Report-only; trial:null: Survivor, OPG-versus-MOC, and GP/A decomposition diagnostics preserve their independent-review limits and grant no return, allocation, or execution claim.
Date: 30 August 2026 Status: CURRENT · REPORT-ONLY · trial:null · variants_allowed:0
Decision: these three completed diagnostics change no rule, threshold, strategy, allocation, promotion decision, or live-cost credit. Every return series is a saved series ending no later than 2021; no post-2021 market-return row appears here. Later policy and killcard metadata were context only.
Supersedes: for these three diagnostics, this round-up supersedes the QUEUED statuses and older ledger row counts printed in edge-deep-dive-2026-08-29.md and feature-backlog-2026-08-29.md. Those previously published documents remain frozen and are not edited here.
1 — Outcomes at a glance
| Diagnostic | Reproduced outcome | Decision boundary |
|---|---|---|
| Survivor lenses | Active-GP/A versus total-momentum Pearson is 0.015 unscaled and −0.047 vol-scaled; vol-scaled 50/50 is 9.97% component-net arithmetic mean ×12, Sharpe 0.70, max DD −34.41%; equal-risk is 37.14%/62.86% | Correlation mixes active and total units; blends omit portfolio-level rebalancing/netting costs; no allocation claim |
| OPG vs MOC | Full-window OPG−MOC CAGR is +6.5 bps/year; 2017–2021 is −62.2 bps/year | Headline sign is robust to the registered fundability screen; t-stats are not reproducible from the saved artifact; size split is an artefact; credit zero OPG edge |
| GP/A decomposition | Registered active arithmetic mean ×12 2.83%/yr versus reconstruction 2.84%/yr, a 0.992 bp/yr gap; GP/A rank correlation is 0.728 with turnover and 0.094 with margin | Observed component difference 0.96%/yr is below the 4.17%/yr MDE: NOT DISTINGUISHABLE |
2 — Two-sleeve survivor lens
All 5,767 saved momentum daily rows compound into 275 GP/A exit intervals from 1 February 1999 through 1 December 2021. GP/A active-versus-EW is retained only as a co-movement lens against momentum total net. Blend arithmetic uses GP/A top-absolute net against momentum total net.
Correlations
| Lens | Momentum | Full Pearson | Full Spearman | 2017–2021 Pearson | 2017–2021 Spearman | Rolling 36m min / median / max |
|---|---|---|---|---|---|---|
| GP/A active vs momentum total — co-movement only | unscaled | 0.015 | −0.049 | 0.125 | 0.209 | −0.647 / −0.062 / 0.370 |
| GP/A active vs momentum total — co-movement only | vol-scaled | −0.047 | −0.067 | 0.090 | 0.130 | −0.601 / −0.065 / 0.270 |
| GP/A total vs momentum total | unscaled | 0.826 | 0.845 | 0.892 | 0.854 | 0.572 / 0.841 / 0.956 |
| GP/A total vs momentum total | vol-scaled | 0.697 | 0.815 | 0.824 | 0.812 | 0.573 / 0.809 / 0.921 |
The approximately-zero active-versus-total correlations are descriptive only. They are not total-book diversification estimates.
Unit-reconciled component-net blends
| Momentum | Series | Component-net arithmetic mean ×12 | Geometric CAGR | Volatility | Sharpe | Max DD |
|---|---|---|---|---|---|---|
| unscaled | GP/A top absolute net | 13.31% | 12.06% | 19.32% | 0.689 | −48.73% |
| unscaled | Momentum total net | 15.01% | 11.06% | 30.41% | 0.494 | −68.39% |
| unscaled | Naive 50/50 | 14.16% | 11.98% | 23.81% | 0.595 | −52.90% |
| unscaled | Inverse-vol / equal-risk | 13.97% | 12.07% | 22.57% | 0.619 | −51.47% |
| vol-scaled | GP/A top absolute net | 13.31% | 12.06% | 19.32% | 0.689 | −48.73% |
| vol-scaled | Momentum total net | 6.63% | 6.17% | 11.41% | 0.581 | −23.79% |
| vol-scaled | Naive 50/50 | 9.97% | 9.34% | 14.23% | 0.700 | −34.41% |
| vol-scaled | Inverse-vol / equal-risk | 9.11% | 8.56% | 13.21% | 0.689 | −30.37% |
| Momentum | Blend | GP/A weight | Momentum weight | Diversification ratio | GP/A / momentum risk-budget share |
|---|---|---|---|---|---|
| unscaled | Naive 50/50 | 50.00% | 50.00% | 1.044 | 37.84% / 62.16% |
| unscaled | Equal-risk | 61.16% | 38.84% | 1.047 | 50.00% / 50.00% |
| vol-scaled | Naive 50/50 | 50.00% | 50.00% | 1.079 | 64.99% / 35.01% |
| vol-scaled | Equal-risk | 37.14% | 62.86% | 1.086 | 50.00% / 50.00% |
Equal-risk weights are full-window in-sample diagnostics. Component series are net only within their own constructions; the blends omit portfolio-level rebalancing and netting costs. Aggregate returns cannot answer LL-4's opposite-trades-in-overlapping-names question because holdings and signed intended trades are absent.
3 — Risk-budget consistency audit
All nine published policy-v1 calibration checks reproduce exactly: screening-session count; vol-scaled H1 dates; vol-scaled 252-arm H2; vol-scaled soft-breach dates; vol-scaled retirement dates; comparison-only unscaled H1 dates; unscaled 252-arm H2; unscaled soft-breach dates; and unscaled retirement dates.
- Vol-scaled H1: 12 strict breaches under −3.2022%; unscaled counterfactual under the permitted threshold: 285.
- Continuous-HWM vol-scaled H2: peak 7 March 2000, crossing 18 July 2002, trough 12 March 2003 at −27.63%, recovery 7 September 2005, 792 inclusive crossing-to-recovery sessions.
- The policy document's reset-every-252-sessions diagnostic remains 0 vol-scaled H2 breaches. It answers a different question from the continuous-HWM episode.
- Soft bands: 7 vol-scaled breaches and 1 synthetic historical retirement. The 11 January 2010 arm breached at sessions 15 and 21, retiring on 9 February 2010. This was not a live retirement.
- The comparison-only unscaled calibration replays H1=15, H2=0, soft breaches=8, and retirements=1. It is not a permitted strategy variant.
Worst 10 narrated drawdown episodes — vol-scaled
Narrations are general knowledge only, not derived from the saved return series and not causal proof. Length is peak-to-trough observed sessions.
| Rank | Peak date | Trough date | Depth | Length | Recovery date | Market context |
|---|---|---|---|---|---|---|
| 1 | 2000-03-07 | 2003-03-12 | −27.63% | 755 | 2005-09-07 | Post-dot-com bear market through the run-up to the Iraq war. |
| 2 | 2008-06-19 | 2009-03-09 | −21.60% | 180 | 2011-02-07 | Global financial crisis trough and subsequent momentum-factor reversal. |
| 3 | 2015-07-23 | 2016-02-09 | −19.71% | 138 | 2017-09-19 | China-growth fears, oil collapse, and early-2016 risk-off reversal. |
| 4 | 2020-02-20 | 2020-03-19 | −15.67% | 20 | 2020-11-24 | COVID-19 crash and exceptionally fast cross-asset deleveraging. |
| 5 | 1999-01-12 | 1999-01-13 | −15.37% | 1 | 1999-02-01 | Late-1990s dot-com-era factor and technology volatility. |
| 6 | 2018-09-05 | 2018-12-24 | −13.53% | 76 | 2020-01-16 | Q4 2018 selloff amid Fed tightening and US-China trade tension. |
| 7 | 2006-05-11 | 2006-06-14 | −13.09% | 23 | 2007-05-31 | May–June 2006 global risk-off selloff amid inflation and rate concerns. |
| 8 | 1999-02-01 | 1999-02-10 | −12.24% | 7 | 1999-03-10 | Late-1990s dot-com-era style rotation and volatility. |
| 9 | 2011-05-02 | 2011-10-04 | −11.40% | 108 | 2012-09-14 | Euro-area sovereign-debt crisis and US debt-ceiling volatility. |
| 10 | 2007-10-19 | 2008-01-22 | −10.14% | 63 | 2008-06-19 | Early global-financial-crisis credit stress and equity selloff. |
Worst 10 narrated drawdown episodes — unscaled counterfactual
| Rank | Peak date | Trough date | Depth | Length | Recovery date | Market context |
|---|---|---|---|---|---|---|
| 1 | 2000-03-07 | 2002-07-24 | −74.19% | 596 | 2015-03-19 | Dot-com bust, 2001 recession/9-11 aftermath, and 2002 accounting crisis. |
| 2 | 2020-02-20 | 2020-03-19 | −40.82% | 20 | 2020-07-06 | COVID-19 crash and exceptionally fast cross-asset deleveraging. |
| 3 | 2015-07-23 | 2016-02-09 | −36.11% | 138 | 2017-10-05 | China-growth fears, oil collapse, and early-2016 risk-off reversal. |
| 4 | 2018-09-05 | 2018-12-24 | −31.23% | 76 | 2020-01-16 | Q4 2018 selloff amid Fed tightening and US-China trade tension. |
| 5 | 2021-02-16 | 2021-05-11 | −30.26% | 59 | unrecovered | 2021 reopening, reflation, and sharp growth/value and momentum rotations. |
| 6 | 1999-04-14 | 1999-04-20 | −26.23% | 4 | 1999-10-29 | Sharp 1999 technology/growth rotation during the dot-com boom. |
| 7 | 1999-01-12 | 1999-01-13 | −15.37% | 1 | 1999-02-01 | Late-1990s dot-com-era factor and technology volatility. |
| 8 | 2000-01-03 | 2000-01-07 | −13.35% | 4 | 2000-01-19 | Y2K/dot-com turn-of-year volatility before the 2000 peak. |
| 9 | 1999-02-01 | 1999-02-10 | −12.24% | 7 | 1999-03-10 | Late-1990s dot-com-era style rotation and volatility. |
| 10 | 2020-09-02 | 2020-09-08 | −11.72% | 3 | 2020-10-05 | September 2020 technology/growth correction after the post-COVID rally. |
No observed breach is used to tune or relax a threshold. This screening replay cannot observe H3 reconciliation/data failures or H4 buying-power/short events.
4 — OPG versus MOC execution audit
Across 1999–2021, OPG minus MOC CAGR is +6.5 bps/year and next-close minus MOC is −86.2 bps/year. Over 2017–2021 they are −62.2 and −101.8 bps/year. A constrained rerun using the registered $10 and $10M ADV screen preserved the headline sign: full OPG−MOC moved from +9.4 to +6.5 and 2017–2021 from −19.9 to −62.2 bps/year.
| Window | Cost/side | MOC CAGR | OPG CAGR | Next-close CAGR |
|---|---|---|---|---|
| full period | 10 bps | 15.56% | 15.62% | 14.70% |
| full period | 25 bps | 14.08% | 14.14% | 13.23% |
| full period | 50 bps | 11.65% | 11.71% | 10.82% |
| 2017–2021 | 10 bps | 22.48% | 21.87% | 21.47% |
| 2017–2021 | 25 bps | 21.00% | 20.39% | 20.00% |
| 2017–2021 | 50 bps | 18.57% | 17.97% | 17.59% |
The matched buy-event overnight gap exceeds same-name non-rebalance gaps by 13.9 bps; the report gives NW(0) monthly t=2.720091. The sell difference is 11.1 bps, t=2.153492. The JSON preserves summaries but not the monthly series, so those t-statistics cannot be independently reproduced from the artifact; lag zero is not justified against serial dependence from persistent holdings.
The current-metadata 2017–2021 size split reports OPG−MOC of −62.4 bps/year for Large/Mid and −27.4 for Small/Micro, but it is artefact-contaminated by non-PIT labels, unclassified targets, stale/non-global month-end allocations, last-print-to-cash treatment, and no fill-day liquidity/staleness screen. It supports no size mechanism.
Live-lane conclusion: retain 10 bps/side base and 25 bps/side stress, and credit ZERO OPG edge pending auction, impact, and terminal-price validation. No queue, spread, auction imbalance, partial-fill, impact, or broker-routing model was present.
5 — GP/A decomposition
The reconstruction matches the registered series closely enough for the stated gate, but is not byte-identical machinery:
- registered active arithmetic mean ×12: 2.83%/yr;
- reconstructed active arithmetic mean ×12: 2.84%/yr;
- reproduction gap: 0.992 bp/yr over 275 aligned months — PASS against the ≤5 bp/yr gate;
- maximum monthly difference: 14.18 bp; monthly-difference RMSE: 1.93 bp.
Mean monthly cross-sectional Spearman rank correlations are 0.094 for GP/A versus gross margin and 0.728 for GP/A versus asset turnover. In 2017–2021 they are 0.037 and 0.749. GP/A therefore tracks turnover far more closely in this diagnostic, but this is descriptive, not causal.
| Full-sample cell | Active arithmetic mean ×12 | NW t(0) | NW t(3) | Return months |
|---|---|---|---|---|
| margin low / turnover low | −4.92% | −3.34 | −2.95 | 275 |
| margin low / turnover high | 1.25% | 0.98 | 0.96 | 275 |
| margin high / turnover low | −2.03% | −1.85 | −1.73 | 275 |
| margin high / turnover high | 1.13% | 0.68 | 0.58 | 275 |
Top-margin-tercile active arithmetic mean ×12 is 0.51%/yr; top-turnover-tercile is 1.47%/yr. Their absolute difference is 0.96%/yr, versus a two-sided α=0.05, 80%-power sensitivity MDE of 4.17%/yr. The difference is NOT DISTINGUISHABLE.
The four extreme-tercile corners do not span the registered top-quintile sleeve: middle-tercile or component-missing observations account for a 93.09% residual share of reconstructed active arithmetic mean. The correlations and tercile formation ranks use a 286-month component-valid subset, while return comparisons use the 275-month aligned series. SF1 is a current vendor snapshot without PIT revision history; sector and cost labels are also current/non-PIT. These limitations forbid a component promotion or mechanism claim.
6 — Custody and non-action
Sources were the saved survivor-series, opg-vs-moc, and gpa-decomposition diagnostic bundles; the two survivor Markdown outputs; the registered preregistrations and policy-v1 document/event; and the three independent review records. Local absolute paths are intentionally omitted.
No post-2021 market-return rows were read. Undated current metadata and later policy/killcard metadata were context only. All three diagnostics remain trial:null, report-only, and non-promotable.