Graveyard
Candidates decided against in this research program. Each delta is in percentage points (pp) against the base named in its row, on the window named in its row. Deltas in different rows were measured against different bases and do not rank against each other.
Outcomes: a candidate can pass the bar and still not be adopted, be rejected on its measurements, or be stopped by a diagnostic before a backtest could bind.
Bases
Each delta was measured against the base named in its row:
- Oct 11 rules: the live rules from 2026-10-11 to 2026-10-13.
- Oct 13 starting rules: the rules adopted on 2026-10-13, before event rotation (replaced the same day).
- event-rotation finalist: the Oct 13 starting rules plus daily event rotation, before the admission rule (a research base on 2026-10-13).
- Oct 13 rules: event rotation plus the leader-emergence admission, live 2026-10-13 to 2026-10-14.
- Oct 14 rules: the Oct 13 rules plus the pre-release boost on the top name, from 2026-10-14; the variant with the smaller leveraged-ETF sleeve in the pre-release window followed the same day.
- Oct 15 rules: the Oct 14 rules after the 2026-10-15 changes (volatility target, release-session hold, release estimate without the fallback rule), live 2026-10-15 to 2026-10-16.
- Oct 16 rules: the Oct 15 rules after the 2026-10-16 changes (post-release cut added, carry cap removed), live 2026-10-16 to 2026-10-22.
- Oct 14-22 rules: the rules in use from 2026-10-14 to 2026-10-22 (the Oct 14, Oct 15 and Oct 16 rules above); rows that name only this range were measured on one of them.
- Oct 22 rules: the Oct 16 rules plus the dip rule on the top name with a smaller leveraged-ETF sleeve, live 2026-10-22 to 2026-10-28 (replaced by the quality strip on 2026-10-28).
- early-October combined base: a combined variant that served briefly as the research base in October 2026, before 2026-10-13.
- retired base: an earlier research base, retired before the October 2026 bases; those rows ran on a window ending 2026-09-18.
- published-factor test base: a one-off variant of the rules (a regime-gated, capped weighting) used as the base for the published-factor batch.
- matched 5-sleeve all-momentum control: a five-sleeve all-momentum portfolio built as the control for the event-sleeve study.
Passed the bar, not adopted (5)
| Candidate | Headline delta | Reason | Deciding facts |
|---|---|---|---|
| Speed blend (a half-and-half blend of the Oct 13 starting rules and the event-rotation finalist) | -8.10pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Concentration (a few names or years carry the result) | Half starting rules / half event-rotation finalist stock book returned 114.72% CAGR (+7.50pp vs the Oct 13 starting rules, -8.10pp vs the finalist) with fixed-January walk-forward 103.6%, and its leave-one-year-out fixed-January walk-forward fell below 100% without 2017 (95.7%), 2020 (82.3%) or 2024 (84.2%), so it passed the bar against the Oct 13 starting rules but was not adopted over the finalist. |
| Event-rotation near-finalists (50-day average confirmation; split buffer; release-day window of 5) | +0.50pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Mechanism overlap | Three sibling versions beat or matched the finalist on single headline figures (the 50-day average confirmation 124.6% CAGR and 116.1% fixed-January walk-forward; the split buffer 117.3% fixed-January walk-forward at 120.7% CAGR; the 5-day window maximum drawdown -53.5% at 117.0% CAGR) but none was adopted, because the finalist won the family walk-forward (114.9%, picked in all 12 years, vs 102.4% for the 50-day-average family) and the rolling four-year median. |
| Consensus-miss exclusion, any coverage (20 sessions) | +4.40pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Mechanism overlap | The post-hoc any-coverage form of the high-coverage consensus-miss exclusion passed its pre-registered retest marginally (+4.4pp CAGR, January walk-forward 84.5% vs 80.2%, DSR 0.00, 91.6% coverage of reports) and sits not adopted. On the Oct 14-22 rules it was skipped, because no point-in-time consensus source is usable live, so it is not an open live option. |
| Event-driven rebalancing (daily rotation out of the buffer) | +6.40pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Superseded by an adopted variant | Rotating a held name out at the next open as soon as it became ineligible or fell outside a rank buffer of 12 (fixed schedule kept) added +6.4pp CAGR vs the Oct 11 rules (101.2%), with a January walk-forward of 88.0% vs 80.2%, a four-year win rate of 74% and +18.7pp from 2022 onward, but it sat below the 10pp adoption bar, with maximum drawdown -62.1% vs -57.3%, a deflated Sharpe ratio of 0.00 and stress forced liquidations of 193 vs 120; the pre-registered confirmation then failed: the result spikes at one buffer size (buffers 10 to 14 read +2.6 / +3.6 / +6.4 / +11.9 / +6.8pp, a 5.5pp step from 12 to 13 against a 4pp limit, and buffer 9 read -0.4pp in the original grid), and the four-year-fold walk-forward read 100.0% vs the control's 100.4%. Its stronger descendant, with a statistical margin on the replacement test, is the adopted daily check; this standalone form was not adopted. |
| Cost-aware vol target under Robinhood-style margin (train-best 0-breach vt 0.75) | +1.60pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Not chosen | Choosing the train-best vol target with zero breaches under Robinhood-style margin (vt 0.75) met the mechanical bar marginally at +1.6pp CAGR vs live Oct 11 rules (train +1.4, fresh 2022+ +2.1, 4y win 89%, family January walk-forward 81.6% vs 80.2%, 0 breaches under Robinhood-style margin) but was not adopted: it is a leverage dial (MaxDD -58.6% vs -57.3%), DSR 0.00, and the family walk-forward picked the grid edges (vt 0.50 / 0.80) in all 12 years, never vt 0.75; the same family at vt above 0.70 had already been rejected. |
Rejected (195)
| Candidate | Headline delta | Reason | Deciding facts |
|---|---|---|---|
| Effective-date index reconstitution lead-capture sleeve on the Oct 22 rules | -58.20pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Catching names in the stock book for five sessions around index effective-date reconstitutions cut fixed-January walk-forward by 58.20pp vs Oct 22 rules (exposure-matched gap +55pp): the forced-flow timing is not free altitude on this concentrated book. |
| Short-squeeze leverage boost (short-interest data), on the Oct 22 rules | -12.04pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Adding leverage when short-interest data signalled a likely short squeeze cut the fixed-January walk-forward by 12.04pp vs the Oct 22 rules, even though it beat the exposure-matched test: short-squeeze timing does not add return on these rules. |
| Adverse 8-K cluster admission veto on the Oct 22 rules | -11.49pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Vetoing admissions after an adverse 8-K cluster (W=5, H=21) cut fixed-January walk-forward by 11.49pp vs Oct 22 rules: cluster timing does not lift the book. |
| Leverage boost after a new buyback authorization (8-K filings), on the Oct 22 rules | -10.25pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | A short leverage boost after a company filed a new share-buyback authorization (from 8-K filings, not the corporate-actions feed) cut the fixed-January walk-forward by 10.25pp vs the Oct 22 rules: buyback-announcement timing hurts on these rules. |
| Lead selling-climax volume brake on the Oct 22 rules | -12.90pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Cutting the lead after a selling-climax volume print cut fixed-January walk-forward by 12.90pp vs Oct 22 rules: microstructure climax brakes lose on this concentrated book. |
| 8-K Item 2.01 acquisition rank-exit stickiness on the Oct 22 rules | -11.55pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Holding rank-exits after 8-K 2.01 acquisition completions cut fixed-January walk-forward by 11.55pp vs Oct 22 rules. |
| Leverage boost on heavy buyback execution (10-Q filings), on the Oct 22 rules | -9.56pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Boosting leverage when quarterly 10-Q filings showed heavy share repurchases cut the fixed-January walk-forward by 9.56pp vs the Oct 22 rules (a separate test from the buyback-authorization boost). |
| Deferred-revenue and operating-cash dial on the pre-release boost on the Oct 22 rules | -15.53pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Dialing the pre-release boost from deferredrev∧ncfo YoY cut fixed-January walk-forward by 15.53pp vs Oct 22 rules: release-window dose is not free headroom. |
| NFCI continuous stock-sleeve scaler on the Oct 22 rules | -13.44pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Scaling the stock sleeve continuously with Chicago Fed NFCI cut fixed-January walk-forward by 13.44pp vs Oct 22 rules: financial-conditions compression of the stock book is hostile here. |
| Dual-list calendar stock leverage pulse on the Oct 22 rules | -10.18pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Raising stock leverage on dual-list calendar days cut fixed-January walk-forward by 10.18pp vs Oct 22 rules despite clearing kinship kills vs the post-release cut. |
| Stock-book cut on a weekly jump in the St. Louis Fed financial stress index, on the Oct 22 rules | -13.10pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Cutting the stock book when the St. Louis Fed's weekly financial stress index jumped from the week before cut the fixed-January walk-forward by 13.10pp vs the Oct 22 rules, the same direction as the rejected Chicago Fed financial-conditions scaler. |
| Index-addition sleeve between announcement and effective date (re-run with announcement dates), on the Oct 22 rules | -9.61pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Once index-change announcement dates were added to the data, a sleeve holding newly announced index additions until they took effect passed the pre-run mechanics check but still cut the fixed-January walk-forward by 9.61pp vs the Oct 22 rules (the candidate's walk-forward was 162.00%). |
| Lead-weight cap when days-to-cover is high (short-interest data), on the Oct 22 rules | -11.07pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Capping the top name at 0.75 of its weight when its days-to-cover (short interest divided by daily volume) was in the top fifth cut the fixed-January walk-forward by 11.07pp vs the Oct 22 rules (the candidate's walk-forward was 160.54%). |
| Shift into other assets on the S&P 500 futures basis, on the Oct 22 rules | -20.24pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Moving money into other assets based on the S&P 500 futures basis (from a free data source) cut the fixed-January walk-forward by 20.24pp vs the Oct 22 rules (the candidate's walk-forward was 151.36%): the futures basis adds nothing on these rules. |
| FINRA short-volume breadth whole-book/TQQQ risk resize on the Oct 22 rules | -12.57pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Resizing risk from FINRA short-volume breadth cut fixed-January walk-forward by 12.57pp vs Oct 22 rules. |
| Admission freeze while several holdings report earnings (earnings-calendar data), on the Oct 22 rules | -24.64pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Freezing new admissions when two or more held names had reported earnings in the last five sessions cut the fixed-January walk-forward by 24.64pp vs the Oct 22 rules. |
| ZN/NQ futures-state TQQQ-only sizing on the Oct 22 rules | -12.40pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Sizing only the TQQQ sleeve from ZN/NQ futures state cut fixed-January walk-forward by 12.40pp vs Oct 22 rules. |
| Small sleeve of newly spun-off companies (corporate-actions feed), on the Oct 22 rules | -11.93pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | An 8% sleeve holding newly spun-off companies for 10 sessions cut the fixed-January walk-forward by 11.93pp vs the Oct 22 rules. |
| Leverage boost when a passive holder first crosses 5% (13G filings), on the Oct 22 rules | -11.00pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | A short leverage boost when a passive investor first filed a 13G stake in the top name cut the fixed-January walk-forward by 11.00pp vs the Oct 22 rules (the activist 13D version was dropped at the pre-run mechanics check). |
| Small sleeve of companies that just spun off a unit (corporate-actions feed), on the Oct 22 rules | -16.55pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | An 8% sleeve holding the parent companies of fresh spin-offs for 10 sessions cut the fixed-January walk-forward by 16.55pp vs the Oct 22 rules. |
| Small sleeve of new listings (corporate-actions feed), on the Oct 22 rules | -11.53pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | An added 8% sleeve holding newly listed companies for 10 sessions, with no overlap with the rules' own holdings during the hold, still cut the fixed-January walk-forward by 11.53pp vs the Oct 22 rules. |
| Continuous regime-model exposure scale (published paper, arXiv 2402.05272), on the Oct 22 rules | -21.69pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Scaling exposure continuously by a statistical jump-model regime estimate cut the fixed-January walk-forward by 21.69pp vs the Oct 22 rules (the candidate's walk-forward was 149.91%, and it was 0.24pp behind the exposure-matched test). |
| Volume-price divergence admission (published paper, arXiv 2508.14656), on the Oct 22 rules | -79.83pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Admitting names on a divergence between trading volume and price cut the fixed-January walk-forward by 79.83pp vs the Oct 22 rules (the candidate's walk-forward was 91.77%). |
| Regime-ranking overlay, first form (from a published-paper review), on the Oct 22 rules | -13.92pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | A regime-ranking overlay, charged for its trading, cut the fixed-January walk-forward by 13.92pp vs the Oct 22 rules (the candidate's walk-forward was 157.68%): a second sleeve or overlay dilutes these rules. |
| Regime-ranking overlay, second form (from a published-paper review), on the Oct 22 rules | -12.79pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | A second regime-ranking overlay, charged for its trading, cut the fixed-January walk-forward by 12.79pp vs the Oct 22 rules (the candidate's walk-forward was 158.82%). |
| Structural idea from a published-paper review: form 1, on the Oct 22 rules | -46.19pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | The first structural change drawn from a review of published papers cut the fixed-January walk-forward by 46.19pp vs the Oct 22 rules (the candidate's walk-forward was 125.42%). |
| Structural idea from a published-paper review: form 2, on the Oct 22 rules | -17.45pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | The second structural change from the same review cut the fixed-January walk-forward by 17.45pp vs the Oct 22 rules (the candidate's walk-forward was 154.15%). |
| Structural idea from a published-paper review: form 3, on the Oct 22 rules | -55.87pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | The third structural change from the same review cut the fixed-January walk-forward by 55.87pp vs the Oct 22 rules (the candidate's walk-forward was 115.73%). |
| Structural idea from a published-paper review: form 4, on the Oct 22 rules | -16.90pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | The fourth structural change from the same review cut the fixed-January walk-forward by 16.90pp vs the Oct 22 rules (the candidate's walk-forward was 154.70%). |
| Structural idea from a published-paper review: form 5, on the Oct 22 rules | -13.47pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | The fifth structural change from the same review cut the fixed-January walk-forward by 13.47pp vs the Oct 22 rules (the candidate's walk-forward was 158.13%). |
| Stock-book scale from the SPY-QQQ lead-lag (published paper, arXiv 2201.08283), on the Oct 22 rules | -12.82pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Scaling the stock book by whether SPY or QQQ was leading cut the fixed-January walk-forward by 12.82pp vs the Oct 22 rules; it beat the exposure-matched test by +2.64pp but still lost to the rules. |
| Admission damper on heavy short-sale volume (FINRA short-volume data), on the Oct 22 rules | -36.46pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Damping admissions of names with heavy daily short-sale volume cut the fixed-January walk-forward by 36.46pp vs the Oct 22 rules. |
| Two-rate regime scale from the Fed funds and 3-month T-bill rates (published paper, arXiv 2503.11499), on the Oct 22 rules | -43.14pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Scaling exposure by a regime read from the Fed funds rate and the 3-month Treasury bill rate cut the fixed-January walk-forward by 43.14pp vs the Oct 22 rules. |
| Demoting names whose gains come overnight (published paper, arXiv 1410.5513), on the Oct 22 rules | -80.63pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Demoting names whose returns came mostly overnight cut the fixed-January walk-forward by 80.63pp vs the Oct 22 rules. |
| Demoting names slow to recover from drawdowns (published paper, arXiv 1403.8125), on the Oct 22 rules | +4.72pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Demoting names slow to recover from their largest drawdown added +4.72pp to the fixed-January walk-forward (and beat the exposure-matched test by +33.07pp) but missed the +5pp bar in force then, so it was rejected. |
| Admission screen: high three-month realized volatility, on the Oct 22 rules | -68.82pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by high three-month realized volatility cut the fixed-January walk-forward by 68.82pp vs the Oct 22 rules. |
| Demoting names with deep drawdowns (published paper, arXiv 1403.8125), on the Oct 22 rules | -113.34pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Demoting names by the depth of their largest drawdown cut the fixed-January walk-forward by 113.34pp vs the Oct 22 rules. |
| Admission screen: left-tail sensitivity to market falls (beyond their normal market sensitivity), on the Oct 22 rules | -65.80pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by left-tail sensitivity to market falls (beyond their normal market sensitivity) cut the fixed-January walk-forward by 65.80pp vs the Oct 22 rules. |
| Admission screen: crash risk (negative skew of their returns), on the Oct 22 rules | -56.70pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by crash risk (negative skew of their returns) cut the fixed-January walk-forward by 56.70pp vs the Oct 22 rules. |
| Admission screen: year-over-year net income, bottom 5%, on the Oct 22 rules | +11.40pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year net income, bottom 5% added +11.40pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (it failed one of them), so it was rejected. |
| Admission screen: year-over-year net income, bottom 10%, on the Oct 22 rules | +15.20pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year net income, bottom 10% added +15.20pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (the walk-forward and 2022-onward checks), so it was rejected. |
| Admission screen: slowing year-over-year net income growth, bottom 5%, on the Oct 22 rules | -18.10pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by slowing year-over-year net income growth, bottom 5% cut the fixed-January walk-forward by 18.10pp vs the Oct 22 rules. |
| Admission screen: year-over-year change in cash to assets, bottom 4%, on the Oct 22 rules | -10.08pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year change in cash to assets, bottom 4% cut the fixed-January walk-forward by 10.08pp vs the Oct 22 rules. |
| Admission screen: year-over-year rise in debt to assets, top 4%, on the Oct 22 rules | -8.20pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year rise in debt to assets, top 4% cut the fixed-January walk-forward by 8.20pp vs the Oct 22 rules. |
| Admission screen: year-over-year change in current ratio, bottom 4%, on the Oct 22 rules | -41.02pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year change in current ratio, bottom 4% cut the fixed-January walk-forward by 41.02pp vs the Oct 22 rules. |
| Admission screen: year-over-year earnings per share, bottom 4%, on the Oct 22 rules | +11.20pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year earnings per share, bottom 4% added +11.20pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (the risk checks), so it was rejected. |
| Admission screen: year-over-year return on assets, bottom 4%, on the Oct 22 rules | +10.27pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year return on assets, bottom 4% added +10.27pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (the risk checks), so it was rejected. |
| Admission screen: year-over-year asset growth, top 4%, on the Oct 22 rules | -105.69pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year asset growth, top 4% cut the fixed-January walk-forward by 105.69pp vs the Oct 22 rules. |
| Admission screen: revenue at or above the median with net income in the bottom quarter, on the Oct 22 rules | -1.49pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by revenue at or above the median with net income in the bottom quarter cut the fixed-January walk-forward by 1.49pp vs the Oct 22 rules. |
| Admission screen: year-over-year net margin, bottom 4%, on the Oct 22 rules | +9.64pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year net margin, bottom 4% added +9.64pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (maximum drawdown and the 2022-onward era), so it was rejected. |
| Admission screen: year-over-year revenue growth, top 4%, on the Oct 22 rules | -96.57pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by year-over-year revenue growth, top 4% cut the fixed-January walk-forward by 96.57pp vs the Oct 22 rules. |
| Admission screen: two quarters in a row of falling year-over-year net income, on the Oct 22 rules | +15.33pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by two quarters in a row of falling year-over-year net income added +15.33pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (maximum drawdown and the 2022-onward era), so it was rejected. |
| Admission screen: asset growth in the top quarter with revenue at or below the median, on the Oct 22 rules | -7.15pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by asset growth in the top quarter with revenue at or below the median cut the fixed-January walk-forward by 7.15pp vs the Oct 22 rules. |
| Admission screen: unusually high trading volume, top 4%, on the Oct 22 rules | -9.60pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by unusually high trading volume, top 4% cut the fixed-January walk-forward by 9.60pp vs the Oct 22 rules. |
| Admission screen: unusually low trading volume, bottom 4%, on the Oct 22 rules | -20.77pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by unusually low trading volume, bottom 4% cut the fixed-January walk-forward by 20.77pp vs the Oct 22 rules. |
| Admission screen: slowing momentum, bottom 4%, on the Oct 22 rules | -52.85pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by slowing momentum, bottom 4% cut the fixed-January walk-forward by 52.85pp vs the Oct 22 rules. |
| Admission screen: one-month return, top 4% (exhaustion), on the Oct 22 rules | -66.12pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by one-month return, top 4% (exhaustion) cut the fixed-January walk-forward by 66.12pp vs the Oct 22 rules. |
| Admission screen: daily high-low range relative to price, top 4%, on the Oct 22 rules | -73.58pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by daily high-low range relative to price, top 4% cut the fixed-January walk-forward by 73.58pp vs the Oct 22 rules. |
| Admission screen: a rising one-month price on falling volume, on the Oct 22 rules | -81.53pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by a rising one-month price on falling volume cut the fixed-January walk-forward by 81.53pp vs the Oct 22 rules. |
| Admission screen: thin dollar volume among strong-momentum names, bottom quarter, on the Oct 22 rules | +0.31pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by thin dollar volume among strong-momentum names, bottom quarter added +0.31pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (the rolling four-year median and the year-by-year walk-forward), so it was rejected. |
| Admission screen: down-day versus up-day volatility, top 4%, on the Oct 22 rules | +1.16pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by down-day versus up-day volatility, top 4% added +1.16pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (the rolling four-year median and the year-by-year walk-forward), so it was rejected. |
| Admission screen: net signed volume, bottom 4%, on the Oct 22 rules | +1.16pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by net signed volume, bottom 4% added +1.16pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (the rolling four-year median and the year-by-year walk-forward), so it was rejected. |
| Admission screen: price far below the 63-day high (broken momentum), bottom 4%, on the Oct 22 rules | +3.55pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by price far below the 63-day high (broken momentum), bottom 4% added +3.55pp to the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks (the rolling four-year median and the year-by-year walk-forward), so it was rejected. |
| Admission screen: accelerating momentum, top 4%, on the Oct 22 rules | -29.69pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by accelerating momentum, top 4% cut the fixed-January walk-forward by 29.69pp vs the Oct 22 rules. |
| Admission screen: turnover (dollar volume relative to market value), top 4%, on the Oct 22 rules | -101.60pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by turnover (dollar volume relative to market value), top 4% cut the fixed-January walk-forward by 101.60pp vs the Oct 22 rules. |
| Admission screen: weak one-week return among strong-momentum names, bottom quarter, on the Oct 22 rules | -69.36pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by weak one-week return among strong-momentum names, bottom quarter cut the fixed-January walk-forward by 69.36pp vs the Oct 22 rules. |
| Admission screen: weak open-to-close moves, bottom 4%, on the Oct 22 rules | -20.08pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Screening out names by weak open-to-close moves, bottom 4% cut the fixed-January walk-forward by 20.08pp vs the Oct 22 rules. |
| Combined screen: the leading quality screen plus an inventory-coverage screen, on the Oct 22 rules | +14.43pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Combining the leading quality screen of that week with an inventory-coverage screen measured +14.43pp on the fixed-January walk-forward vs the Oct 22 rules but failed the second-stage checks, so it was rejected. |
| Combined screen: the leading quality screen plus a debt screen (bottom 10%), on the Oct 22 rules | +23.92pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Combining the leading quality screen with a debt screen measured +23.92pp on the fixed-January walk-forward vs the Oct 22 rules but failed the rolling four-year median and year-by-year walk-forward checks. |
| Combined screen: the leading quality screen plus inventory-coverage and debt screens, on the Oct 22 rules | +20.41pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Combining all three screens measured +20.41pp on the fixed-January walk-forward vs the Oct 22 rules but failed the rolling four-year median, year-by-year walk-forward and headline-model breach checks. |
| Combined screen: debt to assets relative to industry peers, top 10%, on the Oct 22 rules | +7.66pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Adding a peer-relative debt-to-assets screen measured +7.66pp on the fixed-January walk-forward vs the Oct 22 rules but was rejected. |
| Combined screen: net income relative to debt, bottom 10%, on the Oct 22 rules | +11.53pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Adding a net-income-to-debt screen measured +11.53pp on the fixed-January walk-forward vs the Oct 22 rules but was rejected. |
| Combined screen: cash relative to debt, bottom 10%, on the Oct 22 rules | +0.00pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Adding a cash-to-debt screen changed nothing (+0.00pp on the fixed-January walk-forward vs the Oct 22 rules), so it was rejected as inert. |
| Combined screen: interest coverage relative to industry peers, bottom 10%, on the Oct 22 rules | +21.38pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Adding a peer-relative interest-coverage screen (operating profit over interest expense) measured +21.38pp on the fixed-January walk-forward vs the Oct 22 rules but was rejected. |
| Combined screen: the leading quality screen plus a peer-relative inventory-coverage screen, on the Oct 22 rules | +25.47pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | This combination measured +25.47pp on the fixed-January walk-forward vs the Oct 22 rules (the candidate's walk-forward was 197.07%, just short of 200%) but failed the rolling four-year median, year-by-year walk-forward and headline-model breach checks. |
| Denser strict combinations of the quality and peer-relative inventory screens, on the Oct 22 rules | +23.15pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Stricter, denser combinations of these screens lifted the candidate's walk-forward to about 194-197% (the best +23.15pp vs the Oct 22 rules) but failed the rolling four-year median and headline-model breach checks; the best candidate that passed in this round was a strict balance-sheet screen at 195.93%, still short of 200%. |
| Strict high-quality combinations that reached about 204% but failed the risk checks, on the Oct 22 rules | +32.80pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Strict combinations of the quality, inventory-coverage and large-company screens reached a walk-forward of 203.1-204.4% (the best +32.80pp vs the Oct 22 rules) but failed the rolling four-year median and headline-model breach checks; the version that passed every check was softer, at 200.88%. |
| Earnings-floor and cash-flow screens added to the leading candidate, on the Oct 22 rules | -8.00pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Added to the leading candidate of the time (the balance-sheet and accrual screen, 212.40%), an earnings-floor screen still passed at 204.92% but gave up about 8pp, and strict free-cash-flow and cash-conversion screens failed the headline-model breach check: these families do not add return. |
| Market-wide 200-day trend gate: the whole strategy (stocks and the leveraged Nasdaq sleeve) moved to cash whenever the S&P 500 ETF closed below its 200-day moving average, on the Oct 14-22 rules | -47.46pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | Going fully to cash whenever SPY closed below its 200-day moving average (the classic Faber / Paul Tudor Jones rule) cut CAGR from 162.0% to 114.5% and the fixed-January walk-forward from 160.4% to 105.2%, and did not reduce the worst drawdown at all (-52.0%, a 2021 leader drawdown while SPY was in an uptrend): the gate was off 16% of the time and sat out the sharp rebounds the book earns most in (2020, 2022-23, 2025); the same rule on QQQ was worse on drawdown (-65.5%). On SPY alone the rule kept its textbook trade, 9.9% a year at -20.9% worst drawdown against buy-and-hold's 14.4% at -33.7%. |
| Leadership handoff after a bad earnings reaction: the strategy's largest holding dropped from the stock book for 20 sessions after its share price fell relative to the Nasdaq on its earnings release, so the book rotates into the next leader, on the Oct 14-22 rules | -14.16pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | When the book's largest name reacted badly to its earnings release (the strategy already trims the whole stock book then), also dropping that company for the next 20 sessions so the book rotated into the next-ranked leader cut CAGR from 162.0% to 147.8% and the fixed-January walk-forward from 160.4% to 148.3% (122.0% vs 128.2% with 2020 left out), deepened the worst drawdown (-55.1% vs -52.0%) and cost 32pp in the 2022-onward era; it won only 38% of the rolling four-year windows, and the same exclusion moved 40 sessions later lost about as much (walk-forward 150.5%), so the cost is the forced rotation out of the leader rather than anything tied to the release. |
| Fundamentally confirmed leadership score: the strategy's trend ranking blended half and half with a ranking of each company's past-year revenue and earnings growth as first reported, on the Oct 14-22 rules | -78.27pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Ranking names half on the strategy's own price trend and half on how fast their revenue and earnings had grown over the past year (as first reported, used only after each filing), to pick fundamentally backed leaders earlier, shared fewer than half of its top four candidates with the trend ranking and cut CAGR from 162.0% to 83.7% and the fixed-January walk-forward from 160.4% to 82.1%, with a deeper worst drawdown (-67.2% vs -52.0%); it won none of the rolling four-year windows, and the same blend with the growth half held neutral cost only 5pp, so the loss came from the growth ranking itself. |
| Fed balance-sheet expansion as a TQQQ re-entry state: the leveraged Nasdaq sleeve switched back on while its volatility gate is off whenever the Federal Reserve's total assets had grown more than 5% over the last 13 weeks, on the Oct 14-22 rules | +2.21pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Concentration (a few names or years carry the result) (also: Failed the walk-forward) | Switching the leveraged Nasdaq sleeve back on during a volatility exit whenever the Federal Reserve's balance sheet was expanding fast (more than 5% in 13 weeks, read only after each weekly release) added 2.2pp of CAGR (164.2% vs 162.0%) and 2.3pp to the fixed-January walk-forward (162.7% vs 160.4%), but every re-entry inside the walk-forward windows fell in the spring 2020 crash: with 2020 left out the walk-forward is unchanged to the decimal, and the change won only 44% of the rolling four-year windows (it lost none; the rest were identical), short of the 60% required. |
| Systematic-share sizing: the stock book scaled between 0.75x and 1.25x by how much of its holdings' last quarter of moves was market- and factor-driven rather than unexplained, on the Oct 14-22 rules | -10.88pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the exposure-matched check (also: Failed the walk-forward) | Sizing the stock book up when its holdings' recent gains were mostly explained by market factors (expected to continue) and down when they were mostly unexplained (expected to reverse) cut CAGR from 162.0% to 151.1% and the fixed-January walk-forward from 160.4% to 149.8%, below a constant scale of the same average size (159.1%), with a deeper worst drawdown (-54.5% vs -52.0%) and 16 margin breaches on the headline model against 6; it won none of the rolling four-year windows. |
| Systematic-only momentum: the strategy's trend ranking replaced by each name's past six months of market- and factor-driven return only, with the unexplained part left out, on the Oct 14-22 rules | -106.63pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Ranking names by the part of their last six months of returns explained by five statistical market factors (and by earnings-release days), instead of the strategy's own trend measure, a test of the research finding that such moves continue while unexplained moves reverse, cut CAGR from 162.0% to 55.4% and the fixed-January walk-forward from 160.4% to 51.2%, with a deeper worst drawdown (-63.2% vs -52.0%); it beat the strategy in under 1% of rolling four-year windows. |
| Cash-burn exclusion: names whose cash share of assets fell most over the past year (the bottom fifth of the universe) made ineligible, on the Oct 14-22 rules | -57.92pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Excluding, every day, the fifth of the universe whose cash had fallen most as a share of assets over the past year (a genuine cash-flow signal, not disguised asset growth) changed the lead name on about half of rebalances and cut CAGR from 162.0% to 104.1% and the fixed-January walk-forward from 160.4% to 114.8%, with a deeper worst drawdown (-76.1% vs -52.0%) and one more margin breach on the headline model; it trailed the strategy in 11 of 17 years. |
| Credit-conditioned cash filter: the strategy's cash-and-distress screen applied only while the Baa corporate spread is above its one-year median, on the Oct 14-22 rules | -40.68pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Switching the cash-and-distress screen off whenever credit was easier than its one-year median (57% of days) changed the lead name on a third of rebalances and cut CAGR from 162.0% to 121.3% and the fixed-January walk-forward from 160.4% to 133.2%, with a deeper worst drawdown (-67.6% vs -52.0%); it trailed in 14 of 17 years and won 3% of rolling four-year windows. |
| Volatility-conditioned growth filter: the strategy's revenue-acceleration screen applied only while VIX is below its one-year median, on the Oct 14-22 rules | -68.61pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Switching the revenue-acceleration screen off whenever VIX was above its one-year median (44% of days), a re-test of an earlier regime-switched quality filter that had passed a walk-forward on an older version of the strategy, cut CAGR from 162.0% to 93.4% and the fixed-January walk-forward from 160.4% to 66.8%, and deepened the worst drawdown to -93.4%; 2020 alone was 513 points behind. |
| Both screens state-conditioned: the cash-and-distress screen on credit stress and the revenue-acceleration screen on VIX, together, on the Oct 14-22 rules | -88.50pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Conditioning both screens at once cut CAGR from 162.0% to 73.5% and the fixed-January walk-forward from 160.4% to 62.9%, with a worst drawdown of -87.3%, and lost every rolling four-year window: the two conditionings did not offset each other. |
| Rate-duration exclusion: names whose returns fall most when 10-year yields rise (the bottom fifth of the universe's market-adjusted beta to yield changes) made ineligible, on the Oct 14-22 rules | -86.88pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Excluding, every day, the fifth of the universe whose returns had fallen most with rising 10-year yields over the past year changed the lead name on about two thirds of rebalances and cut CAGR from 162.0% to 75.1% and the fixed-January walk-forward from 160.4% to 61.4%; the worst drawdown deepened from -52.0% to -81.1%, and it did not even protect the book in 2022, the rate shock it was built around (-44.4 points that year), trailing the strategy in 15 of 17 years and every rolling four-year window. |
| Dollar-volume lead: the stock sleeves' market-cap weights re-sized by each name's 60-session median dollar volume instead of its market value, on the Oct 14-22 rules | +1.82pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | Keeping the same selected names and the same concentration, but sizing them by recent trading value instead of market value, changed the lead name on 35% of rebalances and added 1.8pp of CAGR (163.8% vs 162.0%), yet the fixed-January walk-forward fell 9.2pp (151.2% vs 160.4%), 2022 onward fell 48.4pp, the worst drawdown deepened from -52.0% to -65.6% (-66.2% on the headline model) and headline margin breaches rose from 6 to 17; it trailed most in 2020 and 2024. |
| Revenue lead: the stock sleeves' market-cap weights re-sized by each name's trailing-four-quarter revenue instead of its market value, on the Oct 14-22 rules | -100.99pp CAGR vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | Sizing the same selected names by trailing revenue (as first reported, point in time) instead of market value changed the lead name on 63% of rebalances and cut CAGR from 162.0% to 61.0% and the fixed-January walk-forward from 160.4% to 48.9%; it was pre-registered also as a drawdown guard, but the worst drawdown deepened from -52.0% to -67.8%, and it trailed the strategy in 13 of 17 calendar years and in every rolling four-year window. |
| Item 1.02 exclusion (252 sessions) on the Oct 13 starting rules | +4.14pp CAGR vs the Oct 13 starting rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Concentration (a few names or years carry the result) (also: Negative in the fresh era (2022 onward)) | Excluding Item 1.02 names for 252 sessions added +4.14pp CAGR on the Oct 13 starting rules (111.36% vs 107.22%), but fresh 2022+ was -0.42pp and leave-one-ticker-out showed the gain was an ALGN slot-reallocation artifact of the market-cap weighting plus NVDA (excluding ALGN keeps 44% of the gain, NVDA 58%, MU 82%), with DSR 0.007. |
| Item 3.02 unregistered-sale exclusion on event-rotation | +0.66pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Dropped by decision (also: Concentration (a few names or years carry the result)) | Excluding Item 3.02 names added +0.66pp CAGR and +1.04pp fixed-January walk-forward on the event-rotation finalist, but AMD carried 88.3% and MELI 87.0% of the gain in leave-one-ticker-out (each individually necessary) and the shifted placebo read z +0.42 (8 of 24 displaced masks match or beat the treatment); dropped by decision on 2026-10-13. |
| Stale high-coverage negative-surprise avoidance on the Oct 14-22 rules | -6.14pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Avoiding stale high-coverage negative earnings surprises cost -6.14pp CAGR on the Oct 14-22 rules (fixed-January walk-forward 103.67% vs 115.13%, rolling four-year mean -9.06pp, fresh 2022+ -6.29pp, protocol WF picked delta -9.50pp), confirming the rejection already recorded on Oct 11 rules. |
| Congressional-cluster exclusion (3+ members buying, 30 days) on Oct 11 rules | -3.37pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | Excluding congressional-cluster names cost -3.37pp CAGR on Oct 11 rules (92.88% vs 96.25%) with MaxDD -59.60% vs -57.38% and -4.09pp on the 2012-07-27 data sub-period; an independent check on the Oct 11 rules build read +0.1pp full-period with fresh 2022+ -4.1pp and January walk-forward 72.3% vs 72.4%. |
| Negative EPS surprise exclusion; SUE variant | +0.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | Excluding negative EPS surprises added +0.2pp CAGR on Oct 11 rules (SUE variant +2.9pp), but fresh 2022+ was -3.5pp (SUE -4.0pp) and January walk-forward read 69.0% / 70.6% vs the 72.4% control, with DSR 0.02 / 0.09. |
| Congressional-cluster plus negative-EPS-surprise combo exclusion | +1.50pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The congressional-cluster plus EPS-surprise combination added +1.5pp CAGR on Oct 11 rules (Robinhood; +4.1pp on the headline model), but fresh 2022+ was -0.9pp, January walk-forward read 70.3% vs 72.4%, and every 4y-fold walk-forward fold that picked it lost on test (-4.7 / -0.6 / -3.7pp). |
| High-coverage consensus-miss exclusion | -1.60pp CAGR vs Oct 11 rules (independent-check build); window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed independent replication | The high-coverage consensus-miss exclusion lost -1.6pp CAGR on Oct 11 rules (88.2% vs 89.8%) with train -4.7pp and January walk-forward 74.8% vs 75.6% (picked 1/12 folds), and the originating claim (99.42% CAGR, walk-forward +12.10pp) does not reproduce on this build. |
| Large negative earnings-reaction exclusion (prior-day reaction <= -10%) | +5.70pp CAGR vs the Oct 13 starting rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Concentration (a few names or years carry the result) | Excluding names after a large negative earnings reaction (EAR <= -10% vs SPY, 20 sessions) added +5.7pp CAGR on the Oct 13 starting rules (112.9% vs 107.2%), but MaxDD deepened to -64.3% vs -59.4%, the -10% threshold was a spike (the -7% and -15% neighbours read -0.1pp and -0.7pp, and the spike point's own fresh era was negative), and the sold names beat SPY by +2.5% over the next 20 sessions with a per-exit edge median of -2.2%. |
| Held-name GAF keep rule (sell a held name the day it loses GAF) | -20.80pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | Keeping held names that lose GAF eligibility cost -20.8pp CAGR vs Oct 11 rules (73.9% vs 94.8%) with fresh 2022+ -37.0pp, every neighbour and every anchor losing and DSR 0.00; the mask pays as an admission rule and destroys value as an exit rule. |
| Margin-binding fill priority (cause-aware fills when margin binds) | -3.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Cause-aware fill priority when margin binds cost -3.2pp CAGR vs Oct 11 rules at the primary (91.6% vs 94.8%) with a 4y win rate of 1% and switch score 1/10, and the plain binding-day variant (+0.3pp) also failed. |
| Receivables / net-working-capital rank tilt | -29.30pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed independent replication | The receivables / net-working-capital rank tilt lost -29.3pp CAGR vs the Oct 11 rules in the independent implementation (65.5% vs 94.8%), not reproducing the original +9.64pp rolling four-year mean claimed in-sample; an exact second build traced 29.1 of the 29.4pp gap to switching from percentile ranks to raw z-scores, and that second build read +0.1pp with the walk-forward picking the control in 4 of 4 folds. |
| Receivables-level rank tilt | +10.51pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward (also: Concentration (a few names or years carry the result)) | The receivables-level rank tilt gained +10.51pp in the rolling four-year mean in-sample vs the Oct 11 rules, but the walk-forward picked it in only 3 of 5 folds against a 4-of-5 bar, with its +8.99pp out-of-sample mean concentrated in the fourth and fifth folds (+33.68/+12.03pp), and the third fold's pick lost -0.74pp. |
| Bipower (jump-robust) volatility target | -1.00pp CAGR vs the Oct 13 starting rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | Replacing the realized-vol estimator with a jump-robust bipower estimator cost -1.0pp CAGR vs the Oct 13 starting rules (106.2% vs 107.2%) with fresh 2022+ -3.5pp, every variant losing and DSR 0.017. |
| Correction-state rotation pause (corr-pause) | +1.00pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Concentration (a few names or years carry the result) | Pausing event-row rank exits in correction states added +1.00pp CAGR vs the event-rotation finalist (123.81% vs 122.81%; fixed-Jan 115.2%), but the neighbours failed (fast-10 -2.09pp, fast-42 -0.08pp, SPY-reference +0.24pp), 4y windows failed, the literal GHM split read -7.16pp, and the gain was five episodes with 2020 alone contributing -27.0pp from a paused switch (DSR 0.010). |
| Event rotation on the buffer-7 book | -4.78pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Running the event-rotation event rotation on the buffer-7 book lost -4.78pp CAGR vs the Oct 14-22 rules (119.66% vs 124.43%); fixed-January walk-forward rose +1.10pp to 116.23% but the adoption battery failed (rolling four-year median -10.39pp, MaxDD -64.27%, headline-model breaches 4 vs 2, train -11.12pp against fresh +17.09pp), so the live buffer stays. |
| Event-row local re-weighting (cross-check) | -3.30pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Mechanism overlap | Re-weighting locally at event rows lost -3.30pp CAGR vs the finalist (121.13% vs 124.43%) with fixed-January walk-forward 108.39% (-6.74pp) and fresh 2022+ -13.20pp, even though it retained 80.6% of the finalist's edge over the Oct 13 starting rules and shallowed MaxDD by 8.58pp: full re-concentration at event rows was load-bearing, not a cost. |
| GAF/DC mask dated at the 8-K release, on the Oct 14-22 rules | -19.66pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Dating the GAF/DC mask at the 8-K release instead of the filing date cost -19.66pp CAGR on the Oct 14-22 rules (104.49% vs 124.15%) with fresh 2022+ -32.81pp, switch score 0/10, anchors -17 to -21pp and DSR 0.000; filing-date availability is the correct dating. |
| GAF/DC mask dated at the 8-K release, original Oct 11 rules study | +3.40pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward (also: Failed the placebo check) | Dating the GAF/DC mask at the 8-K release added +3.4pp CAGR headline on Oct 11 rules (98.2% vs 94.8%) but failed the drawdown leg (MaxDD -62.6% vs -57.3%) and the causal-family January walk-forward (76.9% vs 80.2%), while its strong neighbours (max30 +7.9pp, max15 +2.6pp) failed the shuffle check as look-ahead artefacts. |
| More / faster stock sleeves on the Oct 14-22 rules | -1.72pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | More or faster stock sleeves lost -1.72pp CAGR at the primary on the Oct 14-22 rules (122.43%), every variant lost full-period CAGR (-0.57 to -5.78pp), the family January walk-forward picked Surge in 11/12 folds, and DSR was <= 0.008; four sleeves at the current cadence is the best point in the tested family. |
| event-rotation sibling families (dominated rotation forms) | No single primary delta (the sources record a range across variants) | Mechanism overlap | The dominated event-rotation rotation families all lost to the adopted conjunction (a rank buffer, a statistical margin on the replacement test, eligibility rules and rank-based exits) on the decision metrics, vs the Oct 13 starting rules: exit persistence -1.7 to -12.8pp CAGR at every K/B, release-day-only checks -5.6 to -14.0pp, the conjunction -4.4pp, and sleeve-count variants -5.2 to +0.8pp. |
| 52-week-high proximity (filter and rank tilt) | -11.50pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The 52-week-high proximity filter cost -11.5pp CAGR vs Oct 11 rules (83.3% vs 94.8%) and the rank tilt -19.4pp (train -25.7pp), with every proximity threshold losing (-2.2 to -20.2pp). |
| ATR trailing stops | -29.50pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | ATR trailing stops cost -29.5pp CAGR vs Oct 11 rules in the all-held-names form (65.3%) and -31.1pp in the top-name-only form, with fresh 2022+ -59.5pp / -63.1pp and roughly 1,600 whipsaw exits in the all-held-names form. |
| Index-inclusion drift A (pre-inclusion admission) | +0.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Concentration (a few names or years carry the result) | Admitting names shortly before index inclusion added +0.2pp CAGR vs Oct 11 rules, but the entire effect was one year (2024 +36.9pp) and January walk-forward was a tie. |
| Index-inclusion drift B (post-inclusion penalty) | -10.00pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Penalising recently included names cost -10.0pp CAGR vs Oct 11 rules with a 4y win rate of 1%, and every penalty size lost (-4 to -13pp). |
| Index-membership seasoning and lingering | -10.62pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Excluding a name's first 63 sessions in the index (season63) cost -10.62pp rolling four-year mean vs Oct 11 rules (the displaced placebo, sessions 126-189, cost -4.52pp, so about -6pp is event timing), while extending eligibility 63 sessions past removal (linger63) was noise at +0.30pp rolling four-year mean. |
| Multi-horizon return rank | -2.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The multi-horizon return rank lost -2.2pp CAGR at the primary vs Oct 11 rules, and its post-hoc small tilts (+5.9/+6.7pp) existed only at the live anchor, reading -1.5 to +2.4pp across shifted anchors with train negative in 9 of 12 shifted cells. |
| Conviction concentration (raise the weight of top-conviction names) | -28.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Concentrating into top-conviction names cost -28.2pp CAGR vs Oct 11 rules, every variant lost (-19.7 to -35.2pp), and the 4y win rate was 3%; the current market-cap weighting already concentrates as much as pays. |
| Prior-year winners satellite | +1.70pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Concentration (a few names or years carry the result) | The prior-year winners satellite added +1.7pp CAGR vs Oct 11 rules but failed the drawdown leg (MaxDD -59.8% vs -57.3%), its response was monotone in k (a leverage dial, not a selection effect), and family WF picked losers, then max-k. |
| Universe breadth brake | -4.80pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The universe breadth brake cost -4.8pp CAGR vs Oct 11 rules, every neighbour lost, and the cost concentrated in V-shaped recoveries (2020: -68pp of that year). |
| Entry-timing filter (skip new buys when the 5-day average is falling) | -10.00pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | Skipping new buys while a name's 5-day average was falling cost -10.0pp CAGR vs the Oct 11 rules, -18.1pp from 2022 onward, and the reverse rule lost too (-23.0pp). |
| QQQ trend gate for the TQQQ sleeve (AGITQ) | No single primary delta (the sources record a range across variants) | Failed the walk-forward | QQQ trend-gate variants for the TQQQ sleeve lost -2.2 to -4.8pp CAGR vs Oct 11 rules, and walk-forward picked the control 12/12 in every family. |
| Trim top name before earnings | -16.10pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | Trimming the top name before earnings cost -16.1pp CAGR vs Oct 11 rules with fresh 2022+ -30.5pp, and every neighbour lost. |
| Insider-buying clusters | +0.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | Insider-buying clusters added +0.2pp CAGR vs Oct 11 rules with fresh 2022+ -2.4pp, and the signal touched 0.002% of book weight; Form 4 cluster forms tested separately (batches 40/41) were flat in avoidance form and catastrophic in require form (-27.5 to -39.8pp). |
| 13F institutional-ownership tiebreaker | -6.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The 13F institutional-ownership tiebreaker cost -6.2pp CAGR vs Oct 11 rules, and every neighbour lost. |
| Buyback shrink tiebreaker | -4.30pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The buyback shrink tiebreaker cost -4.3pp CAGR vs Oct 11 rules, and every neighbour and every anchor lost. |
| Supplier-to-customer spillover tiebreaker | -3.10pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | The supplier-to-customer spillover tiebreaker cost -3.1pp CAGR vs Oct 11 rules with fresh 2022+ -9.8pp, and the dose-response was monotone worse in the spillover weight. |
| EDGAR lazy prices (filing-attention tiebreaker) | -25.50pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The EDGAR lazy-prices tiebreaker cost -25.5pp CAGR vs Oct 11 rules with a 4y win rate of 7%. |
| Peer lead-lag tiebreaker | -5.90pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The peer lead-lag tiebreaker cost -5.9pp CAGR vs Oct 11 rules, and both signs lost. |
| Volume-confirmed momentum tiebreaker | -15.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | The volume-confirmed momentum tiebreaker cost -15.2pp CAGR vs Oct 11 rules with fresh 2022+ -19.3pp, and every form lost. |
| Parking in gold (GLD) while the gate is off (family) | +1.50pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed independent replication (also: Negative in the fresh era (2022 onward)) | Parking in GLD while the gate was off (first form) passed the originator's bar narrowly (+1.5pp CAGR vs the Oct 11 rules), but independent validation read +2.24pp out-of-sample in 3 of 5 folds, below the +3pp bar, with a worse maximum drawdown; the form that held GLD only above its 200-day average read +0.4pp, the Treasury-bond (TLT) forms +0.6/+0.6pp with -2.0/-1.0pp from 2022 onward, and the blends +1.1/+1.0pp. |
| BTAL hedge satellite | +1.00pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Concentration (a few names or years carry the result) | The BTAL hedge satellite added +1.0pp CAGR at the primary vs Oct 11 rules, but two-way walk-forward lost at every anchor and half the gain was the 2020 spell. |
| QQQ up-streak / IBS satellite; TRIN oversold satellite | +0.00pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The QQQ up-streak / IBS satellite added +0.0pp CAGR vs Oct 11 rules (the signal fires 0.84 times per year, too rare to matter at any tested size) and the TRIN oversold satellite lost -0.6pp. |
| Defensive satellites (VRP; defensive TSMOM sleeve; GLD volume-breakout; post-bear re-risk; momentum-factor trend brake) | -0.30pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | Five defensive satellites all lost vs Oct 11 rules: VRP -0.3pp, defensive TSMOM sleeve -2.1pp, GLD volume-breakout -2.0pp, post-bear re-risk -0.1pp and momentum-factor trend brake -5.8pp CAGR, with fresh 2022+ deltas of -2.1 / -5.3 / -3.5 / -3.0 / -9.5pp respectively. |
| High-water-mark circuit-breaker overlay | -15.40pp walk-forward out-of-sample mean vs retired base; window 2010-02-11 to 2026-09-18; the earlier, larger starting account | Failed the walk-forward | The high-water-mark circuit-breaker overlay lost -15.4pp in walk-forward out-of-sample mean CAGR (46.71% vs the base's 62.08%); training-only selection picked the same variant (a 35% trigger with a 30% repair) in 5 of 5 folds and it still lost out-of-sample, while in-sample all 9 trigger-and-repair variants lost 23-37pp CAGR for at most 7pp of drawdown relief. |
| Sleeve-structure variants on Oct 11 rules | -5.49pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Across 22 sleeve-structure variants in batches 49-50, none cleared the +3pp bar vs Oct 11 rules: TQQQ/QQQ barbell -5.49pp rolling four-year mean, TQQQ vol target -2.45pp, SSO parking -2.11pp, stock-book state trims -1.47 to -3.17pp, sleeve sub-state refinements -0.15 to -0.70pp, and never-fully-exit-the-sleeve -8.68pp. |
| Short sleeves A and B (weakest-names short book) | -1.00pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | The always-on short sleeve (A) lost -1.0pp CAGR vs Oct 11 rules (93.8% vs 94.8%) with every k/N/rank neighbour losing on the full period and on train, and the gate-off-only sleeve (B) lost -0.2pp and was negative in 2022+ at all anchor offsets. |
| State gates on the TQQQ sleeve (7 families, 24 registered forms) | -10.18pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Seven families of TQQQ-sleeve state gates produced no viable form in 24 registered forms: main-form changes in the rolling four-year mean vs the Oct 11 rules were high-yield credit spread level -10.18pp, the VIX against 3-month VIX term structure -9.82pp, the Cboe SKEW index +0.15pp, VVIX -7.19pp, the MOVE bond-volatility index -4.99pp, market breadth -5.35pp and the 10-year minus 2-year Treasury spread -5.19pp, with the reverse rules worse than the main forms in every family. |
| Price-path and risk-shape tilts (18 constructs) | No single primary delta (the sources record a range across variants) | Failed the walk-forward | Eighteen price-path and risk-shape tilts all failed vs Oct 11 rules: the first batch's 12 tilts read -15.44 to -39.21pp rolling four-year mean, close-location persistence -18.53pp (prefer-high) / -18.05pp (prefer-low), overnight-gap asymmetry -31.89 / -35.07pp and up-day balance -22.54 / -40.27pp, none within 15pp of the +3pp bar, so no walk-forward was run. |
| Rank-age and clock tilts (trend age; information age; earnings-clock phase) | -40.25pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Rank-age and clock tilts failed in both directions vs Oct 11 rules: trend age -40.25pp (prefer-young) / -22.58pp (prefer-old) rolling four-year mean, information age -29.01 / -21.64pp, and earnings-clock phase -31.37 / -22.08pp. |
| Dollar-index and put/call market gates | -0.14pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The dollar-index gate was a wash vs Oct 11 rules (-0.14pp rolling four-year mean, binding on 6.5% of defined sessions, MaxDD -59.25% vs -57.38%) and the aggregate put/call gates lost in both directions (-5.01pp contrarian, -8.47pp calm) on the covered span (the put/call archive ends 2019-10-04). |
| EDGAR 8-K event exclusions (non-distress items) | -14.01pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | EDGAR 8-K non-distress exclusions lost vs Oct 11 rules: Item 5.02 departures -14.01pp rolling four-year mean (binding on 16.5% of eligible cells, mostly routine governance), Item 1.01 agreements -12.46pp and Item 2.01 completions -3.73pp. |
| EDGAR regulatory-distress and timeliness exclusions (Items 4.02 / 4.01 / 3.01 / 1.03; NT and late filings) | +0.00pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | EDGAR distress and timeliness exclusions were inert or negative vs Oct 11 rules: Items 4.02, 1.03 and NT/late filings read +0.00pp rolling four-year mean (bit-identical to control, because a flagged name is never at the top of the rank inside the buffer), Item 4.01 -0.01pp, and Item 3.01 -0.74pp CAGR with MaxDD -65.37% vs -57.38%. |
| Schedule 13D / 13G eligibility boost | -1.55pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The Schedule 13D / 13G eligibility boost lost vs Oct 11 rules (13D -1.55pp, 13D/A -1.10pp, 13G -1.94pp rolling four-year mean): the boost binds but the admitted names underperform the names they displace, and tails worsen. |
| Event sleeves on the early-October combined base (PEAD; 13D activist; 13G passive) | -9.34pp rolling four-year mean CAGR vs matched 5-sleeve all-momentum control; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Event sleeves on the early-October combined base all lost vs the matched 5-sleeve all-momentum control (92.06%): PEAD -9.34pp, 13D activist -12.31pp and 13G passive -9.46pp rolling four-year mean, negative across the whole rolling distribution (medians -7.46 / -13.56 / -11.48pp) with pick overlap of only 0.4-3.7%. |
| Corporate-action exclusions (corporate-actions feed) | -9.67pp rolling four-year mean CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Excluding names after corporate actions from the corporate-actions feed all lost vs the Oct 11 rules: forward splits -9.67pp in the rolling four-year mean, dividend initiations -3.70pp, dividend increases -11.78pp, cuts or omissions -9.98pp, high trailing yield -10.22pp and low trailing yield -14.66pp, with shifted-date placebos (-1.62pp splits, -3.46pp negative events) showing the damage comes from the events themselves. |
| Universe expansion to liquid names outside the index | -29.00pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Expanding the universe to liquid names outside the index cost -29.0pp CAGR vs the Oct 11 rules in the main form (72.2%), with maximum drawdown -73.8%, fixed-January delta -25.3pp and the walk-forward picking the candidate in 0 of 12 years; the two other forms were worse (-54.1pp and -42.2pp). |
| Universe widening to the top 1,500 companies by market value | -30.07pp rolling four-year mean CAGR vs the early-October combined base; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Widening the universe to the 1,500 largest companies at each date lost -30.07pp in the rolling four-year mean vs the early-October combined base (CAGR -29.15pp to 69.55%, maximum drawdown -73.55%); the walk-forward picked the candidate in 0 of 5 folds, training CAGR was lower in all 5 folds, and the control rebuilt through the extended pipeline reproduced to within 0.004pp. |
| PIT top-500 universe trim | -21.40pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | The PIT top-500 universe trim lost -21.40pp CAGR on the Oct 14-22 rules (103.03% vs 124.43%; fixed-January walk-forward 101.02% vs 115.13%) after failing the protocol WF in its first test (picked 2/5 folds, out-of-sample picked delta -5.57pp, gain era-concentrated); on Surge it also cleared all margin breaches (0/0/0) and shallowed MaxDD by 6.94pp, a genuine risk-reducer at the price of a fifth of the return. |
| Winner-mimicry (kNN 'looks like a past winner') and concentration requirements | -22.46pp rolling four-year mean CAGR vs retired base; window 2010-02-11 to 2026-09-18; the earlier, larger starting account | Failed the walk-forward | Winner-mimicry and winner-profile requirements failed vs the retired base: kNN 'looks like a past winner' forms read -22.46 to -35.15pp rolling four-year mean across 7 forms, and the first-profit, turnaround and jumper requirements cost -42.85pp, -36.54pp and -28.07pp. |
| Winner-autopsy masks (operating leverage; momentum-earnings divergence) | -3.50pp rolling four-year mean CAGR vs retired base; window 2010-02-11 to 2026-09-18; the earlier, larger starting account | Failed the walk-forward | Both masks proposed by the top-20 winner autopsy failed in-sample vs the retired base, so no walk-forward was run: the operating-leverage mask -3.50pp rolling four-year mean (-6.15pp CAGR) and the momentum-earnings-divergence mask -22.23pp (-28.24pp CAGR, MaxDD -78.6%). |
| Published factors (Whited-Wu; org capital; Amihud) | -41.60pp rolling four-year mean CAGR vs published-factor test base; window 2010-02-11 to 2026-09-18; the earlier, larger starting account | Failed the walk-forward | Published factor batch 14 failed on the published-factor test base: Whited-Wu -41.6pp rolling four-year mean, org capital -33.6pp and Amihud illiquidity -27.1pp, and the remaining members also failed (R&D growth -21.0pp, coskewness -17.0pp, LT reversal -12.3pp, KZ index -4.6pp). |
| Pair trades (4 families) | +2.60pp rolling four-year mean CAGR: none (absolute; 75% rolling four-year mean adoption bar); window 2010-02-11 to 2026-09-18; the earlier, larger starting account | Failed the walk-forward | Four pair-trade families failed the 75% rolling four-year mean adoption bar: the best form (long-short 12-month momentum) returned +2.6% rolling four-year mean on its own, 72pp short of the bar and 16pp short of QQQ on the same windows, with cointegration pairs -2.0%, reversal -14.2% and sector-neutral -11.0%. |
| Insider-buying spike as a TQQQ FAST re-entry trigger (on the Oct 13 rules) | -2.64pp CAGR vs the Oct 13 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the walk-forward (also: Concentration (a few names or years carry the result)) | Sending the TQQQ sleeve to FAST on an aggregate insider-buying spike lost -2.64pp CAGR vs the Oct 13 rules on the current starting account (135.38% vs 138.02%), with MaxDD 4.41pp deeper (-58.80%), fixed-January walk-forward 124.55% vs 128.72%, switch 4/10 / 5/10 and DSR 0.000 - all ten bar items fail; the loss is the 2020 spell (the trigger bought TQQQ eight sessions before QQQ's closing low; leave-2020-out is +0.41pp), and the literal OR-with-D form is a no-op against D's entry guard. |
| Prior-calendar-year winners satellite revisited on the Oct 13 rules (in-engine) | -1.32pp CAGR vs the Oct 13 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the placebo check | Re-run inside the engine on the Oct 13 rules, the prior-year top-5 winners satellite (weight 0.10, January rebalance) lost -1.32pp CAGR (136.99% vs 138.31%) with train -1.29pp and fresh 2022+ -1.76pp, and the exposure-matched mid-5 placebo (-1.11pp) and the losers sign check (-1.26pp) lost the same amount, so there is no winner effect; MaxDD passed both drawdown bars (-54.36%) but headline-model breaches rose 2 to 3, initial margin bound on 14 of 17 satellite fills (mean buy fraction 0.32, zero in 8 of 17 years), and the 2026-10-12 overlay near-miss on Oct 11 rules was its unlimited margin. |
| DRAM nowcast mask (Taiwan memory-maker revenue) on the Oct 14-22 rules | -0.71pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the placebo check (also: Failed the walk-forward) | Masking memory / storage names while the Nanya + Winbond revenue nowcast decelerated lost -0.71pp CAGR vs the Oct 14-22 rules (123.44% vs 124.15%) with fresh 2022+ -3.63pp, fixed-January walk-forward 113.38% vs 114.91% and switch 4/10 / 4/10, and the 6-months-late placebo gained +6.30pp - more than any real variant (neighbours -10.69 to +6.23pp) - so the deltas are selection-path divergence, not signal timing; the mask did exit MU before its 2024 fall, and the sleeve refill lost as much in the same sell-off. |
| Implied-correlation TQQQ boost, on the event-rotation finalist | +1.75pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Raising the TQQQ sleeve weight from 0.5 to 0.75 while Cboe's three-month implied correlation index sat at or above its trailing 80th percentile gained +1.75pp CAGR (125.90%) but failed the bar: fixed-January walk-forward 114.24% vs 114.91%, a four-year win rate of 39%, switch tests 6/10 and 5/10, and headline-model breaches 8 vs 2; crowded margin meant realised TQQQ exposure rose only +0.09 of the planned +0.25, and the same-exposure swap form lost -1.92pp and failed every check. |
| Flow-pressure tilt (60/40 drift + leveraged-ETF flows) on the Oct 14-22 rules | +0.69pp CAGR vs event-rotation finalist; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) (also: Failed the walk-forward) | Scaling the stock book x1.5 / x0.5 on a combined 60/40-drift and leveraged-ETF flow-pressure state added +0.69pp CAGR vs the Oct 14-22 rules but lost -5.53pp on fresh 2022+, won 21% of 4y windows and read fixed-January walk-forward 111.44% vs 114.91% (7 of 10 bar items missed, DSR 0.000); the flow component alone lost -15.44pp, the TQQQ-k form's +1.95pp was all fresh-era with headline-model breaches 5 vs 2, and on the rules with admission the primary lost -0.36pp (not additive). A post-hoc up-only half (+5.76pp) is a lead for a fresh test, not part of this verdict. |
| X claim: buy last month's top stock, rotate monthly | -4.30pp CAGR: none (absolute standalone; SPY +14.5% on the same dates); window 2010-01-29 to 2026-09-30; the earlier, larger starting account | Failed independent replication | Tested honestly (point-in-time S&P 500 + Nasdaq-100 universe, 5 bps/side), the claimed rule - 100% in the prior month's single best performer, rotated monthly - returned -4.3% CAGR with MaxDD -85.7% (SPY +14.5% / -33.7% on the same dates); the claim's 33.9% comes from a survivorship universe with no costs. The best honest form (top-10) read 14.1% / -42.7%, anchored-Jan walk-forward over the variants read 8.3% vs SPY 13.5%, and as a satellite on the early-October combined base (101.17% control) the top-K books cost -4.4 to -10.8pp. |
| Gate-off breakout sleeves A/B (TQQQ weight 0.5 / QQQ weight 1.5 after a QQQ 50-day high) | +0.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward (also: Concentration (a few names or years carry the result)) | Holding TQQQ (A, weight 0.5) or QQQ (B, weight 1.5) for one session after a QQQ 50-day closing high inside a gate-off spell added +0.2pp (A) / +0.1pp (B) CAGR vs Oct 11 rules, but A failed fresh 2022+ (-0.2pp), the 4y win rate (32%) and January walk-forward (80.0% vs 80.2%), and B's 4y win was 50%, not above it; the signal fired on only 7 of 392 gate-off sessions, A's gain is the single May-2020 episode, and where it fires more (20-day highs) it loses -0.6pp. |
| Turn-of-month TQQQ boost (extra weight 0.25 leg, gate on, 2/2 window) | -2.60pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | An extra TQQQ leg (weight 0.25) over the last two and first two sessions of each month, while the gate was on, cost -2.6pp CAGR vs Oct 11 rules (92.2% vs 94.8%) with fresh 2022+ -9.0pp, 4y win 43% and January walk-forward never picking it (family 77.4%); the weight 0.50 and 3/3-window neighbours lost -3.5 / -3.2pp, the 1/1 window's +0.3pp had train +0.0, and 2024 alone cost -54pp spread over Sep-Nov. |
| Turn-of-month gate-off sleeve (hold TQQQ weight 0.5 over the 2/2 window, gate off) | -0.70pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Holding the TQQQ sleeve (weight 0.5) over the turn-of-month window while the gate was off cost -0.7pp CAGR vs Oct 11 rules with train -1.1pp and 4y win 41%, January walk-forward never picked it, and anchor shifts read -1.3 to -1.5pp with 4y win 17-19%; the QQQ weight 1.5 neighbour lost -0.4pp. The gate is off in high-volatility spells, and turn-of-month seasonality does not override it. |
| Options-expiration timing (TQQQ sleeve out the week after monthly opex) | -11.00pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Taking the TQQQ sleeve out for the week after monthly opex cost -11.0pp CAGR vs Oct 11 rules (83.8%) with train -13.8pp, 4y win 1% and anchors -11 to -13pp; half-out cost -5.5pp and the opex-week sign check -5.2pp, so both the 'weak post-opex week' and 'strong opex week' claims lose - being out of TQQQ on any calendar slice costs the sleeve's carry. |
| LEAPS / leveraged-ETF overlays (deep-ITM calls for margin; LEAPS leverage; 2x single-stock ETFs) | -3.70pp CAGR vs Oct 11 rules (same option-data dates, 96.0%); window 2010-02-11 to 2026-06-30; the earlier, larger starting account | Failed the walk-forward | Replacing margin with deep-ITM ~1y calls (delta 0.80) at the same exposure cost -3.7pp CAGR vs Oct 11 rules on the option-data dates (92.3% vs 96.0%, real chains at bid/ask fills; -11.1pp with every name model-priced) - the interest saved is smaller than the time value and spread paid. The leverage forms are not alpha: LEAPS at weight 1.5 gained +17.6pp on the 12 archive names, but the hindsight-free all-names family January walk-forward read 72.4% vs 79.4% with MaxDD -72.4%, and 2x single-stock ETFs read +18.1pp with 32 Robinhood breaches and MaxDD -73.8%. |
| TQQQ dip-buy sleeve while the gate is off (oversold above the 200-day average) | -1.10pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Holding the TQQQ sleeve (weight 0.5) for 3 sessions after an oversold QQQ close (two-day RSI below 10, above its 200-day average) while the gate was off cost -1.1pp CAGR vs the Oct 11 rules (93.7%), with a four-year win rate of 0%, a January walk-forward of 78.8% vs 80.2% and -1.4 to -1.6pp at the start days; it fired on only 7 of 392 gate-off sessions, and 2020 dip buys cost -17 to -80pp a year across variants (deflated Sharpe ratio at most 0.07). |
| Lag7 laggard sleeve (long the weakest Magnificent-7 names) | +3.90pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the placebo check | A margin-financed sleeve on the two weakest Magnificent-7 names (weight 0.10) added +3.9pp CAGR vs Oct 11 rules with 4y win 99%, but the leverage-matched all-7 reference gained +3.1pp and the leaders sign check +4.2pp - the laggards did not beat the wrong-name controls - the response is monotone in k (+2.0 / +3.9 / +7.7, a leverage dial), and the hindsight-free point-in-time top-7-by-cap basket read fresh 2022+ -0.5pp, failing the pre-registered bar; the gain is margin-financed beta on a hindsight-picked basket, not a laggard effect. |
| Sector dip-buy satellite after an SPY down day | -0.20pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Buying the two Select Sector SPDR funds with the best 12-month return (still above their 100-day average) for 10 sessions after an SPY fall of 1.5% or more (weight 0.15) cost -0.2pp CAGR vs the Oct 11 rules, -0.6pp in training; every neighbour lost in training, the family January walk-forward (80.8%) picked the weight-0.25 variant rather than the main one, and 2020 cost -20.8pp. |
| SPY deep-pullback satellite in an uptrend | +0.10pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) | Holding an extra SPY sleeve (weight 0.10, 10 sessions) when SPY closed at its 15-session low with the weakest one-day return of those sessions while above its 200-day average added +0.1pp CAGR vs the Oct 11 rules but read -0.6pp from 2022 onward, with a four-year win rate of 44% and a January walk-forward of 80.2%, identical to the control; neighbours flipped sign (-0.2 to +0.3pp), the gate-off parking form lost -1.1pp (it bought SPY on 2-6 March 2020), and only 78 triggers fired after 2010 (deflated Sharpe ratio at most 0.05). |
| Momentum-crash brake (Daniel-Moskowitz bear state, stock book x0.5) | -0.60pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Failed the walk-forward | Halving the stock book while SPY's 24-month return was negative and the TQQQ gate was off cost -0.6pp CAGR vs Oct 11 rules with 4y win 3% and January walk-forward picking the control 12/12; the state fired on only 33 sessions (2010 and 2020, never after 2021) and cost 2020 -27pp by cutting exposure into the leader-led rebound; the 12-month form read train -3.7pp and the bear-state-alone form -3.6pp. |
| Faster TQQQ re-entry, three rejected forms (short-window volatility; 20-day average trend; half weight) | +2.10pp CAGR vs Oct 11 rules; window 2010-02-11 to 2026-10-02; the earlier, larger starting account | Negative in the fresh era (2022 onward) (also: Failed the walk-forward) | Three faster TQQQ re-entry forms were rejected on the Oct 11 rules (a fourth form was adopted into the starting rules and is not listed): re-entering when 10-day volatility fell below 28% gained +2.1pp CAGR but -0.3pp in 2022 onward and a January walk-forward of 79.0% vs 80.2%, with its 15-day neighbour at -1.5pp; re-entering above the 20-day average lost -1.1pp, -3.2pp from 2022 onward; half weight on the first trigger gained +0.6pp but -1.0pp from 2022 onward and a January walk-forward of 78.6%. The faster volatility signal re-enters on short calm spells inside unfinished selloffs. |
| Turnover-conditioned short-term momentum tilt on the Oct 13 rules | -17.63pp CAGR vs the Oct 13 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | Tilting the rank score toward the last month's winners among the top half of names by turnover lost -17.63pp CAGR vs the Oct 13 rules (120.39% vs 138.02%), with MaxDD 11.11pp deeper, fixed-January walk-forward 103.86% vs 128.72%, fresh 2022+ -26.30pp, a 4% four-year win rate and all four anchors negative; every neighbour lost as well, and the unconditional tilt (-15.73pp) beat the conditioned one, so the turnover conditioning added nothing - the book's leaders are themselves high-turnover names (the top name sits in the high-turnover half on 74.7% of rows), so the tilt acted on the book, not the margin. |
| Late-reporter exclusion on the Oct 13 rules | +1.69pp CAGR vs the Oct 13 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the placebo check | Excluding names whose earnings release ran more than 10 calendar days past its point-in-time estimate read +1.69pp CAGR vs the Oct 13 rules (139.71% vs 138.02%), but the same flags shifted 21 and 63 sessions later - when the names were not late - read the same +1.57pp, the flagged names made up only 0.072% of the book's weight-days against a pre-registered 1% kill line, the 7-day neighbour lost -3.73pp, two of four anchors lost and the four-year median was -1.03pp: the gain is selection-path noise from a handful of forced exits, not a late-reporting effect. |
| VVIX spike / mean reversion as a TQQQ FAST re-entry trigger on the Oct 14 rules | +1.34pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Concentration (a few names or years carry the result) | Re-entering TQQQ FAST once VVIX fell to 0.80x of a spike high in the last 20 rows (the literal spike form was inert - it fires at the volatility peak, where the existing entry guard blocks every row) added 16 FAST rows in 5 of 13 gate-off spells and +1.34pp CAGR vs the Oct 14 rules (142.84% vs 141.51%), positive at all four anchors, fixed-January walk-forward 133.47% vs 132.56% and breaches unchanged, but it failed the neighbourhood item: the VVIX >= 120 neighbour lost -0.64pp, the other two neighbours gained only +0.51pp each and all three lost in 2022 onward; the gain is one episode - seven sessions in June 2010 add +17.3pp to that year, and leaving 2010 out cuts the annualised gain to +0.12pp (DSR 0.007). |
| QQQ IBS mean reversion as a TQQQ sleeve boost on the Oct 14 rules | -0.03pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the placebo check | Raising the gated TQQQ sleeve from weight 0.5 to 0.6 while QQQ's internal bar strength state is long (in below 0.10, out above 0.90; 34% of rows, about 16 entries a year) returned 141.48% vs 141.51% (-0.03pp), with train -0.45, a 4-year-window median -2.45, one of four anchors positive, a 5/10 switch test and -0.03pp at 10 bps; the placebo that boosts the opposite state gained +1.14pp and beat it, because the margin-bound account turns the boost into a swap (TQQQ gross +0.048x, stock book -0.052x on boosted rows) and the deltas are path noise; stressed-margin forced liquidations rose from 433 to 726, and the literal QQQ 0.10 satellite lost -0.09pp. |
| Margin backlog: completing margin-clipped buys on later sessions, on the Oct 14 rules | +0.29pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the placebo check | Completing stock buys that initial margin had clipped, on later sessions whenever headroom above a 5% equity reserve appeared, returned 141.79% vs 141.51% (+0.29pp), but it bought on only 78 sessions in 16.6 years and filled 1.6% of the clipped notional (margin binds on 66% of trade rows, with a mean buy fraction of 0.23 when it does); a plain vol-target raise to the same average stock exposure earned +0.56pp and beat it (-0.28pp, 1 of 10 held-out halves), the switch test was 7/10 and the 4-year-window median fell 0.65pp; with no reserve the gain rose to +2.19pp, but its own matched vol-target raise earned +4.33pp. |
| Pre-release admission: the admission rule's rank gate relaxed inside the candidate's own release window, on the Oct 14 rules | -4.67pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the placebo check (also: Failed the walk-forward) | Letting the leader-admission rule take a bigger name from further down the ranking than the live rule allows, only while that name sat inside the pre-release boost's own window before its estimated earnings release, added 49 sleeve admissions on 17 dates and returned 136.83% vs 141.51% (-4.67pp), with train -5.05, 2022 onward -3.66, 11% of 4-year windows won, fixed-January walk-forward 127.00% vs 132.56%, a 1/10 switch test and all four anchors negative; the same rule on windows shifted a month earlier gained +0.93pp and beat it, the admitted names trailed the names they replaced by 2.0% over 20 sessions, and the same wider gate with no release condition lost 10.34pp. |
| Admission confirmation: two-consecutive-row confirmation of admissions on the Oct 14 rules | -1.56pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the placebo check | Requiring the leader-emergence admission condition to hold on two consecutive decision rows before an admission fires returned 140.11% vs the pinned Oct 14 rules' 141.67% (-1.56pp CAGR; fixed-Jan walk-forward 132.47% vs 132.69%): the confirmation arm's equity path is bit-identical to the placebo that delays every admission by one row unconditionally, so the confirmation effect is exactly zero and the net is the delay cost alone - 0 of the control's 82 admissions fired on time, 70 fired one row late and 12 were lost, the loss concentrates in 2020 (-48.35pp) and 2024 (-23.16pp), the walk-forward family picks the control in all 12 years, and the deflated Sharpe ratio of the delta is 0.000; none of the control's admissions had the condition hold on the preceding row, so on this crossing signal confirmation is structurally a delay, and a five-session-old condition lost -9.51pp. |
| Daily refresh of each stock sleeve's volatility scaling factor between its own rebalances, on the Oct 14 rules | +1.54pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Concentration (a few names or years carry the result) (also: Negative in the fresh era (2022 onward)) | Recomputing each stock sleeve's volatility scaling factor every session between its own rebalances, and resizing the sleeve whenever the fresh factor moved 20% or more from the one it carried, added +1.54pp CAGR vs the Oct 14 rules (143.05% vs 141.51%) and was positive at all four anchors, but the gain is 2020 alone (raises as volatility fell after the crash; with 2020 left out the annualised delta is -0.10pp), fresh 2022+ fell 3.68pp, the four-year-window median fell 2.17pp, the switch test was 8/10 and 7/10, and a plain raise of the volatility target to the same average stock exposure earned +1.64pp; the cut-only form lost -1.71pp, worse than a uniform cut to the same exposure (-0.39pp), because realized volatility peaks after the drop and the 2021 cuts came two sessions after the trough; a copy reading 20-session-old volatility earned +2.69pp, and headline-model breaches rose from 2 to 5 (DSR 0.003). |
| Announcement-return momentum gate: eligibility from a stock's last four earnings-release reactions, on the Oct 14 rules | -27.24pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | Making a stock ineligible, and selling it if held, when the sum of its last four earnings-release reactions against SPY (the two-day move around each results filing) was zero or negative returned 114.26% vs the Oct 14 rules' 141.51% (-27.24pp CAGR): the rule flagged 14% of the book's holdings and its largest holding on 14.5% of days, and removing those leaders lost in training (-23.58pp), in fresh 2022+ (-40.99pp), in all 151 four-year windows, at all four schedule anchors and in every held-out half of the switch test (0/10); the fixed-January walk-forward fell to 101.76% from 132.56%, every neighbour lost (two or eight releases, thresholds of -5% and +5%), the entry side alone lost -28.90pp and the exit side alone -17.02pp, and a placebo summing days 10-12 after each release lost more (-45.27pp) while flagging 2.4 times as many holdings, so the release reactions do not pick worse leaders than an arbitrary three-day sum; four-year windows with a margin breach fell from 47 to 0 at a deeper drawdown (-62.91% vs -54.39%), and the deflated Sharpe ratio is 0.000. |
| Drift-band re-weighting of each stock sleeve back to the market-cap weighting between its own rebalances, on the Oct 14 rules | +1.72pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Concentration (a few names or years carry the result) (also: Failed the walk-forward) | Re-weighting a stock sleeve back to the market-cap weighting at its current size whenever its drifted weights sat 0.10 or more away from that target (156 times in 16.6 years, with the same picks and the same exposure) added +1.72pp CAGR vs the Oct 14 rules (143.23% vs 141.51%), was positive at all four schedule anchors and in fresh 2022+ (+3.18pp), and beat a copy re-weighting on the same days to 20-session-old weights (+0.93pp) and the same number of re-weights on a fixed calendar (-0.09pp); but the four-year-window median fell 1.47pp, the family January walk-forward ended 0.04pp under the control (132.52% vs 132.56%), the switch test was 9/10 without the embargo and 6/10 with it, and the gain sits in 2021 and 2026 (without both years the annualised delta is -0.49pp); margin breaches were unchanged and the deflated Sharpe ratio is 0.013. |
| Post-release tight rank cut: selling a held stock that falls out of the top seven in the five sessions after its own earnings release, on the Oct 14 rules | +8.73pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the walk-forward | Selling a held stock at the next open when it ranked worse than 7th among the eligible names in any of the five sessions after its own earnings-release filing, without waiting for a clearly better replacement, added +8.73pp CAGR vs the Oct 14 rules (150.23% vs 141.51%) with a fixed-January walk-forward of 137.73% vs 132.56%; it was positive in training (+11.50pp), at all four schedule anchors (+4.77 to +8.71pp) and with 2020 or 2024 left out, and it beat the same rule on non-announcement days (-9.62pp) and a plain leverage increase to the same average exposure (+0.85pp); but fresh 2022+ was flat (+0.01pp), the response peaked at the 7th-rank cut (5th +3.70pp, 10th +1.57pp), and the family January walk-forward, which re-picks the window and the rank cut each year, chose the 5th-rank cut in three early years and ended 0.82pp under the control (131.74%); the drawdown was not reduced (-54.73% vs -54.39%), margin breaches were unchanged and the deflated Sharpe ratio is 0.014. |
| Revenue-acceleration rotation margin: a wider replacement gap for held stocks whose revenue growth is accelerating most and a narrower one for the weakest, on the Oct 14 rules | -5.77pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | Scaling the score gap a held stock that has slipped in rank must face before it is swapped out by its revenue-growth acceleration among the eligible names (one cross-sectional standard deviation wider for the top third, one narrower for the bottom third; 69 exits differ in 16.6 years) cost -5.77pp CAGR vs the Oct 14 rules (135.74% vs 141.51%); it lost in training (-6.84pp), in fresh 2022+ (-2.37pp), at all four schedule anchors (-2.07 to -5.78pp), at half and one-and-a-half times the step (-0.55 and -4.90pp) and with halves instead of thirds (-9.36pp), and the same rule on acceleration four quarters stale lost about as much (-6.05pp); its own fixed-January walk-forward was 134.46% vs 132.56% only because the loss sat in 2011 (an early sale of the book's largest holding, which then rose) and 2026 (a kept leader that then fell), while the family January walk-forward ended at 132.31%, the switch test was 5/10 both ways and the deflated Sharpe ratio is 0.000; the drawdown and margin breaches were unchanged. |
| Revenue-acceleration-confirmed drift band: re-weighting a stock sleeve back to the market-cap weighting between its own rebalances only toward a stock whose revenue growth is accelerating most, on the Oct 14 rules | +3.24pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-02; the current starting account | Concentration (a few names or years carry the result) | Re-weighting a stock sleeve back to the market-cap weighting at its current size between its own rebalances, only when one of its stocks sat 0.10 or more below its target weight while that stock's revenue-growth acceleration was in the top third of the eligible names (91 times in 16.6 years, at the same exposure), added +3.24pp CAGR vs the Oct 14 rules (144.75% vs 141.51%) with a fixed-January walk-forward of 135.08% vs 132.56%; it was positive in training (+1.56pp), in fresh 2022+ (+6.92pp), at all four schedule anchors (+1.66 to +2.74pp), at bands of 0.05 and 0.20 and with halves instead of thirds (+2.11 to +2.63pp), the family January walk-forward ended at 132.99%, and it beat the same rule on acceleration four quarters stale (+2.20pp), on a random third of the names (+2.07pp), the same number of re-weights on a fixed calendar (+3.39pp) and the up-drift rule without the acceleration condition (+1.59pp); but three months carried 90.5% of its active return and 70% fell in the first 63 sessions of a calendar year, so the switch test with the embargo that removes those sessions was 7/10 (10/10 without it), and in training the condition did no better than the stale or random ones (-0.19 and -0.26pp); the drawdown and margin breaches were unchanged and the deflated Sharpe ratio is 0.174. |
| TQQQ trend check on top of the volatility gate: the TQQQ sleeve held at zero while QQQ closes more than 2% below its 50-day average, back on above it, on the Oct 14 rules (a drawdown guard) | -13.45pp CAGR vs the Oct 14 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the walk-forward | Holding the TQQQ sleeve at zero while QQQ closed more than 2% below its 50-day average (back on above it), on top of the existing volatility gate, was judged as a drawdown guard and failed: maximum drawdown moved only from -54.39% to -53.62% on the research model and from -57.62% to -57.66% on the headline model (the bar asked for 2.0pp shallower on both), because the worst drawdown was a single-stock fall in early 2021 that the trend check whipsawed through (it avoided 15% of that episode's TQQQ loss); it cost -13.45pp CAGR vs the Oct 14 rules (128.06% vs 141.51%) and the fixed-January walk-forward fell to 120.66% vs 132.56%. It did cut the 2025 drawdown from -53.20% to -39.31%, removed every margin breach, was shallower at all four schedule anchors and beat a 20-session-late placebo, but a plain smaller TQQQ sleeve with the same average TQQQ exposure did better on the headline model (-54.62%) and on the walk-forward (132.67%). |
| Single-company cap without cutting exposure: the largest company's value across the stock sleeves capped at one times equity on every trade, the excess moved to the same sleeves' other picks by market cap, on the Oct 14 rules (a drawdown guard) | -30.79pp CAGR vs the Oct 14 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the walk-forward (also: Failed the placebo check) | Capping the largest company's value, summed over the stock sleeves, at one times equity on every rebalance, entry and resize (it bound on 80% of them), with the excess moved to the same sleeves' other picks by market cap, was judged as a drawdown guard and failed: the largest company's peak fell from 1.94 to 1.19 times equity, but maximum drawdown moved only from -54.39% to -53.73% on the research model and from -57.62% to -57.98% on the headline model (the bar asked for 2.0pp shallower on both), because the second names fell with the leader (the 2025, 2020 and 2018 drawdowns got deeper while 2021 and 2026 got shallower) and margin limits blocked part of the re-weighting's purchases; it cost -30.79pp CAGR vs the Oct 14 rules (110.71% vs 141.51%) and the fixed-January walk-forward fell to 101.85% vs 132.56%. It cut forced liquidations under the stress preset from 433 to 110 and was shallower at all four schedule anchors, but a plain lower volatility target with the same average stock exposure did better on both drawdowns (-53.41% and -55.48%) and on the walk-forward (126.31%), and the same cap applied on randomly shifted days was shallower than the rule itself. |
| Low-correlation co-leader: a quarter of every stock sleeve dominated by the largest company held in the best-ranked eligible stock whose recent daily returns are nearly uncorrelated with it, on the Oct 14 rules (a drawdown guard) | -35.00pp CAGR vs the Oct 14 rules; window 2010-02-11 to 2026-10-02; the current starting account | Failed the walk-forward (also: Negative in the fresh era (2022 onward)) | Holding a quarter of every stock sleeve that the book's largest company dominates in the best-ranked eligible stock whose daily returns over the prior six months had a correlation of 0.30 or less with that company (swapped at equal value, so the exposure stayed the same) was judged as a drawdown guard and failed: maximum drawdown deepened from -54.39% to -55.11% on the research model and from -57.62% to -59.04% on the headline model (the bar asked for 2.0pp shallower on both). The low correlation held in calm markets but not in the sell-offs: the co-leader helped when the leader fell alone (the early-2021 and 2025-26 drawdowns got shallower), but in 2020 it was a stock that fell 61% while the leader fell 24%, and the 2025, 2020, 2011 and 2018 drawdowns got deeper. It cost -35.00pp CAGR vs the Oct 14 rules (106.50% vs 141.51%), -45.08pp in fresh 2022 onward, and the fixed-January walk-forward fell to 96.80% vs 132.56%. It cut forced liquidations under the stress preset from 433 to 135 and was shallower than the same rule with a random, a most-correlated or a market-cap-chosen second stock, but a plain lower volatility target with the same average stock exposure did better on both drawdowns (-53.35% and -56.15%) and on the walk-forward (130.28%). |
| Low implied correlation as a TQQQ warning: the TQQQ sleeve halved while Cboe's three-month implied correlation index sits in the bottom fifth of its prior two years, on the Oct 14 rules with the smaller leveraged-ETF sleeve (a drawdown guard) | +1.34pp CAGR vs the Oct 14 rules with the smaller leveraged-ETF sleeve; window 2010-02-11 to 2026-10-02; the current starting account | Failed the placebo check | Halving the TQQQ sleeve's target, while the volatility gate had it on, whenever Cboe's three-month implied correlation index closed at or below the 20th percentile of its prior two years (38% of the days the sleeve was on; combined with the existing TQQQ cut by the smaller multiplier, not the product) was judged as a drawdown guard and failed the placebo check. Maximum drawdown improved from -54.39% to -51.84% on the research model and from -57.62% to -55.20% on the headline model, it added +1.34pp CAGR vs the Oct 14 rules with the smaller leveraged-ETF sleeve (150.15% vs 148.82%), the fixed-January walk-forward rose to 143.86% vs 141.23%, the drawdown was shallower at all four schedule anchors, and it beat a plain smaller TQQQ sleeve with the same average TQQQ exposure (-54.70% and -55.61%, walk-forward 140.53%). But the same cut on the same pattern of days moved six months later was shallower still on both models (-50.64% and -54.65%), so the drawdown gain came from a cut clustered in time, not from what low correlation predicts. The state was often on at the peaks but switched off inside most drawdowns and missed the 2025-26 one, and in early 2021 the stock book used the freed margin and lost more in its largest stock. Margin breaches fell from 1 / 2 / 47 to 0 / 1 / 0 (research path / headline path / four-year windows) and forced liquidations under the stress preset from 428 to 405. |
| Top-name break re-weight: capping the book's largest company at half of each stock sleeve while it trades 10% or more below its 63-session high, the rest moved to the sleeve's other picks, on the Oct 14-22 rules (a drawdown guard) | -21.97pp CAGR vs the Oct 14-22 rules; window 2010-02-11 to 2026-10-14; the current starting account | Failed the walk-forward (also: Failed the placebo check) | Capping the book's largest company at half of each stock sleeve's weight whenever it closed 10% or more below its 63-session high (19.7% of days), with the excess moved to the sleeve's other picks at the same total exposure, was judged as a drawdown guard and made drawdowns deeper, not shallower: maximum drawdown -59.96% vs -57.92% on the research model and -60.02% vs -58.89% on the headline model, CAGR 139.51% vs 161.48%, fixed-January walk-forward 129.15% vs 157.38%, and it lost in every one of the 151 rolling four-year windows. |
| Pre-release boost held one more session (starting one session earlier): the boost and the TQQQ cut kept on through the session after the top holding's estimated earnings release, on the Oct 15 rules | +1.25pp CAGR vs the Oct 15 rules; window 2010-02-11 to 2026-10-14; the current starting account | Mechanism overlap | Holding the pre-release boost (and the smaller leveraged-ETF sleeve) one session longer, through the session after the top holding's estimated earnings release (lo 0 instead of 1), was given a full pre-registered validation on the Oct 15 rules control and rejected on its neighbours: it added +1.25pp CAGR (162.41% vs 161.16%) and +1.31pp to the fixed-January walk-forward and passed the four-year windows, anchors, switch, leave-out, cost and placebo items, but a smaller boost (x1.20) at the same setting lost 0.29pp while a larger one (x1.30) gained 2.40pp, and one extra boosted session placed away from the release earned most of the gain (+0.93pp), so the gain is extra exposure on boosted days, not timing around the release. Drawdowns and margin breaches were unchanged. |
| Leader-dip boost, only in an uptrend: the stock book raised by a quarter while the largest holding trades 5.6% or more below its 52-week high and the Nasdaq-100 is above its 200-day average, on the Oct 14-22 rules | +4.35pp CAGR vs the Oct 15 rules; window 2010-02-11 to 2026-10-14; the current starting account | Negative in the fresh era (2022 onward) (also: Failed the exposure-matched check) | A retry of the leader-dip boost that only bought the dip while the Nasdaq-100 was above its 200-day average, a filter chosen on 2010-2021 to skip crash onsets. It added 4.4pp of CAGR (165.5% vs 161.2%) and 4.3pp to the fixed-January walk-forward, but from 2022 on it trailed the strategy without it (-2.0pp) and a plain boost with the same average exposure (-5.7pp): the filter removed exactly the dips that paid after 2021. It did not make the worst drawdown any shallower, because that drawdown (early 2021) was a single-stock break inside a bull market. It failed the pre-registered out-of-sample and exposure-matched checks. |
| Leader-dip boost: the stock book raised by a quarter while the largest holding trades 5.6% or more below its 52-week high, on the Oct 14-22 rules | +10.64pp CAGR vs the Oct 15 rules; window 2010-02-11 to 2026-10-14; the current starting account | Failed the exposure-matched check | Raising the stock book by a quarter whenever its largest holding had pulled back 5.6% or more from its 52-week high (37% of days; the margin limit let only about a quarter of the boost through) added 10.6pp of CAGR (171.8% vs 161.2%) and 11.9pp to the fixed-January walk-forward, held out of sample from 2022 (+19.9pp), and beat ten time-shifted copies of itself. But against a plain boost with the same average exposure it made drawdowns slightly deeper (-60.0% vs -59.8% on the research model, -60.6% vs -60.0% on the headline model) and added margin breaches in the early 2021 and 2026 sell-offs, because buying the leader's dip also buys the start of real crashes. It failed the pre-registered exposure-matched check. |
| Dip-step: two rungs of stock-book dose on the lead's 252-session high, at the live dip's average gross | -0.64pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward (also: Deepened the worst drawdown) | Replacing the single 5.6% dip boost with two rungs (weights 1 and 2 at 5.62% and 11.24% below the lead's 252-session high), scaled so the average stock gross matched Oct 22 rules, cut the fixed-January walk-forward from 171.61% to 170.96% and deepened the worst drawdown from -55.28% to -57.32%. Headline breaches rose from 17 to 19. Turnover fell slightly. The deep rung was real (17% of rows, 64% of the boost) and was not killed at step 0. |
| Daily appreciation resize: re-size held stocks after account gains when margin headroom permits, on the Oct 14-22 rules | +1.47pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the exposure-matched check (also: Failed the walk-forward) | Re-sizing the held stock book after account gains added 1.47pp to the fixed-January walk-forward, short of the 2pp bar, and increased stress breaches from 311 to 475. It finished 0.07pp behind a constant exposure increase with the same average stock gross, so the gain came from extra exposure rather than the appreciation timing. Re-sizing every day was worse than the control by 2.73pp and produced 779 stress breaches. |
| Ten interactions among the dip boost, pre-release boost, post-release cut and leveraged-ETF routing, on the Oct 14-22 rules | +1.46pp fixed-January walk-forward vs Oct 22 rules; window 2010-02-11 to 2026-10-16; the current starting account | Failed the walk-forward | None of ten pre-stated interactions among the current stock and leveraged-ETF rules passed the card. The best arm turned the dip boost off while the post-release cut was active: it added 1.46pp to the fixed-January walk-forward with unchanged worst drawdown, headline breaches and stress, but missed the 2pp bar. The only arm with a materially shallower drawdown gave up 10.81pp of walk-forward. |
| Dip-state leveraged-ETF routing: raise the stock book and cut the leveraged Nasdaq sleeve while the lead stock is below its high, on the Oct 14-22 rules | +11.71pp fixed-January walk-forward vs the Oct 14-22 rules (Oct 14 rules); window 2010-02-11 to 2026-10-16; the current starting account | Deepened the worst drawdown | Routing exposure from the leveraged Nasdaq sleeve into the stock book while the lead stock was below its high added 11.71pp to the fixed-January walk-forward and beat a constant exposure increase by 6.65pp. It nevertheless deepened the worst drawdown by 4.32pp on the research model and 3.43pp on the headline model, raised headline breaches from 6 to 16, and raised stress breaches from 364 to 566, so it failed the risk conditions. |
Stopped by a diagnostic (14)
| Candidate | Headline delta | Reason | Deciding facts |
|---|---|---|---|
| Directionless stand-down for the TQQQ sleeve: the sleeve to cash while QQQ's efficiency ratio over 63 or 126 sessions is below a threshold, on the Oct 14-22 rules (a drawdown guard) | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Stopped at the pre-registered selection step, with no test cells run. Standing the TQQQ sleeve down whenever QQQ's daily efficiency ratio (net move divided by path length) over 63 or 126 sessions fell below 0.10, 0.15 or 0.20 did cut the sleeve's losses on flat versions of 2010-2021 (by 1.6 to 8.5 points a year), but at every one of the six settings it cost 13.9 to 26.1 points of 2010-2021 CAGR against a 2-point limit. Daily QQQ moves are noisy even in a strong bull market (the ratio's median was 0.11 to 0.13), so the rule kept the sleeve out of the market for 50% to 93% of the trend. |
| Short program (short the weakest names) | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | The pre-registered forward-return check (834 twenty-session windows) found the short candidates lag the eligible-pool average but still beat SPY by +2.3pp/yr and +2.6pp/yr under the earlier and the current market-cap weighting, with break-even borrow at -2.3% against the required +5% assumption and a 49% hit rate, so the check failed, no short book was ever simulated, and the decision on 2026-10-13 was 'long only for now'. |
| Short book on Oct 11 rules - engine feasibility | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | The pre-registered feasibility check failed before any run: the step2 engine cannot express short stock positions (selection takes rank-top only and every weighting scheme returns non-negative weights), its margin model prices a short on signed (netted) market value, and it carries no borrow rate, fee, recall or dividend-paid model. |
| Winter-rotation sleeve (GLD/TLT/BIL momentum on idle cash) | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | The sleeve was inert, not negative (measured delta +0.00pp CAGR and +0.00pp walk-forward OOS): mean idle cash is 0.12% of equity (median 0%), so there was no idle capital to rotate and the premise was refuted by measurement before the construct could bind. |
| Put/call fear extreme as a TQQQ FAST re-entry trigger on the Oct 13 rules | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at mechanics, with no strategy run and no trials: a 10-day mean equity put/call ratio above 0.80 falls on 38 gate-off rows in four fear spells (2011-08, 2018-12, 2020-03, 2022-11/12), but every trigger row sits at 0.846-1.000x its spell's running volatility peak, so adding it to the existing re-entry trigger adds exactly one FAST row in 16 years (2022-12-02); only the hold form acts - 38 FAST rows at the volatility peak, before the low - which is the form already rejected for the insider-spike trigger. |
| Coppock-curve regime for the TQQQ sleeve, on the Oct 14 rules | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at the mechanics check, with no strategy run: the monthly SPY Coppock curve (a 10-month weighted average of the 14- and 11-month rates of change) turned up from below zero only in 2009-05, 2016-05 and 2023-03 (in 2020 it never went below zero), so letting TQQQ through the volatility gate for 12 months after each signal changes 8 sessions in 16 years, all in the May 2010 flash crash at its volatility peak; the gate is already on in 98.6% of those sessions, which makes the QQQ version the already-rejected post-bear re-risk satellite. |
| Overnight equity-premium park for the leveraged-ETF sleeve's idle dollars on the Oct 14 rules | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at mechanics, with no strategy run and no trials: holding QQQ worth half of equity only overnight (bought at the close, sold at the next open) while the leveraged-ETF sleeve's volatility gate is off would have needed a positive close-to-open return net of 10 bps a night on the 2010-2021 gate-off nights, but on those 184 nights (6.1% of sessions) QQQ's overnight return was -0.033% gross and -0.133% net, while its intraday leg was +0.089%; over all nights the overnight premium was real (+0.056%, t 4.0) but sat in the calm regime and was about half the round-trip cost. |
| Book-level industry weight cap on new fills on the Oct 14 rules | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at mechanics, with no trials: a cap skipping any new stock whose industry already held 70% of the stock book would bind on 94.1% of sessions, and on 99.3% of those one name alone held 70% (the largest name's median share of the stock book is 99.4%), so the sleeves' overlap is single-name concentration rather than industry concentration; the 2021 margin breach and the five worst drawdown troughs each sat 98.9-100% in one name, which a cap acting only on new fills cannot reduce. |
| Stale-leader replacement (the exit mirror of leader admission) on the Oct 14 rules | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at mechanics, with no trials: swapping a sleeve's largest name for a top-2 leader at least half its size once the largest name had decayed to rank 9 or worse fires on only 6 dates in 16.6 years (20 sleeve swaps), and its trigger covers 0.50% of the stock book's weight-days, because the rank buffer, the daily check and the leader admission already clear decayed dominant names; on the six swaps the incoming name trailed the outgoing one by 15.5% over the next 20 sessions. |
| Earnings-record-conditioned hold of the pre-release boost through the top holding's release session, on the Oct 14 rules | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at mechanics, with no trials: keeping the pre-release boost on through the largest holding's estimated earnings-release session only when that stock's last four release reactions against SPY summed to more than zero would almost never have differed from holding through every time, because the record was positive on 91.5% of the 47 sessions where the rule could act (the pre-registered limit was 90%), so it could have skipped only 4 of them, and the holding rose over the release session on all 4 (+2.25% on average, about the same as the other 43); the unconditional hold was tested separately. |
| Release-keyed re-weight: re-weighting a stock sleeve back to the market-cap weighting on the session after a holding's own earnings release, on the Oct 14 rules | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at mechanics, with no trials: re-weighting a stock sleeve back to the market-cap weighting on the session after one of its stocks' own earnings-release filing, when that stock's target weight sat 0.05 or more above its drifted weight, would have fired only 22 times in 16.6 years (0.15% of sleeve-days, none after 2019), under the pre-registered minimum of 40 re-weights and 1% of sleeve-days. |
| Leader change inside a sleeve: re-weighting a stock sleeve back to the market-cap weighting when a different holding becomes its largest company, on the Oct 14 rules | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at mechanics, with no trials: re-weighting a stock sleeve back to the market-cap weighting when its largest company by market cap was not its largest holding and led it by 5% or more would have fired in only 23 episodes in 16.6 years (11 of them in 2021), under the pre-registered minimum of 30. |
| Margin-cushion re-target: every sleeve and the leveraged-ETF sleeve resized back to their own targets when the account's equity falls to 1.25x its maintenance requirement, on the Oct 14 rules with the smaller leveraged-ETF sleeve (a drawdown guard) | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Dropped at mechanics, with no trials: the guard was to resize every stock sleeve and the leveraged-ETF sleeve back to their own targets at the next open (same picks and weights) whenever the account closed with equity at or below 1.25 times its maintenance requirement, with a five-session cooldown, and the pre-registered check needed it to fire at least 5 times. On the research margin model it fired on only 3 closes in 16.6 years (March 2021, March 2025 and March 2026), each with 16-28% of that drawdown still ahead; each time it trimmed the drift above target (stock exposure down 0.11-0.17x equity, the leveraged-ETF sleeve back to 0.50x, no buys), but the 2020, 2011 and 2018 drawdowns never got that thin. On the headline model, whose requirement adds a concentration stress term, the same threshold is routine (7.4% of closes, 70 firings), so it becomes a frequent re-target that often buys rather than a crash guard. |
| Industry run-up crash guard: the stock book scaled to three quarters while the dominant holding's industry has doubled against the market over two years with at least two bubble warning signs, on the Oct 14-22 rules (a drawdown guard) | No delta (stopped by a diagnostic before a backtest) | Stopped by a diagnostic | Scaling the stock book to three quarters while the largest holding's industry had more than doubled against the market over two years and showed at least two of four bubble warning signs (accelerating gains, rising volatility, share issuance, many young firms) was stopped at its pre-registered mechanics check: the signal was on at the peak of only two of the six largest drawdowns (early 2021 in carmakers, late 2024) and silent in the other four (2011, the 2020 crash, both 2026 drawdowns), so it could not be a general drawdown guard. No trials were charged. |