PBC-1 — the public site publishes momentum's broker book ("69 names") while the same site's own notes describe a 50-name sleeve, and nothing on the page relates the two
status: DISMISSED 2026-10-30, panel run 47 · raised as a carried handoff by the steward on 2026-10-26/27/28/29 ("Public site shows momentum's raw position count (69), not its tradeable count (~48). Cosmetic, still unfiled; the panel or triage files it") · class: published-record clarity (NORTHSTAR §1 asset 1) · effort ~0.15 pd · judgement: NO — not now
Plain-language summary. The public dashboard's momentum page shows "current book — 69 names". The strategy selects 50. Both numbers are true: 69 is what the broker account holds, 50 is what the monthly selection targets, and the difference is residue the book has not finished selling. Nothing on the page says so, and the site's own published memos quote "50-name" repeatedly, so a careful reader finds a 38% discrepancy with no explanation. Nothing is mis-stated and no decision depends on it.
What was measured (EXECUTED, committed state only)
web/public/data/fleet.jsonmomentumlatest_snapshot(generated 2026-10-29T15:30:27Z):n_positions69,gross_leverage0.2735, equity $13,680.71,long_mkt_value$3,741.33.- Provenance:
execution/public_export.py:242copiesn_positionsverbatim from the sleeve's own snapshot;execution/daily.py:660writes it aslen(broker.get_positions())— broker truth, every position the account holds. The export is a pure renderer, so there is nothing wrong in it. - Rendering:
web/app/sleeves/[sleeve]/page.tsx:150-155— label "current book", value "69 names", sub "0.27x gross exposure". Accurate as labelled. - The target set is 50:
data/processed/kelly_ledger.jsonlast_targets.weights(dated 2026-10-01) holds exactly 50 symbols summing to 0.2709 — which the live book matches to 0.0026 (0.2735 actual). The live book is on target; this is not an execution-fidelity gap, and that is worth recording on its own. - PRIOR ART, found and deliberately not relitigated. The owner's own
research/2026-09-02_self_audit.md:212item 9 already adjudicated the mechanism: "Dust. 22 positions under $10 ($62 total) that no run can sell (min_order_notional$10); health excludes them; a kill-switch liquidation would leave them. Hygiene only." The code path ispipeline.compute_orders(src/thales/execution/pipeline.py:124-131):if abs(diff) < min_notional: continue, whose tripwire logs only the entry case (current == 0.0 and target > 0.0, the 2026-06-09 finding), so a full EXIT below the floor is skipped silently. The chair also tested whether that dust breaks KSP-1's observation clause and found it explicitly tolerated:daily.py:37-41sets_KILL_SWITCH_FLAT_GROSS = 0.05with a comment naming "residual unsellable names ... a few percent", and the dust is ~0.5-1.4% of equity. Both halves checked out; neither is a finding. - So the 19-name difference is un-targeted residue, and breadth has ratcheted 47 (2026-06-11) → 53 → 58 (07-02) → 63 (08-05) → 67 (09-02) → 70 (10-02) → 69 (10-13): roughly +3 to +5 at each monthly selection, never falling back.
Why it is dismissed rather than filed
- The real question already has a row, with a sharper tripwire.
queue/dismissed/fin-2-minimum-viable-equity-floor.mdowns the breadth overshoot and its economics, and its reopen condition 2 is numeric: "Average position falls below $50, or breadth exceeds 75 names on the committed snapshot." Evaluated today: average position $3,741.33 / 69 = $54.22 and breadth 69 — NOT FIRED, on both legs, measured not argued. A label row filed beside that tripwire would duplicate it at lower resolution. - The steward's "~48 tradeable" figure is not derivable here and is not
relied on. No per-symbol holdings are committed (
portfolio_snapshots.jsonlcarries aggregates only) — the same BUILD-BRIEF GAP FIN-2 records. Treat the ~48 as unverified; the verified numbers are 69 held and 50 targeted. - Nothing is wrong. No trade, alarm, gate, ledger or published number is incorrect; the page states what the account holds. On a paper account at 0.27x gross the residue is ~$54 a name.
One neighbouring tripwire checked and ALSO not fired
FIN-2's reopen condition 4 is "the build-brief gap closes — any committed
source makes the median position derivable without a broker pull". It has NOT
fired. long_mkt_value ($3,741.33) is a committed field, written from
broker truth at daily.py:651-653, so the mean position is derivable
without a broker pull and is what this file quotes. The median FIN-2 asks
for still is not: no per-symbol holdings are committed anywhere. Condition 4
is written on the median, so it stands unfired, and FIN-2's build brief is
still under-specified for the same reason it was on 2026-10-05.
Reopen conditions (any one)
- FIN-2's reopen condition 2 fires — average position below $50 OR breadth above 75 on the committed snapshot. Forecast from the measured ratchet: the 2026-11-02 selection takes breadth to ~72-74 (still inside), and the 2026-12-01 selection is the first likely breach. At that point the published 69-and-rising stops being residue and starts being a book the validated 50-name engine never ran; reopen as part of FIN-2, not separately.
src/thales/execution/public_export.pyis touched for any reason and neither this clarification nor PTS-1's one-line fix rides along. The cheapest form is one extra field — the target count is committed (kelly_ledger.last_targets) — rendered as "69 names · 50 targeted". This is the same free-rider conditionqueue/dismissed/pts-1-*.mdcondition 3 already carries on the same file; whoever next edits it should take both.- Real capital is deployed (FIN-2 condition 1, the expected path). A public page understating how much of a real book is off-strategy is a different question from the same page understating a paper one.
Why this file exists at all
The steward has carried this item forward in four consecutive daily records
with no verdict anywhere, because a carry has no closing move. This file is
the verdict, so the carry can stop. That is the whole point of
ops/RESEARCH_PANEL.md "What matters here" #2: a dismissal with a live
reopen condition is a standing tripwire, and an unfiled observation is rot.