Versioned event rule
Each formation uses a fixed detector and definition version. Only closed candles and information available by the signal close may enter the event.
TLAP Pattern Lab
How Pattern Lab calculates historical results
Definitions, event independence, comparable baselines, late-period checks and limitations of the published research.
Method in brief
Independent outcomes
Overlapping outcome windows are removed. Up, down and unchanged counts use the same denominator for each exact configuration.
Comparable baseline
The baseline uses the same instrument, timeframe, horizon and higher-timeframe condition without the pattern trigger.
Late-period verification
Direction and any data-driven choice are made on earlier data; the later portion is retained as a separate check.
Multiple testing
Adjusted q-values are retained within the published family and scan scope. A high percentage alone is not confirmation.
Limits
Results describe historical price movement, not a trading strategy, probability forecast, investment recommendation or broker execution quote.
Chart-formation detectors · definition version 1.0
Only closed candles are used. A pivot requires 3 closed candles on each side. Boundaries are fixed from confirmed pivots; completion requires a close beyond the boundary, rather than a touch.
- Formation length: 12–120 candles; no more than 20 candles from the last pivot to completion. Prior movement uses the close 10 candles earlier.
- Head and shoulders: shoulder difference ≤ 25% of head height; neckline slope ≤ 25% of formation height.
- Double top/bottom: extreme difference ≤ 25% of retracement depth; retracement ≥ 1.5 ATR(14).
- Flags and pennants: impulse ≥ 3 ATR(14) in 20 candles; channel/convergence 4–20 candles; channel range ≤ 60% of the impulse.
- Triangles and wedges use at least two confirmed extrema on each side. A wedge’s late range is at most 70% of its early range. Up/down breakouts are evaluated separately.
The comparable baseline excludes detector-event windows with a complete valid outcome before thinning, including overlap with those signal windows. A rejected signal with a data gap, incomplete horizon or invalid price does not remove a control window. This is the same approved baseline policy used in RU; no statistical method changes with the interface language.
Reading statuses and missing data
“Confirmed”, “Preliminary”, “Not confirmed” and “Insufficient data” are separate stored research states. A missing row, network error or timeout is never presented as a zero result. Filters do not silently relax, and one pattern variant never supplies another variant’s chart or statistics.
The core and extended market scopes remain separate. Publications are not the same as every planned test; the coverage registry lists planned cells and reasons for gaps without inventing results.