Pick a market, then check off what actually happened today — a gap up, Asia sweeping yesterday's high, and so on. We show every pre-verified finding that matches, with its measured rate against the base rate, sample size, and confidence interval. Nothing here is computed on the fly — every row already survived the research pipeline's multiple-comparisons verification (Bonferroni correction across sectors, correlated-cluster checks). See methodology for how that works. Covers 6 pairs, pooled across weekday and weekend.
When we said a level had an X% chance of being touched, how often did it actually happen? Out-of-sample, bucket by bucket.
Mean gap
1.39 pp
Checks measured
707.8M
707,782,112
Out-of-sample window
—
PredictedRealized
Mean gap
1.70 pp
Checks measured
117M
117,041,556
Out-of-sample window
—
PredictedRealized
Mean gap
1.35 pp
Checks measured
236.5M
236,491,175
Out-of-sample window
—
PredictedRealized
These are touch probabilities from the model's own reliability check, not graded trade calls — for those, see the live Scorecard.
Per-sector breakdown, not per-symbol — the source data does not carry a per-symbol (e.g. ES vs NQ, or AAPL vs MSFT) calibration split, only per-sector/scope. 20-point-percentage bins (coarser than the 10pp futures/fx/stocks breakdown above) to keep per-sector cells sampled. Stocks: historical period 2020-2026, these 12 names selected ex-post (survivorship) — calibration applies to these names, not a random stock.
What happened today?
Pick one value per condition — or leave it at "not selected" to ignore it.
Matching findings
Select at least one condition above to see matching findings.
CI method: wilson score interval, 95%, computed on rate_held/n_held (ci_lo_held/ci_hi_held columns, unchanged). Source: crypto_conditional_tables;daypart=pooled;2026-09-30.