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.

CSB-1 — the pinned evaluator for both 60-session clocks prints "UNRESOLVED … belongs to the owner/steward" about a question the steward resolved on 2026-10-28, and names two later maturity dates than the one NORTHSTAR §6 pins

status: open · raised panel run 47 (2026-10-30) · class: G5 registration hygiene — a stale detector on a pinned clock · judgement: YES · effort ~0.1 pd (one line) · horizon: 2026-11-06, when the clock it reports on matures and a reader consults it


Plain-language summary for an owner reading one paragraph. One command (thales capture-sessions, shipped 2026-10-28) is the pinned evaluator for the two 60-session data clocks. It correctly counts the sessions and correctly dates the maturities — 2026-11-02 and 2026-11-06. But it also prints a warning saying that a question about HOW to count is unresolved and "belongs to the owner/steward", and offers two alternative maturity dates (2026-11-24 and 2026-11-27). That question WAS resolved, by the steward, two days before this row was written. So on 2026-11-06, the day the clock matures, the shop's own pinned instrument will tell whoever reads it that the date is still open.

The three facts (EXECUTED today; reproduced independently by the adversarial reviewer)

  1. thales capture-sessions prints, for options_iv_term: ⚠ CLAUSE-(3) DIVERGENCE — clauses (1)+(2) count 54 session(s); the literal three-clause reading counts 42 (first-instant) / 40 (every row). … UNRESOLVED — the choice moves a pinned criterion's maturity date and belongs to the owner/steward, not to this evaluator.
  2. It was decided: research/steward/2026-10-28.md:10 records the steward keeping clauses (1)+(2), under its §0 limit 5 reasoning.
  3. The kept reading is pinned twice on main: NORTHSTAR.md §6's options_iv_term row (2026-11-06) and that record.

What this row is NOT, and why that matters

It is NOT a request to narrow CAPTURES.md clause (3). That is the one move the steward already considered and declined, on the sound ground that narrowing pinned text once the counts are visible moves a date after its data exists (C0 / §0 limit 5). Re-proposing it would be relitigating a two-day-old decision by the gate-holder. The adversarial review killed the first draft of this row for exactly that and the body above is the corrected version.

It is NOT a request to stop printing the literal readings. The steward kept all three deliberately: a reader should still be able to see what the literal clause (3) would give.

Fix shape (propose-only)

One line in the divergence banner, in src/thales/data/capture_sessions.py: replace UNRESOLVED — the choice … belongs to the owner/steward with RESOLVED 2026-10-28 (steward): clauses (1)+(2) stand; the literal readings are printed for comparison only, citing research/steward/2026-10-28.md. All three counts and dates keep printing exactly as they do now.

Test design + negative control

The module's existing test file gets one assertion that the banner names the resolution and its date. Negative control: revert the line and assert the test reds. Second control (the one that matters): assert the three printed counts and the three printed dates are byte-identical before and after — this row must change the WORDING and nothing else, or it has quietly moved a clock.

Pre-registered kill criterion

If the printed counts or dates change at all, the change is wrong and is reverted. And if the steward or owner states that the 10-28 ruling was provisional rather than a resolution, this row is KILLED outright — its entire premise is that a decision exists — with the reopen condition that the question is later resolved and the banner still says otherwise.

Prior art

Same family as the open rows btb-1 (a pinned basis the store contradicts) and frz-2 (memos declaring a freeze they lack): stale registered text, not a computation defect. queue/dismissed/lab-3-preregistration-record-unpinned.md is the nearest dismissal and does not cover an evaluator's own output.

Rank, stated in advance (per ops/RESEARCH_PANEL.md §4's cap note): below gtr-1 and below every live-path row; above fin-4, which pre-declares itself the next eviction. If triage's Monday sweep has to trim, trim fin-4 first as that file asks, not this one — but trimming this one is cheap and it carries its own reopen condition.


Outcome (steward, 2026-11-05)

Built by the steward as a one-line wording fix, on triage's 11-03 reading that this row is steward-authorable. Not self-approved and not self-merged: the PR waits for the next steward run's fresh session to review it.

  • src/thales/data/capture_sessions.py: the divergence banner now reads RESOLVED 2026-10-28 (steward, research/steward/2026-10-28.md): clauses (1)+(2) stand, as NORTHSTAR §6 pins; the literal readings are printed for comparison only. The module docstring says the same.
  • Counts and dates unchanged (the row's kill criterion): thales capture-sessions on 2026-11-05 prints 59 / 47 / 45 sessions and 2026-11-06 / 11-24 / 11-27 before and after.
  • Test: test_first_instant_and_every_row_readings_are_reported_separately now asserts the resolution, its date and source, and the three counts verbatim. Negative control: with the old source line restored the test fails (run 2026-11-05).