R0268: WATCH

Descriptive statistics, no trading costs in the probabilities (the backtest columns alone use estimated costs). Rules were discovered and tuned on 2022-01 → 2024-12 only, frozen, then tested once on 2025-01 → 2026-10. Millions of hypotheses were searched, so read the corrected p-values and the out-of-sample columns, not the training hit rates. Not investment advice; not a trading system.

Rule: When price vs prior-day high > -0.25% (≈-61 pts at 25,000) AND 1-min Bollinger %B > 0.566 AND size of the opening gap > 0.12% (≈29 pts at 25,000) at 09:30

Target: price reaches the prior-day high after 09:30 (days not yet at PDH) · Decision time: 09:30 IST (uses bars up to 09:30 only) · Method: rulefit · Family: level / structure

69%train hit (n=32) vs base 26% · lift 2.67

83%OOS hit (n=18) vs base 23% · lift 3.55

[61%–94%]OOS Wilson 95% CI

2.5e-06OOS BH q (all) · family BH q 1.5e-06 · Holm 1.3e-04

5.0e-04OOS matched-random placebo p

Verdict criteria

✔ Out-of-sample lift in the same direction as train
✔ OOS Wilson 95% CI excludes the OOS base rate
✔ OOS BH q ≤ 0.05 across all candidates tested OOS AND within its target family
✔ Train BH q ≤ 0.05 (across all hypotheses searched)
✔ Stable by year (same direction in ≥80% of years with ≥10 days, incl. every OOS year with ≥10)
✔ OOS matched-random placebo p ≤ 0.05
✘ OOS sample ≥ 20 days

Train p 8.7e-07; train BH q 7.5e-05; train Holm 1.000 (corrections over all 9,242,869 hypotheses); whole-pipeline shuffled-label placebo family-wise p 0.347 (100 permutations). Extra conditions beat the best single condition OOS: True.

By year

year	rule days	hit	base (all days)	lift
2022	7	86%	22%	3.91
2023	12	58%	25%	2.32
2024	13	69%	30%	2.31
2025 (OOS)	9	78%	27%	2.92
2026 (OOS)	9	89%	19%	4.59

Spot-points backtest (estimated costs)

enter long at 09:30 price, take profit at PDH; stop 0.4% of entry; exit at close. Costs: Nifty futures round-trip model from nifty-setups-backtest/src/costs.py (brokerage, exchange, SEBI, stamp, GST, STT by date: 0.0125% → 0.02% from Oct-2024 → 0.05% from Apr-2026) + 0.5 pt slippage per side; spot used as a futures proxy (basis ignored); stop-first on ambiguous bars.

window	trades	win rate	avg net pts	avg gross pts	total net	max DD pts	PF
train 2022–24 (in-sample)	32	50%	-28.9	-22.5	-924.2	+924.2	+0.3
OOS 2025–26	18	78%	-1.8	+9.2	-32.2	+175.8	+0.9

equity curve

trades.csv · Generated 2026-10-10 00:44 IST.