Verdict: Chart Pattern Scanner (M-015) is rated “NOT RELIABLE” in its validation review dated 3 Sep 2026.
Key result: Backtested cells beating matched controls: 0 of 36.
Key result: Pennants: inverted (underperform the control).
How it was tested: the record does not state separate training and test periods; out-of-sample design: none.
Horizon: up to 40 sessions to resolve; forward outcomes read at 10/20 sessions from the breakout bar. Coverage: ~1,700-name panel (S&P 500 + Nifty + curated watchlists); symbols need >=300 bars.
Stage: Context-only. Version v1; the hub record is as of 6 Nov 2026.
The verdict
NOT RELIABLE as a trading signal — the detector's own 36-cell backtest found no cell beating a matched control — but the no-lookahead construction and honest context-only framing are sound and should be preserved as-is.
Development summary: A carefully no-lookahead-built detector (formation-bar stamping, EB pooling) whose own 36-cell backtest found 0 cells beating matched controls and pennants inverted — correctly kept context-only rather than presented as an edge.
What the review allows
As a descriptive 'what shape printed' read, exactly as labelled on its tab.
The formation-bar / no-lookahead construction as a template for future pattern-style detectors.
What the review rules out
Trade any of the 9 patterns on the odds shown — 0 of 36 tested cells beat a matched control.
Treat pennants as continuation signals — they measured inverted.
What it predicts
Item
Detail
Question it answers
Which of 9 classic chart geometries (flags, pennants, triangles, cups, doubles) are forming or have just broken out, and does this detector's own measured record support trading them?
Method
Zigzag pattern detector
How it works
ATR(14)-scaled zigzag pivots -> geometric detection at the formation bar (no lookahead) -> 40-session backtest walk to breakout/failure -> hierarchical EB pooling (pattern x region)
Inputs
ATR(14) zigzag pivots, 9 chart geometries, Breakout within 40 sessions, Pattern × region EB pooling, Daily OHLC history (300+ bars)
Output
Per-symbol pattern flag (forming / breakout) with EB-pooled brk_rate/tgt_rate/up10/up20 odds
Horizon
up to 40 sessions to resolve; forward outcomes read at 10/20 sessions from the breakout bar
Training data: not stated. Test data: not stated. Out-of-sample design: none.
Headline results from the model's development record, as the hub publishes them.
Measure
Result
Compared with
Note
Backtested cells beating matched controls
0
36 tested
9 patterns x forward outcome cuts
Pennants
inverted
matched control
underperform rather than confirm the textbook direction
Zigzag threshold
ZZ_K=1.5 ATR
3.0 first tried
3.0 made a pivot's confirmation move as tall as the pattern itself — untradeable by construction
The hub's standard metrics for this model class (only those the record measured).
Metric
Sample / window
Compared with
Value
Range (CI)
Cells surviving
2026-07-15 audit · corrected
36 tested
0
n/a
What the validation review found
Is the design sound? Well-designed as an honesty exercise: formation-bar stamping avoids lookahead, and a matched-control comparison is the right counterfactual. The finding (0/36) is taken at face value and correctly drives the context-only status.
Backtest. 36 cells (9 patterns x outcome cuts) against matched controls; pennants inverted. Regime span and train/test structure of this specific 36-cell result are not detailed in the reviewed evidence.
Limitations
0 of 36 backtested cells beat their matched control; pennants actively invert.
Context-only status means the tab still displays odds numbers a user could act on despite the null result — the framing (context, not signal) is the control, not a technical restriction.
No held-out/train-test split is described beyond the matched-control comparison; regime coverage of the backtest is not detailed in the reviewed evidence.
Questions
Is the Chart Pattern Scanner validated?
Its validation review dated 3 Sep 2026 rated it NOT RELIABLE as a trading signal — the detector's own 36-cell backtest found no cell beating a matched control — but the no-lookahead construction and honest context-only framing are sound and should be preserved as-is.
What does the Chart Pattern Scanner predict?
It answers: Which of 9 classic chart geometries (flags, pennants, triangles, cups, doubles) are forming or have just broken out, and does this detector's own measured record support trading them? Output: Per-symbol pattern flag (forming / breakout) with EB-pooled brk_rate/tgt_rate/up10/up20 odds. Horizon: up to 40 sessions to resolve; forward outcomes read at 10/20 sessions from the breakout bar. Coverage: ~1,700-name panel (S&P 500 + Nifty + curated watchlists); symbols need >=300 bars.
How was the Chart Pattern Scanner tested?
Training data not stated, test data not stated, out-of-sample design: none. Backtested cells beating matched controls: 0 of 36; Pennants: inverted (underperform the control); Status: context-only — displayed, not a validated edge.