Internal desk accountability only. Friday September 4 settled prices; written September 6 for the September 8 capital window. Monday September 7 is the Labor Day holiday. This grades desk calls and derived paper-book records, not the user's holdings.

Trust before action

The fresh scoreboard has 110 published calls and 20 BUY calls. At 30 calendar days, 3 of 11 eligible BUY calls beat SPY: 27.27%. Eight missed; their mean underperformance was 10.37 percentage points. Nine BUY calls are excluded from that horizon because their window is incomplete or has no usable data. At 90 days, none of 20 BUY calls has a complete eligible window: both hit rate and mean miss are unavailable, not zero.

These are publication-date raw-price returns against SPY, not triggered-entry or executed-trade returns. Conditional BUY calls can therefore score poorly even when their actual entry condition was never met. The result measures the desk's published directional timing, and does not establish the profitability of the campaign's execution conditions. New campaign calls have no mature grade. Campaign1 has no predecessor campaign to grade; its ledger30/90-day cells remain pending.

The last 30 days: what failed and what the campaign changes

The review window is August 6–September 4. Eight older July 8 BUY calls reached their nominal 30-day anniversary inside it, all on August 7. The five misses below are a maturity-window slice, not calls first published during the last 30 days. The July 5 MU/STX/WDC misses belong in the all-eligible Trust statistic but matured before this review window.

Would this campaign have caught them? It would require company proof before admission and an explicit confirmation sequence before a BUY; jagged entries become WAIT. That addresses the mechanism. It is not a point-in-time replay of July evidence, so the desk cannot claim it would have avoided every historical miss. September knowledge cannot repair a July decision retroactively.

The prior four autopsies repeatedly described armed rungs without agent verdicts, fixed numbers masquerading as moving conditions, incomplete qualitative conjunctions, and covered lanes without company decisions. This run found those failures alive: older targets still said OPEN after their text recorded confirmation or replacement; eleven calls in the September 5 article needed reconciliation, not just the headline five; two source-only July 18 watches had no target at all. Targets alone were an incomplete inventory.

Names closed 22 overlapping live targets and restored one MU watch's recorded August 28 close. Trust's fresh projection found the two targetless watches; their missing targets are backfilled CLOSED as superseded without inventing historical success or failure. Final supersessions are 24, plus one repaired historical close. Original published stances remain for grading. This is follow-through on the desk's own stale decisions, not an engine fix.

Open-call check

call-check completed on 79 open published calls, 41 with numeric triggers; the remaining 38 have prose-only conditions and cannot be called mechanically cleared. The output marks pre-existing invalidation flags on PL, DUK and NTLA, and existing in-zone conditions elsewhere. A deduped flag means the alert already existed, not that somebody adjudicated it. DHR's documented qualitative conjunction remains a known unresolved bell issue. Campaign publication will add the 16 new source calls and receive a final numeric parse/flag check.

The projection held malformed OHLC candidates while selecting valid canonical alternatives. Four unrelated terminal targets lack a resolution date; their date remains null. These warnings are recorded, not fabricated into clean historical outcomes. The missing-call and stale-resolution findings belong in the existing engine task list; no new guard is authorized.

Paper performance and promotion

The paper-performance skill's direct ledger and positions steps conflict with this campaign's explicit ban. The permitted alternative is the already-derived paper strategy aggregate, generated September 5 and valued on September 4 where applicable. The table copies that aggregate rather than recomputing constituent positions. The cash-only dip-buyer has no valuation date and no mark is inferred.

These are projected account returns, not trade hit rates. Bench-signals lacks a comparable return denominator, so its current value cannot prove the research system works or loses. Yolo's higher projected return alone does not establish repeatable edge; unequal start windows and risk exposure prohibit a causal strategy ranking. Tiny differences from the older equity snapshots are retained in the receipts; this table consistently uses the aggregate projection.

The supplied promotion ladder was generated August 12 and is stale for a September 6 decision. Its seven paper actors were all HOLD, with drawdown not computed. Requirements include at least 40 scored trades, two quarters of history, positive median 30-day excess, at least 50% hit rate, discipline checks and a valid drawdown box. Neither fresh promotion nor a current demotion can be inferred from this snapshot. No ladder regeneration through banned ledger reads, no paper trades, and no policy tuning occurred.

What the manual campaign actually tested

It forces a lane decision and then a company decision; it cannot guarantee either will be right. Memory and electricity remain primary while the physical-AI component case waits for direct economics. One new company dive supplies an entry receipt, and fifteen watches keep their unresolved timing or evidence conditions visible. The explicit MU historical-RSI exception is a reviewer-visible judgment to grade, not a silently softened screen. Two manual campaigns remain the prerequisite for deciding what to automate.

Sources and reproducibility

• Metrics: python3 research/issues/2026-09-06-campaign-01-trust-receipts/call-metrics.py --input research/issues/2026-09-06-campaign-01-trust-receipts/desk-calls.json.
• Frozen code output: research/issues/2026-09-06-campaign-01-trust-receipts/call-metrics-phase4.json; original projection command deno task research:terminal-projections --only desk-calls; deno task market -- call-check output archived alongside it.
• Paper and promotion receipts: paper-performance-aggregates.json, paper-five-snapshots.json, paper-ladder-snapshot.json in the same directory. They contain derived paper-only aggregates, no constituent holdings.
• Prior autopsies, process-only; August 25, August 17, August 11 and July 31 sources retain their dates.
• Names review; climate; theses.