Verdict: Swing Excess-Return Ranker (M-022) is rated “FIT WITH CONDITIONS (US ranked candidate list; unchanged from v1)” in its validation review dated 17 Sep 2026.
Key result: US rank IC (WF 2016–25): +0.0244 [+0.0067, +0.0435].
Key result: Non-US pooled rank IC (OOS 2019–25): +0.0472 [+0.0303, +0.0628], unhaircut.
How it was tested: training data 2003-01→2025-06, test data 2016-01→2025-08; out-of-sample design (how it was checked on data it never saw while being built): walk-forward yearly.
Forward record: 21 of 38 published lists have reached their 21-session horizon. Average pick return minus SPY's over the same dates: -2.86 percentage points (standard error 0.44). 37% of the matured picks beat SPY over the horizon.
Stage: Validated (US list) — FIT WITH CONDITIONS; non-US blocs in SHADOW. Version v2; the hub record is as of 6 Nov 2026.
The verdict
FIT WITH CONDITIONS (US ranked candidate list; unchanged from v1). US rating unchanged — v2 raised nothing and lowered nothing (0/7 challengers, 0/4 rules, S3h FAIL). Non-US: NOT FIT as a bloc-publishing system; only India and Korea carry enough verified evidence to publish a SHADOW list.
Development summary: US unchanged: real ranking power (IC +0.024) and a positive risk-adjusted top-10, no book. v2 adds a pooled 22-market model (non-US IC +0.047, unhaircut) whose lists stay SHADOW — only India and Korea are cleared to publish one.
What the review allows
As one input among several into a discretionary swing shortlist, in the momentum sleeve, provided a 12-1 momentum ranking is run beside it and the two lists are compared before selection.
As a research artefact and as the incumbent in the vendor benchmark, provided the champion's own failure of BENCHMARK_PROTOCOL section 4 is disclosed to every entrant.
As the source of a settled-vintage outcome log (picks log) — the forward record is the most useful evidence this model can now produce.
Non-US, India and Korea only: a top-10 SHADOW candidate list within that single market, read as one input, never as a book and never compared across markets.
Non-US, Japan / Canada / France / Switzerland / Italy / UK: the within-market ranking only — no published list.
What the review rules out
Be run as an autopilot 10-name book expected to beat SPY: alpha −0.4%/yr with t = −0.08, +0.28%/yr against SPY's +14.9% over 2021-2026, drawdown 1.23x SPY at 0.64x its Sharpe.
Be described as a 'validated signal', as having met its pre-registered criteria, or as chosen by a pre-registered champion rule.
Be claimed to add value over 12-1 momentum, or to beat any benchmark on the strength of the full-decade average without the 2016-2020 / 2021-2026 split.
Have its 2016-2025 economics re-quoted after the EDGAR survival-coverage defect is fixed without re-running the walk-forward.
Be described as validated in any non-US market: no non-US bloc is validated, every non-US figure is unhaircut for survivorship, and Hong Kong, Germany, China, Taiwan, Australia, Netherlands, Spain, Sweden and Singapore are NOT FIT.
Publish a Europe bloc list (SIGNAL-ONLY, no picks), a list for Denmark, Finland, Norway or Poland (no measurable IC), or any cross-market comparison of two markets' ICs — the ranks and labels are within-market by construction.
What it predicts
Item
Detail
Question it answers
Which liquid US names will beat the universe over the next 21 sessions — ranked weekly for a 10-name swing book held about a month? India and Korea are ranked the same way, in SHADOW.
Weekly ranked list; production pick = top-10 by score/ATR, one per sector; lift bands by rank bin (rank-valid only). Non-US output is a SHADOW list for India and Korea only — relative ranking within one market, unhaircut for survivorship, never a book.
Validated (US list) — FIT WITH CONDITIONS; non-US blocs in SHADOW
Version
v2, first recorded 9 Sep 2026
Validation evidence
Training data: 2003-01→2025-06. Test data: 2016-01→2025-08. Out-of-sample design: walk-forward yearly.
Headline results from the model's development record, as the hub publishes them.
Measure
Result
Compared with
Note
US rank IC, walk-forward 2016–25
+0.0244
12-1 momentum +0.0071 · random +0.0002
US risk-adjusted top-10 excess per entry
+1.05%
12-1 momentum +1.30% · ATR rank -0.32%
US v2 challengers passing the swap rule
0 of 7
gate S3h also FAILS (CI [−0.43, +1.58])
Non-US pooled rank IC, OOS 2019–25
+0.0472 [+0.0303, +0.0628]
12-1 momentum +0.0251 (CI spans 0) · US-only transfer +0.0239
Non-US blocs cleared to publish a list
2 of 21 (India, Korea)
6 ranking-only · 9 NOT FIT · 4 not measurable · Europe bloc NOT FIT
CSCV PBO (v1 ledger; stale for v2)
0.10
30 configurations (F-V3-05 open)
The hub's standard metrics for this model class (only those the record measured).
Metric
Sample / window
Compared with
Value
Range (CI)
Rank IC (headline)
WF 2016–25 · US
12-1 momentum +0.0071
0.0244
[0.0067, 0.0435]
ICIR
WF 2016–25 · US
9 of 10 years positive
0.1783
n/a
Excess return at horizon
WF 2016–25 · US top-10 · gross
12-1 momentum +1.24
+0.74 pp
[-0.09, 1.8] pp
Forward record: 21 of 38 published lists have reached their 21-session horizon. Average pick return minus SPY's over the same dates: -2.86 percentage points (standard error 0.44). 37% of the matured picks beat SPY over the horizon.
What the validation review found
Is the design sound? A 21-session within-date-demeaned excess return is a coherent rank estimand but not the estimand for a book judged vs SPY: the label says nothing about the universe's own return, and that universe lagged SPY by 6.7pp/yr. Lengthening the horizon and dropping the take-profit answer M-021 correctly. But the champion rule was written after results existed, and as written the criteria select nothing.
Data quality. Labels and block-A features reproduce exactly from the raw panel; the EDGAR after-open reaction rule is correct; XBRL uses first-filing dates. Two defects: the universe is ~95% survivor-only across the test decade, recovered delistings flat-padded at zero volume; and the EDGAR symbol map matches 0% of delisted names, so coverage flags encode survival. XBRL lacks the 8-K block's time-of-day guard.
Backtest. The book engine is confirmed independently (CAGR 11.97% vs 12.0%, beta 0.943 vs 0.94, alpha t −0.08 both). The book is a beta-0.94 portfolio with no alpha: −3.4pp/yr against SPY on CAGR, Sharpe 0.57 vs 0.89, drawdown −41% vs −34%. Split in half, +28.6%/yr in 2016-2020 and +0.28%/yr in 2021-2026 — the second half also lost to its own equal-weight universe by 9.4pp/yr.
Against simple benchmarks. The ladder is well built and the protocol offered to vendors is fair. The champion beats random (+1.62pp, CI > 0) and 1-month reversal (+1.01pp, CI > 0) but not 12-1 momentum (−0.25pp, CI [−1.77, +1.11]) — the very test the protocol imposes on vendors. Nulls confirm the rule is not doing the work: constant score / ATR one-per-sector returns −0.74% and the ten lowest-ATR names −0.72%. v2: the global gates never require beating a benchmark (F-V3-12) — a free 12-1 momentum rule beats the model on both legs in Hong Kong and Germany, which is why both publish no list despite passing the developer gates.
Independent replication. Complete. Pooled and per-year rank IC, all four selections, the acceptance window, the benchmark ladder, the book CAGR/beta/alpha and today's ten picks all reproduce from independently written code to <= 1e-4 (ranks exact). No unexplained discrepancy was found in any reported number. v2: exact in both parts — US, 10 configurations x 2 metrics to 6+ decimals with maximum absolute difference 0.0; global, 18 markets x (IC, CI, top-10 excess, acceptance) with max |dIC| 0.0 and max |dCI bound| 5.7e-07pp, and the champion plus the empty override set re-derived from the SPEC rule.
Sensitivity. Robust to cost (+0.24% vs SPY at 50bp; negative only at 100bp), entry (close +1.07% vs open +1.05%), gap veto (+1.11%) and floor ($100M +0.80%). Fragile to the risk control: an ATR<=5% cap halves it to +0.46%, <=3% kills it (+0.07%), and the 3xATR stop costs 0.26pp/entry vs SPY and takes years-above-SPY from 6 to 4. K is inert above 12 (only 12 sectors). Seeds move the same statistic twofold.
Monitoring. A monitor exists and runs daily (freshness, coverage, top-30 overlap, settled 21-session outcomes with a RED at n>=60 and mean-net CI below zero) — better than the brief assumed. It cannot fire the documented retrain trigger (no settled-vintage IC is computed), has no input-drift/PSI panel, and never compares the settled picks with the benchmark this validation shows to be decisive. v2: adds per-bloc freshness (RED above 5 days), coverage against the walk-forward median (AMBER below 70%), seed rank agreement (AMBER below 0.70), list turnover and settled 21-session outcomes for both the production and the shadow challenger list; the first run (2026-09-17 10:01) is GREEN with no alerts.
Limitations
Rank IC +0.024 is a real but small signal; the top-10 is a noisy basket (share of negative trades ≈ 0.50, 5th percentile -19.62%).
Under the pre-registered production rule the champion does NOT beat the 12-1 momentum benchmark: paired difference −0.25pp per entry, CI [−1.77, +1.11]; −6.5pp/yr at book level (validation C3).
Two-period split (validation C4): 2016–2020 book +28.6%/yr vs SPY +16.0% and its own EW universe +15.2%; 2021–2026 book +0.28%/yr vs SPY +14.9% and EW +9.7% — since 2021 the selection lost to its own pond by 9.4pp/yr; the 'pond lagged SPY' reconciliation is withdrawn.
Survivorship (validation V-06): ~95% of the decade is survivor-only; the EDGAR symbol map matches 0% of delisted names, so the coverage flags encode 2026 listing status — economics must not be re-quoted until the map is rebuilt (C6).
Signal decays in recent years (linear-score IC 0.033 in 2016–21 → 0.017 in 2022–25).
XBRL fundamentals exist only from 2009; several ratios have < 50% coverage; missingness is informative and kept native.
Questions
Is the Swing Excess-Return Ranker validated?
Its validation review dated 17 Sep 2026 rated it FIT WITH CONDITIONS (US ranked candidate list; unchanged from v1). US rating unchanged — v2 raised nothing and lowered nothing (0/7 challengers, 0/4 rules, S3h FAIL). Non-US: NOT FIT as a bloc-publishing system; only India and Korea carry enough verified evidence to publish a SHADOW list.
What does the Swing Excess-Return Ranker predict?
It answers: Which liquid US names will beat the universe over the next 21 sessions — ranked weekly for a 10-name swing book held about a month? India and Korea are ranked the same way, in SHADOW. Output: Weekly ranked list; production pick = top-10 by score/ATR, one per sector; lift bands by rank bin (rank-valid only). Non-US output is a SHADOW list for India and Korea only — relative ranking within one market, unhaircut for survivorship, never a book. Horizon: 21 sessions (weekly entries, overlapping tranches). Coverage: US listings ≥$10M/day, ≥$5, ≥6m listed (~2,300), trained ≥$2M/day incl. recovered delistings 2016→; 21 non-US markets ≥USD 5M/day, ≥USD 1, survivor-only 2016→.
How was the Swing Excess-Return Ranker tested?
Training data 2003-01→2025-06, test data 2016-01→2025-08, out-of-sample design: walk-forward yearly. US rank IC (WF 2016–25): +0.0244 [+0.0067, +0.0435]; Non-US pooled rank IC (OOS 2019–25): +0.0472 [+0.0303, +0.0628], unhaircut; Non-US blocs cleared to publish a list: 2 of 21 (India, Korea), both SHADOW. Forward record: 21 of 38 published lists have reached their 21-session horizon. Average pick return minus SPY's over the same dates: -2.86 percentage points (standard error 0.44). 37% of the matured picks beat SPY over the horizon.
Out-of-sample: results measured on data the model never saw while it was built or tuned.
Walk-forward: the model is refit on data up to a date and tested on the period after it, then the window rolls forward, so every test is out of sample.
Rank IC: information coefficient: the rank correlation between a model's scores and the returns that followed; 0 means no skill.
Percentage points (pp): the plain difference between two percentages: 12% against 10% is 2 pp.
Basis points (bp): hundredths of a percentage point: 25 bp is 0.25%.
Standard error (s.e.): the typical size of the error in an estimate; a result more than 2 standard errors from zero is unlikely to be luck.
Confidence interval: the range that likely contains the true value (95% unless stated); [a, b] on this page is that range.
SPY: the SPDR S&P 500 ETF, the US market benchmark used here.
ATR: average true range: a stock's typical daily price range, often expressed as a percentage of its price.
Shadow: scored and tracked on paper only; no list is published and nothing is acted on.
12-1 momentum: the return over the past 12 months, leaving out the most recent month.
Equal-weighted (EW): every name counts the same, whatever its size.
Survivorship: a sample that leaves out companies that later delisted, which flatters historical results.
PSI: population stability index: how far today's inputs have drifted from those the model was built on.