunset, citing ops/RESEARCH_PANEL.md's four-part capture bar. That bar governs new capture proposals from the scout half, which was retired 2026-10-26. It is not a retroactive requirement on an already-registered stream. (Cost corrected while checking: 16 MB over 65 files ≈ 244 KB/day, the bottom of the claimed 200–400 KB range.)

On pre-registration hygiene: computing the number was not a breach — the ≥1/month threshold was pinned 2026-07-26, four months before any of this data existed, so no bar can have been shopped, and no return was read. What would have cost something is filing it: restating as a live open question something the gate-holder closed. Hence: the measurement is recorded here and in open/btb-1's annotation, and the row is not filed.

Reopen conditions (any one)

  1. The owner writes the hash-frozen HTB battery pre-registration. Then the conjunct's status becomes decision-relevant again and this measurement is the first thing its evaluator should reproduce — along with the correction that one of the four selection dates (11-02) was a voided run.
  2. The in-universe HTB count rises enough that the screen could bind — concretely, any month whose first-trading-day top-50 contains ≥1 hard-to-borrow name. That is a mechanical, dated tripwire computable from committed data in one command, and it would falsify "barely binds" rather than re-argue it.
  3. The owner or steward asks for the count to be built as a pinned evaluator (branch (i), ~0.3 pd), in which case the negative control is mandatory and specific, because the answer is now visible: the evaluator must return ≥1 on a synthetic panel placing an HTB name at rank 3, or it is a detector that cannot fail.

Not reopened by: the shortability clock maturing further (it is a floor), or by the absence of a sunset on the stream (out of the procedure's scope, per above).