candlebench

Run

Settings

Trials Seed Holdout fraction Minimum dates Bootstrap samples Experiments tried Windows Source Lookback days Reward ×R Stop buffer Max hold bars Cost model Fixed slippage bps Symbols Rank by

Timeframes

Patterns

Leaderboard

EDGE — corrected discovery evidence confirmed on later held-out dates NOISE — no confirmed edge NEGATIVE — reliably loses INSUFFICIENT — too few trades, dates, or no usable control
★ = random-entry control, the noise floor for its direction. A pattern that does not beat it has shown nothing.
Click any row to see its equity curve and the individual trades behind it.

Compare runs

Baseline Against

Drill-down

Evidence sample

Cumulative R in the order the trades happened the dashed line is the comparison the statistics used: this pattern’s matched random entries (same direction and stop distance, entering 1–5 bars later), scaled to the same number of trades; runs saved before those were stored show the direction’s random-entry reference row instead. The shaded band is the deepest drawdown

Breakdown one ticker carrying the result, and the opening auction, are the two usual explanations for an apparent intraday edge

Group by

Charts

Expectancy with 95% interval pointwise market-date cluster intervals; verdicts use corrected tests

Edge over the random-entry control a positive average alone does not establish an edge; read paired intervals and adjusted p-values

Run

Result

Leaderboard

Compare runs

Drill-down

Charts