Validation record
Research tool · not investment advice. Full disclaimer →
New here? How to read
Every number on this page names its source — start with “How to read”.
The falsification record — published, not curated.
Descriptive record only — past data and published method.
Strategy verdicts
26 strategiesSource: the research record · per run.
| Strategy | Category | Cost basis | Net return/vol | CAGR | Max drawdown | Annual cost | Capacity | Verdict | Book |
|---|---|---|---|---|---|---|---|---|---|
| RISKADJ | public-factor | flat-0.3%/turnover | 1.29 | 35.4% | -41.9% | — | — | survives | — |
| RISKADJ monthly flat-cost (the headline) | cost-tested | flat-0.3%/turnover | 1.29 | 35.6% | -41.9% | 3.8% | ₹30 cr | tested | — |
| MOM12 | public-factor | flat-0.3%/turnover | 1.20 | 37.6% | -49.6% | — | — | survives | — |
| LOWVOL_MOM | public-factor | flat-0.3%/turnover | 1.12 | 18.7% | -33.9% | — | — | survives | — |
| HI52 | public-factor | flat-0.3%/turnover | 1.09 | 24.2% | -38.6% | — | — | survives | — |
| QUAL_MOM | public-factor | flat-0.3%/turnover | 1.04 | 23.5% | -29.0% | — | — | survives | — |
| RESID_MOM | public-factor | flat-0.3%/turnover | 1.02 | 30.2% | -50.7% | — | — | survives | — |
| MOM6 | public-factor | flat-0.3%/turnover | 1.01 | 30.0% | -51.3% | — | — | survives | — |
| VAL_MOM | public-factor | flat-0.3%/turnover | 0.91 | 24.4% | -63.9% | — | — | did-not-survive | — |
| Nifty500 buy & hold (quarterly marks) benchmark | baseline | realistic (tier spread + 0.5xATR slip) | 0.89 | 15.3% | -29.2% | 0.0% | Not modelled | benchmark | — |
| DEFENSIVE(lowvol+qual) | public-factor | flat-0.3%/turnover | 0.87 | 12.0% | -25.5% | — | — | survives | — |
| QMV(qual+mom+val) | public-factor | flat-0.3%/turnover | 0.86 | 20.4% | -51.1% | — | — | did-not-survive | — |
| LOWVOL | public-factor | flat-0.3%/turnover | 0.84 | 9.6% | -26.5% | — | — | did-not-survive | — |
| LOWVOL_MOM quarterly largecap realistic+band | cost-tested | realistic (tier spread + 0.5xATR slip) | 0.79 | 13.3% | -25.4% | 8.3% | ₹190 cr | tested | — |
| SECTOR_ROT_V17 | sector-rotation | etf 0.15%/side | 0.79 | 14.7% | -39.2% | — | — | champion-candidate | Nifty Financial Services 30% Nifty Auto 29% Nifty PSU Bank 17% Nifty Healthcare 13% Nifty Metal 11% RESIDUAL SLEEVE: CASH (below 200DMA) as of 2026-04-01 |
| LOWVOL quarterly largecap realistic+band | cost-tested | realistic (tier spread + 0.5xATR slip) | 0.78 | 9.6% | -23.1% | 5.4% | ₹168 cr | tested | — |
| QUALITY | public-factor | flat-0.3%/turnover | 0.76 | 14.4% | -43.6% | — | — | did-not-survive | — |
| LOWBETA(BAB) | public-factor | flat-0.3%/turnover | 0.70 | 10.2% | -27.7% | — | — | did-not-survive | — |
| EARN_YIELD | public-factor | flat-0.3%/turnover | 0.70 | 17.5% | -71.4% | — | — | did-not-survive | — |
| SECTOR_ROT_V8 | sector-rotation | etf 0.15%/side | 0.70 | 10.8% | -36.2% | — | — | frozen-champion | — |
| BOOK_YIELD | public-factor | flat-0.3%/turnover | 0.61 | 16.2% | -82.3% | — | — | did-not-survive | — |
| RISKADJ quarterly largecap realistic | cost-tested | realistic (tier spread + 0.5xATR slip) | 0.51 | 10.2% | -43.3% | 15.1% | ₹97 cr | tested | — |
| C-BLEND 50/50 monthly (2026-07-03 champion) | cost-tested | participation (Almgren k=0.6) @ Rs50cr | 0.17 | — | — | — | ₹38 cr | tested | — |
| LOWVOL_MOM monthly realistic + band | cost-tested | realistic (tier spread + 0.5xATR slip) | 0.15 | 1.0% | -50.3% | 19.8% | ₹65 cr | tested | — |
| PEAD book (real BSE dates) | event-study | tiered + ATR event costs | 0.10 | — | — | — | — | did-not-survive | — |
| RISKADJ monthly realistic | cost-tested | realistic (tier spread + 0.5xATR slip) | 0.09 | -1.4% | -68.9% | 36.0% | ₹30 cr | tested | — |
Benchmark: Nifty 500 buy-and-hold at realistic cost, net return/vol 0.89. Colour compares like with like: only realistic-cost rows are judged against it; flat-cost rows are illustrative and stay uncoloured. On realistic costs, no strategy in this record beats buying and holding the index.
The index baseline has no modelled capacity breakpoint. Its source stores infinity as a sentinel because this study does not calculate a stock-position participation threshold for the index. This does not imply unlimited real-world capacity.
Source: strategy_runs · 27 runs recorded.
Realistic figures are net of a per-name cost — the tier spread plus half an average true range of slippage — walk-forward from 2012 to 2026, with no look-ahead. The holdings are drawn from the live Nifty 500 constituents: no delisted tickers, no funds or ETFs.
The record
what survived, and what did notSource: the strategy register · walk-forward.
Every strategy we have built is listed here with the number that decides it — net return/vol, on the cost basis each row states — against the Nifty-500 buy-and-hold benchmark measured at realistic cost. Most rows were tested on a flat cost model, which flatters them; on realistic costs none has beaten that benchmark. We publish them anyway — a research record that only contains winners is not a record.
Return/vol is the annualised mean return divided by its volatility — no risk-free rate is subtracted, so it is not the textbook ratio of that name. Higher is steadier per unit of return; the benchmark's is 0.89. A flat-cost row and a realistic-cost row are not comparable numbers, which is why the table only colours like against like.
Recorded dead
approaches we killed, with their numbersSource: Patearn strategy ledger · in its own words.
| Failure model | Recorded result | Why it blocks |
|---|---|---|
| ACCEL / PULLBACK / DELIV_MOM (standalone) | return/vol 0.42-0.85, MaxDD −44%…−70% | [Scope blocked: standalone short-thrust chasing / dip-buying / delivery% monthly ranking. Reopened by: multi-condition conjunctions or regime-conditional entry.] Short-thrust chasing / dip-buying / delivery% added no standalone edge. |
| BOOK_YIELD (deep value / B-P) | return/vol 0.61-0.63 · alpha −1.8%…−2.2% (NEGATIVE) · beta 1.54-1.56 · MaxDD −82% · fails BOTH halves | [Scope blocked: standalone long-ranking on B/P (top-25 monthly, 2012-26 vs Nifty 500). Reopened by: multi-factor context/veto or quality-guarded value with β≤1.1, lower rebalance frequency, or non-standalone ranker beating recorded net return/vol.] Negative alpha + −82% drawdown + high beta = a value-trap engine. Never a production long-ranker. The β≈1.54 + MaxDD≈82% alone stop us. |
| C-BLEND 50/50 as a FUNDABLE book (2026-07-05c) | Flat-cost return/vol 1.32 (recorded champion) → participation-cost NET 0.52 @Rs25cr · 0.17 @Rs50cr · −0.30 @Rs100cr; beats the index at NO AUM; H2 (honest window) 0.70 @Rs50cr < 0.89; ann cost 22%→86% | [Scope blocked: monthly rebalanced mid-cap tilted C-BLEND book at scale (AUM ≥₹25cr) under Almgren participation cost. Reopened by: quarterly/lower frequency rebalance, large-cap only (e.g. LOWVOL_MOM 1.02 @Rs50cr), or low-impact execution.] The 1.32 was flat-cost only. Monthly rebalance × mid-cap tilt (median capacity ~Rs38cr) makes Almgren participation impact fatal; the RISKADJ core is worse. C-BLEND stays a descriptive/paper overlay (fence holds), never a fundable book. Only participation-fundable corner = quarterly large-cap LOWVOL_MOM (1.02 @Rs50cr, ~Rs100cr ceiling). |
| Concall credibility as a factor | Spearman ≈0; HIGH−LOW excess −10% @12m (inverse, survivorship) | [Scope blocked: linear return ranker on raw concall credibility score (12m horizon). Reopened by: red-flag governance veto, interaction terms with capital allocation, or descriptive/veto classification.] FALSIFIED as a factor → descriptive/veto only. |
| Council-56 breakout swing books S1/S2/S4/S5/S6B, tiers v1-v2 (2026-09-28; settled 2026-09-30; corrected share-action tape 2026-10-01) | Causal read-only replay 2004-04-17 to 2026-09-25 on Codex's vendored runner (next-open fills, gap-aware observed-low stops, a flat 0.55% fee each side at a ₹10 lakh start, not AUM-scaled, 25 slots x 4%, 252-session exits, uniform circuit stress) with primary-source merger/delisting settlements, NSE dividends credited pre-tax, BSE venue bars and a corrected share-action tape (actions re-keyed to the symbol that traded, combined split/bonus filings, split legs inside bonus rows, consolidations; every added leg price-gated); liquidatable NET CAGR: S2-v2 15.85% · S1-v1 12.38% · S2-v1 12.03% · S1-v2 9.18% · S5-v1 7.64% · S4-v1 6.64% · S4-v2 4.41% · S6B-v2 3.94% · S5-v2 0.54% vs Nifty 500 TRI 14.04% (price 12.63%). S2-v2 is 1.81pp above the TRI at max drawdown −57.8%; the other eight trail it; all nine are in-sample survivors of a 35-variant sweep. S2-v2's margin is within the path noise the re-measure itself shows: one ₹1.1 lakh share correction moved its endpoint by −0.76pp, and correcting the candidate scan moved the nine books by +0.66pp on average (sd 1.06pp). It is a measurement, not an edge. The structural stop (v2) lowered CAGR in S1, S4 and S5 and raised it in S2. Superseded: the settled figures (S2-v2 13.12% · S1-v1 12.15% · S2-v1 10.70% · S5-v1 7.46% · S1-v2 7.14% · S4-v1 5.97% · S6B-v2 2.57% · S4-v2 2.53% · S5-v2 0.23%), whose tape missed share events and booked phantom one-session losses on held names. Superseded: the 2026-09-28 price-only figures (S1-v1 8.96% · S2-v1 8.72% · S2-v2 8.66% · S1-v2 5.53% · S5-v1 5.32% · S4-v1 4.76% · S6B-v2 0.38% · S5-v2 −1.80% · S4-v2 −2.92%; runner, no dividends, 65 holdings written to zero). The +34.55% / +25.77% / +41.5% figures that preceded all three came from same-day-close fills, exact-stop fills and a two-condition SQL proxy, and are void. | [Scope blocked: these exact S1-S6B v1/v2 entry-plus-stop configurations as fundable books. Reopened by: a variant that beats Nifty 500 TRI net of cost, with dividends and settlements credited, on a period sealed before the rules were chosen. met the dividend-and-settlement half and corrected the tape; S2-v2 now clears the TRI in-sample, within path noise, and the sealed period is still owed.] Re-walks the Tier-2 Launchpad failure in this file (S1 Coiled-Launchpad +4.0%, "no survivor net of costs") without having cited it. |
| EARN_YIELD (cheap on P/E) | return/vol 0.70 · alpha +0.4% · MaxDD −71% | [Scope blocked: standalone long-ranking on P/E (top-25 monthly, 2012-26 vs Nifty 500). Reopened by: composite valuation/quality filter, earnings yield with growth/quality hurdle, or lower-turnover tenure.] No index-beating edge standalone; deep drawdown. |
| Float-normalised demand (float absorption) as a signal (2026-08-18) | the research record metric Public, 1,379 symbols with >=8 quarterly observations: within-symbol sd of Public% median 1.97pp against cross-sectional sd 18.27pp = 9.3x more cross-sectional than time-varying; within-symbol range median 6.43pp; 613/1,379 (44.5%) moved <=5pp across their ENTIRE history | [Scope blocked: treating float-normalised volume as a distinct demand/timing signal without signed flow. Reopened by: genuine signed flow beating plain delivered volume, delivered value / full market cap, and static float.] The denominator carries almost no time-series information, so within a symbol float-normalised demand is rank-identical to plain delivered volume for ~45% of the universe and near-identical for most of the rest. All timing content sits in the numerator, which IS the simpler baseline. Public float is a STRUCTURAL context field (how much of the company is tradable at all), never a demand or timing signal. A genuine absorption claim needs SIGNED flow, and signed flow here is Accumulation, failed OOS, descriptor-only - so beat that record first. Any new demand metric must separate itself from three frozen challengers: plain delivered value; delivered value / full market cap; public float as static context. A metric that cannot is a rescaling wearing a new name. |
| Accumulation-footprint detector v1 (2026-07-05b) | pre-registered gate FAIL 1/4 (only trade-size cleared δ≥+0.20 vs both controls: +0.329/+0.250); 764/947 episodes had NO pre-public window (SEBI PIT T+2); n=54 usable | [Scope blocked: front-detecting insider filings from raw tape alone at SEBI PIT T+2 granularity. Reopened by: campaign arcs E-04, disclosure drift E-03 under fresh pre-registration.] "Front-detect the insider from the tape" is structurally near-impossible in India at filing granularity. deliv_per showed ~no case elevation (δ≈+0.07) — consistent with Accumulation's alpha failure. Survivor: avg-trade-size ratio = descriptive column only. Follow-ups (campaign arcs E-04, disclosure drift E-03) require fresh pre-registration. |
| Interest cover (or any fragility screen) sold as a FREE filter (2026-08-18) | fund_panel, 37,842 PIT observations / 1,026 symbols, 2012-26, blow-up = fwd_252 < -0.50. DOWNSIDE: icov >= 3 blows up 601/29,280 (2.05%) vs 489/8,562 (5.71%), RR 0.3594, symbol-clustered 95% CI [0.2713, 0.4867]. UPSIDE GIVEN UP: the same split doubles 1,951/29,280 (6.66%) vs 939/8,562 (10.97%), winner-RR 0.6076 CI [0.5162, 0.7189], mean 12m return +18.25% vs +24.72%. Volatility-adjusted downside RR 0.5475 [0.4249, 0.7210]; 0.6678 [0.5315, 0.8681] holding volatility within date. Within-date placebo shuffle centres at 0.85, not 1.00 | [Scope blocked: one-sided marketing of interest cover as free downside protection without disclosing ~33% suppression of upside doubles. Reopened by: risk-budgeting/position-sizing input carrying matched two-tail metrics (downside 0.6678, upside 0.6830).] The separation is REAL and survives industry (MH 0.3665), size (0.3679), both (0.3806) and filing staleness (0.3616) - but it is risk compensation, not skill: the fragile arm blows up 2.8x more often AND doubles 1.6x more often AND earns 6.5pp/yr more on the mean. About half the raw effect is volatility - hold volatility fixed and 0.36 becomes 0.67. Never screen, rank or size on interest cover without carrying the winner-RR 0.6076 and the volatility-adjusted 0.6678 on the same surface; a one-sided rendering of this result is the defect, not the number. Survivorship: 9 of the panel's 1,026 symbols are dead against 2,326 of 5,956 in the price archive, so true blow-up rates are understated by an unmeasured amount. LENDER-CONFOUNDER CUT COMPLETED — left the volatility-controlled half uncomputed (said so in its own output); it is now measured on the published estimator, anchor reproducing digit-for-digit: MH vol×date 0.667754 → 0.667808 excluding financials [0.5300, 0.8685], crude blow-up 0.3600 → 0.3610, winner 0.6073 → 0.5896. A paired clustered bootstrap (the correct test for 96%-overlapping samples) gives MH Δ +0.0004 [−0.0179, +0.0220] and crude Δ +0.0012 [−0.0143, +0.0161], but winner Δ −0.0168 [−0.0450, +0.0056] — nearly excluding zero, so "immaterial" is right for the downside and overstated for the upside. 🔴 The cut is a WEAK instrument and does not clear the lender objection: 73.8% of financial panel rows have NULL icov (a bank has no computable interest cover), so lenders self-exclude before the cut runs and only 49 financial symbols are usable — a power gap, not a clearance. All of it inherits official_sector being LLM-derived, non-primary and non-PIT. 🔴 THIS ROW'S OWN BOTH-TAILS RULE MANDATED A MISMATCHED PAIR, AND THE NUMBER THAT FIXES IT DID NOT EXIST. The instruction above says to carry winner-RR 0.6076 beside volatility-adjusted 0.6678. Those are not two tails at matched controls: 0.6076 is the crude upside and 0.6678 is the volatility-controlled downside — a one-sided rendering of the control, which is this very defect in a new dress. The volatility-controlled winner RR existed nowhere in the estate (computes the right tail crude only, never stratified). ✅ It is now computed, by TWO independent seats that never shared code, each first calibrated against three published anchors (crude blow 0.359393, crude win 0.607569, MH blow 0.667754 — all reproduced exactly): MH vol×date WINNER RR = 0.683004 (ALL; 574 strata, N=37,834) and 0.668738 (ex-financials; 571 strata, N=36,284 / 977 symbols). Both seats landed all six digits on both numbers. THE FINDING: both tails are compressed by ~33%, and they are the same number. Downside 0.667754 vs upside 0.683004 (ALL); 0.667808 vs 0.668738 (ex-fin). Hold volatility fixed within date and icov ≥ 3 suppresses doubling almost exactly as much as it suppresses blow-ups — so it selects lower-dispersion firms, and the entire crude asymmetry (0.3594 down vs 0.6076 up) is a volatility artefact that vanishes with the control. Within a volatility×date bucket, high interest cover carries no evidence about the sign of the next 12 months, only about the width of the distribution. It is legitimate as a risk-budgeting / position-sizing input and is not a ranking key or an alpha screen. 🔴 AND THE GAP IS NOT MEASURABLE — do NOT read 0.683 > 0.668 as a finding. Paired symbol-clustered bootstrap, 1,000 draws over 1,026 symbols, quintiles frozen at full sample (the published boot() convention): MH blow [0.5363, 0.8545], MH win [0.5847, 0.8063], paired Δ(win—blow) = +0.0153, 95% CI [-0.1720, +0.1944], P(Δ<0)=0.450. The two RRs are statistically indistinguishable, which supports they are one number and simultaneously forbids ranking them. CORRECTED RULE, superseding the sentence above: carry 0.6678 with 0.6830 — matched controls, never 0.6678 with the crude 0.6076. Load-bearing caveats: 190 of the 574 contributing strata have zero winner events in BOTH arms, so the estimate rests on 384 strata (81 have an arm with <5 rows; median contributing stratum 51 rows); the winner event is 7.64% of rows, not rare, so RR and OR diverge — MH-OR is 0.601868, and 0.683 must never be quoted beside an odds ratio; overlapping 252-day forward windows leave horizon-overlap dependence that symbol clustering does not address; no CI was computed for the ex-financials winner 0.668738 (point estimate only). Methodological pin, proven not assumed: the ex-fin convention is re-rank quintiles within the subset — that reproduces the 0.667808 anchor exactly, where keeping ALL-rows quintiles gives 0.666683 and does not. Winner predicate read from |
| Accumulation tape as alpha | Deflated-Sharpe DSR 0.45→0.36 when added | [Scope blocked: signed-flow accumulation tape as standalone predictive alpha ranker. Reopened by: descriptive classification, institutional footprint filter, or structural context.] Descriptor-only; adds nothing. Do not re-test as alpha. |
| MOMENTUM BAND + RSI single-name swing (2026-07-22) | Upper-band breakout entry (T=EMA5(HLC3) > EMA13(high)) + RSI/2-fractal managed exit, pre-registered, 45,131 events / 124,832 trades 2012-26: EVENT median 22d excess −0.90%, Cliff's δ vs placebo −0.012 (FAIL-null, negative both halves); BOOK net return/vol 0.53 (h1 0.19 / h2 0.87, both <0.89), raw CAGR 19.4% → net 8.4% (cost −11pp) | [Scope blocked: short-term swing trading with hair-trigger RSI stops churning 5-bar holds under high trade costs. Reopened by: multi-week/multi-month holding tenures with wider structural stops.] The "buy strength" edge of the STREAM BAND band is ALSO an anti-signal; RSI-as-stop churns (phase-1 hair-trigger, 94% exits, 5-bar hold) and the RSI-80 partial hurts (−0.25pp mean = profit-taker law again, 07-14e). Beats random-entry (+0.20) but nowhere near fundable. Full entry: § Study 2026-07-22; cites 07-13/14b/14c/14d/14e. |
| Momentum sold as a FUNDABLE strategy | GROSS return/vol 1.29 → NET ~0.09, CAGR negative, MaxDD −69% under realistic cost (~36%/yr, ~100%/mo turnover) | [Scope blocked: high-turnover top-25 monthly rebalance (~100%/mo turnover) under full Zerodha/participation costs. Reopened by: low-turnover momentum (e.g. semi-annual/annual rebalance reproducing NSE Nifty Momentum index method, turnover ≤30%/yr), multi-quarter tenure, net of real Zerodha cost.] The headline return/vol is a flat-cost illusion. Nothing beats Nifty-500 buy-&-hold (0.89) net of realistic cost. Momentum = a gross selection/analytical lens, not net alpha; any fundable form must be low-turnover (and is then defensive, not alpha). |
| Patearn NS 14-pattern composite as the EXPLANATION of its own risk result (2026-08-18) | Preregistered 3-way ablation on fund_panel, 43,412 PIT monthly observations 2012-26, blow-up = fwd_252 < -0.50: dropping patterns 3/7/11/14 and rescaling the tier thresholds moves relative risk 0.4547 -> 0.4535 with the event counts IDENTICAL (422 good / 904 bad in both arms); on the common subset a two-variable screen (icov >= 3 AND de <= 0.8) scores RR 0.4443 vs the full machine's 0.4616, wins 2012-18 outright (0.3343 vs 0.4072) and loses 2019-26 (0.6554 vs 0.5620) | [Scope blocked: attribution of risk separation to the 14-pattern composite without demonstrating value-add over two-variable screen. Reopened by: verified proof that composite beats simple ratios with symbol-clustered CI excluding 1.0.] The risk separation is REAL and reproducible (stored tiers rebuild from stored inputs with 0 mismatches in 43,412 rows); the ATTRIBUTION to fourteen patterns is FALSIFIED. Patterns 3/7/11/14 are hard-coded (1, False) constants = 25.8% of NS weight carrying zero information; 121 of the 291 sub-signal weight units (41.6%) are constants; the 55/40 cutoffs are calibrated around that floor (renormalising without rescaling admits 2,281 more rows and WORSENS separation to 0.4680); T1 has never fired once in 43,412 observations. Never sell, rank, or cite the composite as fundamental skill without first beating RR 0.4443 on the common subset with the sign holding in BOTH halves. Round 4-5 update (2026-08-18): the separation ITSELF survives its confounder - completeness-stratified Mantel-Haenszel RR 0.5105 (vs 0.4547 unstratified), and a symbol-clustered bootstrap (2,000 draws over the 1,117 symbols underlying the 43,412 rows) gives [0.3919, 0.6565] - excluding 1.0, but NOT clearing the preregistered 0.55 bar on its upper bound. Label is DESCRIPTIVE: teachable with its receipt, never a ranker. The composite's one surviving claim (that it wins where disclosure is complete) DECAYS as power rises - tier-minus-screen RR gap 0.35 at >=13/14, 0.24 at >=12, 0.02 at >=11, the only threshold clearing 100 events/arm. CORRECTION (same session, 2026-08-18): the ordering claim above is WITHDRAWN. The pooled screen/tier RR ratio is 0.9014 with a symbol-clustered 95% CI [0.6925, 1.1813], which INCLUDES 1.0, and the same ratio spans 1.0 at every completeness threshold - [0.2296, 1.1164] at >=13/14, [0.3428, 1.3124] at >=12, [0.6147, 1.4635] at >=11. The screen does NOT demonstrably beat the composite; on this panel they are INDISTINGUISHABLE. The block stands on the weaker and correct ground that the composite has never demonstrated value-add over two ratios in EITHER direction - you cannot sell as skill a difference you cannot show. The promotion bar above is SUPERSEDED: to be retained, the composite must show a symbol-clustered CI on the tier/screen RR ratio that EXCLUDES 1.0. Unaffected because arithmetic not inference: removing patterns 3/7/11/14 leaves event counts IDENTICAL (422/904); 121/291 weight units (41.6%) are hard-coded; thresholds are entangled with that floor; NS_max = 72.148 so T1 cannot fire. |
| PEAD tradeable book (event-time, 2026-07-05) | ALL constructions fail: trailing net return/vol 0.10, no-delivery 0.02, within-season 0.06 (pre-registered), HEDGED −0.58, 1.5× cost −0.32 — vs bench 0.85, both halves | [Scope blocked: event-time trading wrappers around earnings surprises without execution advantage. Reopened by: execution-aware primary XBRL drift models beating recorded numbers under no-leak harness.] Event drift is REAL descriptively (A-study SUE-Q5×DELIV-T3 CAR60 +7.62%, t_cohort 1.92) but no wrapper survives real-time ranks + costs + compounding; the within-season variant (the last untested cell) also failed. Descriptive event lens only. Do not re-attempt any PEAD book without beating these exact numbers under the same no-leak harness. |
| QUALITY standalone | return/vol 0.76 · alpha ~0.0% · fails halves | [Scope blocked: standalone quality ranking (ROCE, D/E, OPM, interest cover; top-25 monthly). Reopened by: quality as veto/risk filter or combined with momentum (QUAL_MOM); non-standalone multi-factor sleeve.] Quality doesn't rank returns alone; only helps attached to momentum (QUAL_MOM). → C is a veto/filter, not a ranker. |
In the failure ledger's own words, with its internal references (decision numbers, file and line citations, session tags) left out. Ratios in these rows are return/vol — mean return divided by volatility, with no risk-free rate subtracted. Where a row names a published statistic from the literature, that name is the literature's, not our label for our own number.
Take it with you
the print / procurement packSource: assembly.
Assemble the print pack → one continuous document: the boundary statements, the pre-registered gates, the coverage matrix and this record — for a reviewer who wants it offline.