Verdict: Swing Setup Score (M-014) is rated “NOT RELIABLE” in its validation review dated 3 Sep 2026.
Key result: Decile 10 vs 1 up-rate (in-sample re-weight test): 56.9% vs 51.8%.
Key result: Decile 10 vs 1 median 10d return: +1.03% vs +0.25%.
How it was tested: the record does not state separate training and test periods; out-of-sample design: none.
Horizon: 10 sessions (the window the decile base rates are measured over). Coverage: ~1,200-1,700-name panel (S&P 500 + Nifty + curated watchlists).
Stage: Retired to heuristic. Version v1; the hub record is as of 6 Nov 2026.
The verdict
NOT RELIABLE as a screening or ranking input: the composite sign-flips out-of-sample and was already removed from M-011 for that reason on 2026-08-15; the tab's own display as descriptive context is the appropriate remaining use.
Development summary: The 2026-07 re-weight replaced flat momentum components with vol-expansion + pullback depth + long-term RS on a modest in-sample decile signal (56.9% vs 51.8% up-rate) that then sign-flipped OOS; dropped from M-011 scoring 2026-08-15.
What the review allows
As descriptive, per-symbol component context (which of RS/vol-expansion/pullback is elevated) on the Swing tab only.
What the review rules out
Use the Setup Score or its deciles to screen, rank, or size positions — it sign-flips out-of-sample and was removed from M-011 for exactly this reason.
Assume the in-sample decile separation (56.9% vs 51.8%) generalizes forward.
What it predicts
Item
Detail
Question it answers
Which names are most entry-ready right now, combining relative strength, trend quality, volume accumulation, base tightness and pullback depth into one 0-100 score?
Method
Weighted composite score
How it works
Weighted composite (empirically re-weighted, not equal-weighted); decile-conditioned forward-10-session base rates backtested over a 4-year multi-market grid
The hub's standard metrics for this model class (only those the record measured).
Metric
Sample / window
Compared with
Value
Range (CI)
Precision at k (headline)
in-sample · top decile up-rate
decile 1 51.8%
56.90%
n/a
What the validation review found
Is the design sound? The underlying instinct (measure which components predict, don't assume) is sound; the failure is that the measurement itself did not generalize out-of-sample, which the owner correctly acted on by downgrading it.
Backtest. In-sample only in the reviewed evidence; the out-of-sample sign-flip is recorded as a fact (and acted on) but its own test design is not in the material reviewed here.
Limitations
Sign-flips out-of-sample — the primary reason it was removed from M-011's scoring 2026-08-15; the mechanism of the flip has not been diagnosed in available evidence.
The in-sample decile separation used to justify the 2026-07 re-weight was itself modest (56.9% vs 51.8%), a thin margin to build a screening input on.
No walk-forward or held-out re-test has been run since the downgrade; status is 'retired to heuristic' rather than actively re-validated.
Questions
Is the Swing Setup Score validated?
Its validation review dated 3 Sep 2026 rated it NOT RELIABLE as a screening or ranking input: the composite sign-flips out-of-sample and was already removed from M-011 for that reason on 2026-08-15; the tab's own display as descriptive context is the appropriate remaining use.
What does the Swing Setup Score predict?
It answers: Which names are most entry-ready right now, combining relative strength, trend quality, volume accumulation, base tightness and pullback depth into one 0-100 score? Output: 0-100 Setup Score per symbol, plus decile-level forward R/up-rate/expected-move context. Horizon: 10 sessions (the window the decile base rates are measured over). Coverage: ~1,200-1,700-name panel (S&P 500 + Nifty + curated watchlists).
How was the Swing Setup Score tested?
Training data not stated, test data not stated, out-of-sample design: none. Decile 10 vs 1 up-rate (in-sample re-weight test): 56.9% vs 51.8%; Decile 10 vs 1 median 10d return: +1.03% vs +0.25%; Out-of-sample stability: sign-flips — dropped from M-011 scoring 2026-08-15.
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.
Base rate: how often the outcome happens with no model at all.
Confidence interval: the range that likely contains the true value (95% unless stated); [a, b] on this page is that range.
ATR: average true range: a stock's typical daily price range, often expressed as a percentage of its price.
Equal-weighted (EW): every name counts the same, whatever its size.