8 popular Nifty intraday option-buying strategies: honest backtest, by period and by market regime

Built 2026-10-27 20:34 IST. Research only, not advice. Read-only data, no orders. Option buyer only (long = buy ATM CE, short = buy ATM PE, nearest weekly with DTE≥1). REAL costs ≈2 pts per round trip. Everything below was pre-declared in PREREG.md (19:20 IST, before any 2026 number). Regime addendum after Satya's 19:17 IST question: “could it be each strategy works for certain periods”.

Short answer.

None of the 8 popular setups is a proven option-buying edge. On the 2024–25 train, all 8 lost money (−2.9 to −19.7 pts per trade). In 2026, 5 of 8 are slightly positive (best: Narrow-CPR +4.8, Supertrend+VWAP +4.4 pts per trade). None is significant: the smallest BH q is 0.67.
Yes, they do work in some periods and not others, and the main driver is the year itself. The rolling-expectancy and month heatmaps show long on/off stretches. Example: ORB15 on spot was +7.0 in 2024, −5.0 in 2025 and +1.0 in 2026. Only 2–4 of 9 option quarters were positive for any strategy. Much of 2026's improvement comes from October 2026 alone (13 sessions): every strategy was positive that month (+1.7 to +53 pts per trade). For Jan–Sep 2026 only P4 (+1.8), P5 (+3.7) and P7 (+0.4) were positive, and P1 was −5.5. One good month is a period effect, not an edge.
Can you tell the good periods in advance? Mostly no. We screened 8 ex-ante regime labels (23 buckets: VIX, 20/50-DMA trend, efficiency ratio, prior-day range, NR7, gap size, expiry position, FII index-futures long %), 713 bucket results in all. Each strategy's “good regime” was chosen on older data by a fixed rule and applied unchanged to the next period, giving 32 out-of-sample tests. 0 of 32 survive BH (best q=0.30). The “recent-performance switch” (trade only if the last 20 trades were net positive) was no better than always-on in 11 of 16 cases.
One lead worth watching: breakouts after an NR7 day (yesterday's range was the narrowest of the last 7). It was picked on train, independently, for P2, P3 and P4, and in 2026 it gave +23 to +28 pts per trade (n=15–29, raw p 0.025–0.10). But it is the same ~20–30 days for all three. On spot it helped in 4 of 5 years and reversed in 2024 (−10.6 vs +8.5). It goes on the forward-watch list, not the trade list.
Combining strategies does not raise the odds (Combinations section). The 8 are not independent: when ORB15 fires, ORB5 fires the same way on 89% of those days, and the gap fade is the ORB15 trade 90% of the time. “≥2 agree” and “≥3 agree” did not beat random (2026: −1.3 and +1.3 pts per trade; train −6.6 and −11.0). The best train pair, PDH/PDL breakout + narrow CPR, made +19.8 per trade in 2026 (n=25, p=0.05), but q=0.13 and it falls to +2.8 without its best 3 trades. 0 of 11 combos count.
Premium-support (“black box”) entries on the option chart do not work either (Premium support section). 4 pre-registered variants × 2 exits: all 8 lost money on train and all 8 lost in 2026 (−1.8 to −15.1 pts per trade), most of them worse than random entries at the same minute. They are negative even before charges, so the earlier harsh-cost result was not a cost artefact.
The trader's discretion could not be reverse-engineered (Trader discretion section). All 218 usable log entries were replayed on real 1m bars. The logged 78% win rate becomes 27% on the real bars, and the Black Box calls (14 of 15 logged wins) become 1 of 15. A tree and a logistic on 17 entry-time features did no better than chance on held-out months (AUC 0.46 and 0.55). The pre-registered filter that came out (enter before 09:52 IST) failed the held-out test (lift 0.41) and 2024–26 history (0 of 4 count). The only robust pattern is that he trades early in the day.
The trader's live YouTube calls, scored from the streams, are not an edge either (YouTube live calls section). 58 fresh calls in 31 sessions; 29 scorable on real 1m bars: +3.1 net pts per trade, win 41%, p=0.31, -4.7 without the best 3. His self-reported log matches only grp mode n ... peak_min_med sig_time_med tv_med 0 NIFTY_raw counter 13.0 ... 15.0 300.0 0.800 1 NIFTY_raw neutral 14.0 ... 12.0 304.5 0.475 2 NIFTY_raw with 14.0 ... 3.0 322.0 0.370 3 SENSEX_raw neutral 1.0 ... 41.0 327.0 0.560 4 SENSEX_raw with 1.0 ... 3.0 339.0 -0.720 5 SENSEX_scaled counter 11.0 ... 7.0 323.0 2.990 6 SENSEX_scaled neutral 22.0 ... 4.0 329.0 2.905 7 SENSEX_scaled with 24.0 ... 11.5 312.5 0.390 [8 rows x 10 columns] of 34 stream calls in the same window.
Expiry-day cheap (~20) counter-trend options do jump 2×–3× fairly often, but buying them loses (Expiry cheap counter-trend section). Nifty 40% / Sensex 38% of signals later printed ≥2× and 21% / 19% ≥3×. A random minute or the with-trend option does about as well, and 83% / 95% settle at ≈0. All 6 pre-registered buyer tests lose in 2026; 0 count.
The ~20 premium level is not special (Why 20? section). Across levels 5–100, the chance of 2×/3×, of expiring ≈0, and the expectancy all change smoothly; 20 sits on the curve (0 of 18 outlier checks). It does not hold better than 17/18/22/23. “Holds at 20 then breaks out” was +9.3 on Nifty in 2026 but −8.5 on train, p=0.29: noise. 0 of 12 count.
“20 without time value” (ITM by ~20, delta ≈1) has no edge. It is rare: 41 Nifty events in 109 expiries, mostly after 14:30. It expires near zero far less often than the OTM-20 (34% vs 76%), with the same chance of doubling (39% vs 40%), but it is just a late spot bet that also loses before costs. 0 of 18 rows count.

Summary table · Works in which regime? · By period · Rules · Method and limits · Combinations · Premium support · Trader discretion · YouTube live calls · Expiry cheap counter-trend · Why 20? · 20 without time value

1. Summary (primary exit: structural stop, trail 50% of peak after 2R, time exit 15:14; REAL cost)

Strategy	Train Oct-24–Dec-25 n / net pts per trade	Test 2026 n / net pts	2026 win%	2026 ₹ total (1 lot)	Pctile vs random entries	p (2026)	BH q (8)	Spot net pts per trade 2022 / 23 / 24 / 25 / 26	Positive quarters spot · options	Works in which regime?
P1 ORB15 breakout	277 / -8.2	181 / -3.2	33%	-37,935	54	0.77	0.77	+3.6 / -6.5 / +7.0 / -5.0 / +1.0	10/20 · 2/9	No stable regime. No filter qualified on the 2024–25 options train. On spot, the walk-forward picked a different label each year (expiry → prior-day range), and the OOS edge was not significant.
P2 ORB5 first 5-min candle	281 / -4.8	189 / -0.0	34%	-508	69	0.50	0.67	+2.5 / -5.4 / +1.7 / -1.9 / +7.0	8/20 · 2/9	Lead: after an NR7 day. Chosen on train (+2.1, n=42). 2026: +28.3 pts (n=29, p=0.025, beats 99% of random subsets), but BH q=0.35. Not proven. On spot, the walk-forward kept picking expiry days, which flipped sign year to year.
P3 Prior-day high/low breakout	128 / -2.9	74 / +0.9	38%	4,293	69	0.42	0.67	-6.8 / -0.1 / +8.2 / +3.2 / +7.9	11/20 · 3/9	Lead: after an NR7 day (2026 +23.0, n=15, p=0.10, q=0.39). On spot, the filter flipped between FII-long and prior-day-range labels. Not proven.
P4 Narrow-CPR breakout + VWAP	77 / -10.1	51 / +4.8	39%	15,849	89	0.24	0.67	– / – / +34.1 / -11.1 / +22.1	4/9 · 2/9	Lead: NR7 / compressed prior day (options 2026 +23.7, n=18; spot 2026 +33, n=28, p=0.009, q=0.30). Only one OOS year exists (no VWAP data before Oct-2024). Not proven.
P5 Supertrend(10,3) flip + VWAP	165 / -5.6	128 / +4.4	44%	36,828	92	0.17	0.67	– / – / -4.0 / -0.2 / +16.9	7/10 · 4/9	Mid efficiency-ratio days (0.2–0.4) chosen on train. 2026 +11.9 (n=33, p=0.13). Not proven. The meta-switch hurt.
P6 EMA 9/21 cross + VWAP	202 / -4.4	152 / +0.8	38%	8,346	75	0.42	0.67	– / – / +21.4 / -1.6 / +9.0	8/10 · 4/9	No. The train-chosen filter (compressed prior day) went negative in 2026 (−3.2).
P7 RSI(14) 30/70 reversal	222 / -3.0	160 / +0.5	28%	5,558	95	0.45	0.67	-6.4 / -0.9 / -9.2 / -4.0 / +3.1	8/20 · 4/9	No. The train-chosen filter (compressed prior day) was −8.0 in 2026, worse than always-on. On spot, no filter qualified in 2025–26.
P8 Gap fade (gap-fill)	46 / -19.7	39 / -3.1	28%	-7,977	55	0.62	0.71	-4.0 / -16.7 / +16.9 / -36.2 / -2.3	7/20 · 2/9	No. No filter qualified on train. The M20 switch was +26 in 2026, but on only 8 trades.

Percentile vs random: 2026 mean compared with 2000 sets of random-day entries at the same minute, direction and R, with the same exit (50 = no better than random). Spot net = index points minus 1 pt. VWAP-based P4–P6 start Oct-2024 because futures volume is missing before that. Positive quarters: spot 2022–26 · options Q4-2024–Q4-2026.

2. Works in which regime?

How it was tested (pre-declared). Every label is known before 09:15. Rule: on the training window, keep the buckets of one label where the strategy averaged above 0 (with at least 10 option trades or 15 spot trades per bucket). Pick the label with the highest training total. Apply it unchanged to the next period. Options: train Oct-2024–Dec-2025 → test 2026. Spot: expanding walk-forward (train 2022..Y → test Y+1; OOS 2023–2026; the requested 2017 start is impossible because no intraday data before 2022 exists). Meta-switch M20: trade only if the strategy's last 20 completed trades summed above 0. BH over the declared family of 32 tests. A regime edge counts only if q<0.10, the OOS filtered mean beats always-on, and it beats ≥90% of random same-size subsets.

Overfitting disclosure. 8 labels × 23 buckets × 8 strategies × 2 datasets, 713 bucket cells (all shown in regime_buckets.csv). With this many cuts, plenty of buckets look great in hindsight. In the heatmaps below, many cells are green on train and a largely different set is green in 2026. That is why only out-of-sample results count.

2a. The 32 out-of-sample regime tests

test	cand	rule	train_n	train_mean	oos_n	oos_mean	always_mean	p	q_bh	pctile_vs_random_subsets	pctile_vs_random_entries_same_regime	edge_counts?
opt_regime	P1	no filter found on train	–	–	0	–	-3.22	1.00	1.00	–	–	False
opt_regime	P2	L5_NR7 in ['NR7']	42.00	2.11	29	28.34	-0.04	0.03	0.34	99.30	99.70	False
opt_regime	P3	L5_NR7 in ['NR7']	23.00	7.93	15	22.98	0.89	0.10	0.39	98.45	96.85	False
opt_regime	P4	L5_NR7 in ['NR7']	25.00	1.90	18	23.69	4.78	0.05	0.34	97.95	90.75	False
opt_regime	P5	L3_ER20 in ['mid0.2-0.4']	47.00	6.46	33	11.86	4.43	0.13	0.43	81.85	85.40	False
opt_regime	P6	L4_PDR in ['compressed<0.8']	75.00	0.98	50	-3.16	0.84	0.69	0.87	24.05	27.05	False
opt_regime	P7	L4_PDR in ['compressed<0.8']	82.00	3.11	55	-7.99	0.53	0.93	1.00	5.55	9.05	False
opt_regime	P8	no filter found on train	–	–	0	–	-3.15	1.00	1.00	–	–	False
spot_walkforward	P1	walk-forward (see wf table)	–	–	428	1.47	-0.97	0.36	0.86	77.55	–	False
spot_walkforward	P2	walk-forward (see wf table)	–	–	298	0.65	-0.00	0.45	0.86	58.00	–	False
spot_walkforward	P3	walk-forward (see wf table)	–	–	277	5.49	4.51	0.09	0.39	67.05	–	False
spot_walkforward	P4	walk-forward (see wf table)	–	–	28	33.02	2.90	0.01	0.30	98.40	–	False
spot_walkforward	P5	walk-forward (see wf table)	–	–	33	30.25	7.72	0.05	0.34	93.80	–	False
spot_walkforward	P6	walk-forward (see wf table)	–	–	112	15.25	3.22	0.04	0.34	97.70	–	False
spot_walkforward	P7	walk-forward (see wf table)	–	–	142	-4.08	-2.84	0.75	0.87	41.30	–	False
spot_walkforward	P8	walk-forward (see wf table)	–	–	88	-3.60	-7.03	0.64	0.86	67.25	–	False
opt_M20	P1	M20	–	–	50	-0.28	-3.22	0.51	0.86	66.25	–	False
spot_M20	P1	M20	–	–	366	-1.38	-0.97	0.62	0.86	44.15	–	False
opt_M20	P2	M20	–	–	74	-5.62	-0.04	0.76	0.87	16.35	–	False
spot_M20	P2	M20	–	–	418	-7.34	-0.00	0.95	1.00	1.20	–	False
opt_M20	P3	M20	–	–	25	0.52	0.89	0.47	0.86	49.25	–	False
spot_M20	P3	M20	–	–	221	0.90	4.51	0.42	0.86	11.00	–	False
opt_M20	P4	M20	–	–	28	-0.64	4.78	0.54	0.86	19.00	–	False
spot_M20	P4	M20	–	–	41	15.32	5.06	0.08	0.39	81.60	–	False
opt_M20	P5	M20	–	–	81	-1.62	4.43	0.64	0.86	4.00	–	False
spot_M20	P5	M20	–	–	166	6.49	6.75	0.15	0.43	47.40	–	False
opt_M20	P6	M20	–	–	68	-4.60	0.84	0.74	0.87	11.10	–	False
spot_M20	P6	M20	–	–	196	0.88	4.70	0.44	0.86	15.65	–	False
opt_M20	P7	M20	–	–	84	-0.99	0.53	0.57	0.86	34.45	–	False
spot_M20	P7	M20	–	–	316	-1.26	-2.84	0.65	0.86	74.50	–	False
opt_M20	P8	M20	–	–	8	26.39	-3.15	0.20	0.55	91.00	–	False
spot_M20	P8	M20	–	–	58	17.86	-7.03	0.12	0.41	99.35	–	False

2b. Spot walk-forward: what was chosen each year and what happened next year

cand	test_year	rule	train_n	train_mean	oos_n	oos_mean	always_mean
P1	2023	L7_Expiry in ['day_before', 'expiry_day']	102	13.01	93	-1.87	-6.55
P1	2024	L7_Expiry in ['day_before', 'expiry_day']	195	5.91	99	4.94	6.95
P1	2025	L4_PDR in ['compressed<0.8', 'normal']	503	4.78	174	-2.05	-5.03
P1	2026	L4_PDR in ['compressed<0.8']	344	6.21	62	10.80	0.96
P2	2023	L7_Expiry in ['expiry_day']	50	24.91	51	-5.92	-5.44
P2	2024	L7_Expiry in ['day_before', 'expiry_day']	199	5.80	101	6.24	1.74
P2	2025	L7_Expiry in ['day_before', 'expiry_day']	300	5.95	105	-5.70	-1.93
P2	2026	L7_Expiry in ['expiry_day']	204	6.45	41	11.34	7.00
P3	2023	L8_FII_long in ['25-45%', '<25%']	36	1.15	48	3.36	-0.09
P3	2024	L4_PDR in ['compressed<0.8']	60	4.53	52	10.21	8.17
P3	2025	L8_FII_long in ['25-45%', '<25%']	145	9.96	103	2.37	3.20
P3	2026	L8_FII_long in ['25-45%', '<25%']	248	6.80	74	7.90	7.90
P4	2025	none (trade nothing)	9	–	0	–	-11.09
P4	2026	L4_PDR in ['compressed<0.8']	53	2.29	28	33.02	22.09
P5	2025	none (trade nothing)	25	–	0	–	-0.20
P5	2026	L3_ER20 in ['mid0.2-0.4']	49	14.59	33	30.25	16.86
P6	2025	none (trade nothing)	30	–	0	–	-1.61
P6	2026	L2_DMA_trend in ['down', 'mixed']	117	9.53	112	15.25	9.02
P7	2023	L5_NR7 in ['NR7']	36	3.54	33	-6.47	-0.91
P7	2024	L1_VIX in ['13-16']	72	5.08	109	-3.36	-9.22
P7	2025	none (trade nothing)	558	–	0	–	-3.98
P7	2026	none (trade nothing)	746	–	0	–	3.09
P8	2023	L4_PDR in ['compressed<0.8', 'normal']	55	10.02	29	-17.44	-16.67
P8	2024	L7_Expiry in ['day_before', 'expiry_day']	51	26.00	25	22.65	16.85
P8	2025	L7_Expiry in ['day_before', 'expiry_day']	76	24.90	18	-44.60	-36.18
P8	2026	L7_Expiry in ['day_before', 'expiry_day']	94	11.59	16	26.57	-2.26

The chosen label changes almost every year for P1, P2, P3, P7 and P8. A regime that has to be re-picked every year is not a stable property of the strategy.

2c. Recent-performance switch vs always-on (OOS mean net pts per trade)

data	cand	8w	M20	M40	always_mean
options_2026	P1	6.63	-0.28	2.90	-3.22
options_2026	P2	-2.85	-5.62	-6.78	-0.04
options_2026	P3	0.98	0.52	-12.10	0.89
options_2026	P4	18.92	-0.64	-4.93	4.78
options_2026	P5	-2.56	-1.62	-1.16	4.43
options_2026	P6	1.71	-4.60	-1.80	0.84
options_2026	P7	-3.21	-0.99	-0.72	0.53
options_2026	P8	-2.01	26.39	–	-3.15
spot_2023_26	P1	0.90	-1.38	0.16	-0.97
spot_2023_26	P2	-5.45	-7.34	-6.39	-0.00
spot_2023_26	P3	4.34	0.90	-0.85	4.51
spot_2023_26	P4	2.24	15.32	13.37	5.06
spot_2023_26	P5	3.89	6.49	13.51	6.75
spot_2023_26	P6	4.39	0.88	0.27	4.70
spot_2023_26	P7	-0.65	-1.26	-3.98	-2.84
spot_2023_26	P8	9.41	17.86	5.11	-7.03

M20 / M40 = last 20 / 40 trades net positive. 8w = last 8 weeks net positive. Only M20 is in the family; M40 and 8w are secondary. Turning a strategy off after losses mostly misses the rebounds: these strategies mean-revert in performance rather than trend.

2d. Lead: NR7 prior day for the breakout strategies (P1–P3 pooled; post-hoc check, not in the family)

year	size_NR7	size_no	mean_NR7	mean_no
2022	91	460	13.35	-0.52
2023	90	474	-2.02	-5.46
2024	106	465	-10.57	8.52
2025	82	497	8.36	-4.01
2026	72	374	49.86	-4.01

year	size_NR7	size_no	mean_NR7	mean_no
2024	23	86	2.76	-14.51
2025	82	495	1.47	-5.95
2026	72	372	22.80	-5.83

Spot net pts per trade by year. NR7 beat non-NR7 days in 2022, 2023, 2025 and 2026, and lost in 2024. The 2026 gap (+50 vs −4) is far larger than in any earlier year, so expect it to shrink. To watch going forward: log every NR7-day ORB/PDH breakout from 27-Oct-2026 and re-check after about 40 trades.

2e. Mean net pts by regime bucket (options; top = train, bottom = 2026)

Spot (top = 2022–25, bottom = 2026)

Full options bucket table

label	bucket	P1_test	P1_train	P2_test	P2_train	P3_test	P3_train	P4_test	P4_train	P5_test	P5_train	P6_test	P6_train	P7_test	P7_train	P8_test	P8_train
L1_VIX	13-16	-9.1	-14.6	-6.9	-7.4	0.2	-4.6	12.1	-7.0	4.0	-7.7	-4.2	-5.4	0.5	-1.0	-12.8	-20.8
L1_VIX	<13	3.1	-3.1	5.0	-3.6	1.7	-2.6	0.8	-8.7	6.2	-0.9	0.5	-3.1	-5.6	-4.5	24.3	-14.9
L1_VIX	>16	-5.3	-10.3	0.2	0.0	-2.2	2.2	2.0	-19.3	-0.3	-13.9	4.4	-6.2	8.8	-4.0	-7.9	-26.2
L2_DMA_trend	down	3.5	-14.9	3.2	-6.4	9.8	-7.0	19.3	-18.2	8.6	-4.8	6.1	-1.5	1.2	-5.4	-4.5	-19.1
L2_DMA_trend	mixed	-10.8	-0.8	-3.3	-3.2	-8.1	2.3	-0.8	4.4	2.8	-12.7	-0.8	-4.4	10.0	-3.1	6.5	-2.5
L2_DMA_trend	up	-9.6	-6.1	-3.5	-4.3	-9.0	-1.6	-3.2	-10.1	-2.2	-3.5	-7.4	-6.6	-8.6	-1.1	-9.5	-31.4
L3_ER20	mid0.2-0.4	-0.6	-2.5	2.9	0.8	2.9	-7.9	12.7	-10.9	11.9	6.5	5.5	-0.8	5.3	0.0	1.8	-35.4
L3_ER20	range<0.2	-4.9	-9.5	0.8	-7.5	-1.1	0.1	8.5	-9.8	2.7	-9.9	-0.2	-4.8	0.0	-2.8	-13.8	-9.9
L3_ER20	trend>0.4	-2.2	-17.4	-6.0	-6.8	5.5	-3.9	-19.3	-9.1	-0.5	-12.6	-2.6	-11.5	-3.6	-13.6	2.9	-23.1
L4_PDR	compressed<0.8	1.7	-3.6	9.3	-0.6	-4.5	2.0	9.1	-2.9	0.1	0.4	-3.2	1.0	-8.0	3.1	1.1	-14.1
L4_PDR	expanded>1.2	-8.0	-13.9	-7.9	-9.3	-2.9	-13.7	-31.0	-44.6	10.6	-11.6	8.4	-13.9	4.2	-7.8	16.4	-21.2
L4_PDR	normal	-3.9	-8.4	-2.1	-5.7	9.7	-4.0	10.2	-21.4	1.7	-6.3	-3.7	-1.8	5.7	-5.9	-27.1	-22.6
L5_NR7	NR7	17.0	-2.2	28.3	2.1	23.0	7.9	23.7	1.9	14.5	-4.7	-8.3	-3.2	-16.2	-9.4	26.5	-11.2
L5_NR7	no	-6.9	-9.3	-5.2	-6.0	-4.7	-5.3	-5.5	-15.9	2.8	-5.7	2.3	-4.6	3.9	-1.8	-7.5	-21.8
L6_Gap	flat<0.25%	0.1	-5.7	2.8	-2.0	1.3	-0.3	11.4	-5.2	7.2	-1.9	1.2	0.6	-4.2	-2.6	–	67.3
L6_Gap	large>0.75%	-12.0	-4.9	-15.0	-2.6	–	–	-1.4	-24.4	20.5	-20.6	10.9	-18.3	11.5	-6.0	-6.2	-25.1
L6_Gap	medium	-4.5	-13.4	1.7	-10.4	-0.1	-13.8	-8.6	-15.6	-4.6	-9.3	-4.0	-10.6	0.9	-3.2	-2.2	-21.0
L7_Expiry	day_before	-2.3	-7.5	1.6	-6.6	10.6	2.0	5.7	-12.6	7.9	-6.1	13.3	-4.4	2.3	1.8	28.8	-22.6
L7_Expiry	expiry_day	-0.7	-7.6	2.0	-4.8	0.5	-12.7	8.5	-15.3	6.1	-2.7	-3.3	-2.7	-7.5	-6.2	1.6	-27.1
L7_Expiry	other	-4.5	-8.8	-1.4	-4.2	-2.5	-0.3	3.0	-7.8	2.6	-6.5	-2.4	-5.1	2.6	-3.5	-15.9	-16.3
L8_FII_long	25-45%	31.9	-12.5	5.5	-10.9	15.1	-5.7	31.9	-7.0	-39.6	-10.7	-28.7	-10.1	-23.8	-3.8	–	-30.1
L8_FII_long	<25%	-3.4	-7.8	-0.1	-4.5	0.7	-2.7	4.2	-10.0	4.8	-4.5	1.0	-3.3	0.7	-2.6	-3.2	-16.5
L8_FII_long	>45%	–	18.8	–	43.9	–	31.9	–	-35.6	–	0.7	–	8.2	–	-8.7	–	-39.5

3. By period

Month heatmaps (net pts per trade)

Options by year

cand	2024	2025	2026
P1	-16.4	-6.8	-3.2
P2	-8.0	-4.2	-0.0
P3	-6.1	-2.2	0.9
P4	-10.4	-10.1	4.8
P5	-10.8	-5.0	4.4
P6	0.1	-4.9	0.8
P7	2.9	-4.1	0.5
P8	-13.3	-21.0	-3.2

Options by half-year

cand	2024H2	2025H1	2025H2	2026H1	2026H2
P1	-16.4	-10.3	-3.2	-5.2	-0.0
P2	-8.0	-3.6	-4.9	-0.5	0.6
P3	-6.1	-0.2	-4.0	-1.0	3.5
P4	-10.4	-11.7	-8.7	16.6	-5.0
P5	-10.8	-9.6	-0.3	6.3	1.0
P6	0.1	-7.3	-2.6	1.9	-0.8
P7	2.9	-4.6	-3.8	1.9	-1.4
P8	-13.3	-26.7	-10.2	-15.4	18.7

Options by quarter

cand	2024Q4	2025Q1	2025Q2	2025Q3	2025Q4	2026Q1	2026Q2	2026Q3	2026Q4
P1	-16.4	-11.5	-8.8	-3.8	-2.7	5.1	-15.2	-6.0	26.0
P2	-8.0	-3.7	-3.5	-3.0	-6.8	4.4	-5.4	-6.2	29.8
P3	-6.1	-3.0	4.6	-3.7	-4.2	7.0	-10.2	-6.8	38.9
P4	-10.4	-17.5	-5.8	-4.6	-13.8	29.8	-7.9	-12.0	53.3
P5	-10.8	-16.2	-3.6	0.0	-0.7	5.8	6.7	-2.0	17.0
P6	0.1	-12.5	-2.7	0.3	-6.2	5.0	-0.8	-3.7	14.9
P7	2.9	1.3	-10.7	-5.4	-2.2	-9.6	12.7	-2.2	1.7
P8	-13.3	-22.6	-37.0	-3.5	-21.0	-37.1	-5.2	22.9	13.2

Spot by year

cand	2022	2023	2024	2025	2026
P1	3.6	-6.6	7.0	-5.0	1.0
P2	2.5	-5.4	1.7	-1.9	7.0
P3	-6.8	-0.1	8.2	3.2	7.9
P4	–	–	34.2	-11.1	22.1
P5	–	–	-4.0	-0.2	16.9
P6	–	–	21.4	-1.6	9.0
P7	-6.4	-0.9	-9.2	-4.0	3.1
P8	-4.0	-16.7	16.8	-36.2	-2.3

Spot by half-year

cand	2022H1	2022H2	2023H1	2023H2	2024H1	2024H2	2025H1	2025H2	2026H1	2026H2
P1	-1.3	8.5	-2.6	-10.7	21.6	-6.6	-11.6	1.5	-3.1	7.4
P2	-2.2	7.0	-2.5	-8.5	8.6	-5.0	-2.2	-1.7	5.9	8.7
P3	-2.6	-12.4	0.2	-0.4	23.8	-5.4	8.7	-1.6	2.0	16.1
P4	–	–	–	–	–	34.2	-11.4	-10.8	40.9	6.6
P5	–	–	–	–	–	-4.0	-10.7	10.5	20.0	11.1
P6	–	–	–	–	–	21.4	-6.2	2.8	9.8	7.8
P7	-9.9	-2.7	-2.8	1.1	-18.1	-0.5	-5.9	-2.4	4.7	0.9
P8	-5.0	-2.8	-18.0	-15.2	49.2	-22.8	-44.4	-19.8	-18.5	26.7

Spot by quarter

cand	2022Q1	2022Q2	2022Q3	2022Q4	2023Q1	2023Q2	2023Q3	2023Q4	2024Q1	2024Q2	2024Q3	2024Q4	2025Q1	2025Q2	2025Q3	2025Q4	2026Q1	2026Q2	2026Q3	2026Q4
P1	-3.8	1.0	4.0	13.2	-0.6	-4.9	-12.3	-8.8	29.2	13.3	3.1	-17.0	-8.2	-15.4	1.0	1.9	10.5	-16.2	-3.5	55.1
P2	-4.9	0.4	7.0	7.0	-0.7	-4.4	-12.2	-4.2	21.1	-4.2	-3.0	-7.2	0.1	-4.4	-0.5	-3.0	7.8	4.0	-2.2	56.4
P3	-10.3	8.3	-9.0	-17.2	-3.1	3.8	-15.8	20.6	20.7	27.2	-14.1	1.0	5.9	13.5	-4.3	0.9	12.2	-9.7	-4.3	85.9
P4	–	–	–	–	–	–	–	–	–	–	–	34.2	-13.8	-9.0	-5.9	-16.9	58.4	8.0	-6.9	119.4
P5	–	–	–	–	–	–	–	–	–	–	26.8	-5.3	-11.4	-10.1	10.3	10.7	16.4	23.2	5.3	43.4
P6	–	–	–	–	–	–	–	–	–	–	21.8	21.4	-5.9	-6.5	4.9	0.2	12.7	7.2	2.6	36.8
P7	-6.1	-13.4	-8.7	4.1	0.7	-6.2	3.5	-0.9	-20.1	-16.1	3.4	-4.0	5.7	-18.3	-5.1	0.3	-16.2	24.3	-0.5	6.4
P8	-17.5	7.6	-12.6	9.7	-17.3	-19.3	-14.3	-16.4	84.8	23.0	-22.5	-23.1	-35.5	-68.3	-9.5	-36.4	-60.2	1.2	26.7	26.6

Monthly CSVs: options · spot.

Secondary exit (structural stop, hold to 15:14), options net pts

cand	test	train
P1	-4.36	-8.65
P2	-0.68	-5.29
P3	1.07	-6.21
P4	4.18	-10.56
P5	3.80	-6.35
P6	0.88	-6.77
P7	-0.32	-3.55
P8	-6.28	-20.20

Combinations: does agreement between strategies raise the probability?

Added after Satya's 21:13 IST question: “what about combination of strategies there probability increases right”. Pre-registered at 21:14 IST in PREREG.md before any combo result. Same data, costs, exit, random benchmark and train/test split. A combo trade is the trade of the confirming (later) signal, one per day. Honesty flag: the NR7 combos (N*) reuse a filter whose 2026 result was already seen, so they are not clean out-of-sample.

C1. Are the 8 independent? No.

Common window 21-Oct-2024 to 23-Oct-2026 (496 days). Signal days: P1 463, P2 474, P3 204, P4 128, P5 294, P6 355, P7 387, P8 86. Cell = % of days the row strategy fired on which the column strategy fired in the same direction.

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	100	89	36	21	27	34	21	17
P2	87	100	34	21	24	33	24	16
P3	82	78	100	29	21	24	8	5
P4	75	77	46	100	15	21	16	2
P5	43	39	14	6	100	84	54	19
P6	45	44	14	8	70	100	50	17
P7	25	29	4	5	41	46	100	12
P8	90	86	13	3	64	72	56	100

ORB15 and ORB5 fire the same way on 87–89% of each other's days. When P3 (PDH/PDL) or P4 (CPR) fires, ORB15/ORB5 fire the same way on 75–82% of those days. The indicator group (P5 Supertrend, P6 EMA) agrees 70–84%. P8 (gap fade) is the ORB15 trade 90% of the time. When two of them “agree”, it is usually the same price move counted twice, not two independent confirmations. Option P&L on same-direction co-fire days is also highly correlated (P1–P2 0.91; P1–P8 1.00):

P&L correlation, opposite-direction rate, direction correlation, Jaccard

Correlation of option net P&L on same-direction co-fire days

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	–	0.91	0.52	0.86	0.43	0.73	0.66	1.00
P2	0.91	–	0.58	0.79	0.65	0.72	0.71	0.94
P3	0.52	0.58	–	0.72	0.28	0.40	0.57	0.22
P4	0.86	0.79	0.72	–	0.14	0.42	-0.10	–
P5	0.43	0.65	0.28	0.14	–	0.76	0.60	0.65
P6	0.73	0.72	0.40	0.42	0.76	–	0.57	0.72
P7	0.66	0.71	0.57	-0.10	0.60	0.57	–	0.84
P8	1.00	0.94	0.22	–	0.65	0.72	0.84	–

% of days with an opposite-direction signal

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	0	9	7	5	32	36	57	2
P2	9	0	9	5	35	38	54	3
P3	16	20	0	5	35	41	76	3
P4	18	20	8	0	45	52	66	9
P5	51	56	24	20	0	10	21	2
P6	48	51	23	19	8	0	23	3
P7	68	66	40	22	16	21	0	3
P8	10	14	8	14	8	14	15	0

Direction correlation on days both fire

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	1.00	0.81	0.68	0.61	-0.09	-0.03	-0.46	0.79
P2	0.81	1.00	0.59	0.59	-0.17	-0.08	-0.39	0.71
P3	0.68	0.59	1.00	0.70	-0.28	-0.26	-0.80	0.12
P4	0.61	0.59	0.70	1.00	-0.54	-0.42	-0.62	-0.66
P5	-0.09	-0.17	-0.28	-0.54	1.00	0.80	0.44	0.78
P6	-0.03	-0.08	-0.26	-0.42	0.80	1.00	0.36	0.69
P7	-0.46	-0.39	-0.80	-0.62	0.44	0.36	1.00	0.62
P8	0.79	0.71	0.12	-0.66	0.78	0.69	0.62	1.00

Jaccard same-direction overlap %

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	100	86	36	20	26	32	20	17
P2	86	100	34	20	24	32	23	16
P3	36	34	100	22	11	11	4	4
P4	20	20	22	100	6	7	5	2
P5	26	24	11	6	100	66	34	17
P6	32	32	11	7	66	100	37	17
P7	20	23	4	5	34	37	100	12
P8	17	16	4	2	17	17	12	100

C2. The 11 pre-declared combos (option net pts per trade, REAL costs, trail-50% exit)

Combo	Train n / mean / win	2026 n / mean / win	2026 mean of its single strategies	2026 random mean (pctile)	p	BH q (11)	Jan–Sep 2026	Oct 2026	2026 without best 3 trades	Counts?
K2_15 ≥2 strategies, same direction, within 15 min	262 / -6.6 / 35%	183 / -1.3 / 33%	+0.4	-3.5 (70)	0.621	0.69	-3.7	+29.7	-5.0	no
K3_30 ≥3 strategies, same direction, within 30 min	160 / -11.0 / 31%	111 / +1.3 / 36%	+0.4	-2.4 (79)	0.401	0.63	-2.9	+48.5	-3.3	no
K2_D ≥2 strategies, same direction, same day	287 / -5.6 / 36%	196 / -2.1 / 32%	+0.4	-4.2 (74)	0.694	0.69	-4.3	+24.6	-5.5	no
N2 NR7 day + P2 (contaminated)	42 / +2.1 / 48%	29 / +28.3 / 55%	-0.0	-5.1 (100)	0.025	0.13	+23.8	+155.7	+9.7	no
N3 NR7 day + P3 (contaminated)	23 / +7.9 / 43%	15 / +23.0 / 53%	+0.9	-0.9 (97)	0.097	0.21	+13.5	+155.7	-4.6	no
N4 NR7 day + P4 (contaminated)	25 / +1.9 / 28%	18 / +23.7 / 61%	+4.8	-5.1 (99)	0.047	0.13	+16.3	+149.4	+2.2	no
N234 NR7 day + ≥2 of P2/P3/P4 (contaminated)	29 / +4.6 / 38%	21 / +8.7 / 38%	+1.0	-1.9 (86)	0.258	0.47	+1.4	+155.7	-12.1	no
PAIR_BEST Best pair on train = P3+P4	34 / +0.7 / 38%	25 / +19.8 / 56%	+2.5	-0.8 (96)	0.049	0.13	+14.6	+79.8	+2.8	no
PAIR_P1P5 ORB15 + Supertrend	72 / -12.1 / 31%	53 / +0.3 / 43%	-0.1	-3.5 (72)	0.482	0.66	+0.4	-2.9	-7.3	no
PAIR_P2P5 ORB5 + Supertrend	66 / -8.3 / 36%	50 / -3.0 / 40%	+1.8	-3.1 (52)	0.673	0.69	-3.7	+32.3	-11.3	no
PAIR_P3P4 PDH/PDL + narrow CPR	34 / +0.7 / 38%	25 / +19.8 / 56%	+2.5	-0.8 (96)	0.049	0.13	+14.6	+79.8	+2.8	no

“Counts” needs q<0.10, a 2026 mean above its single strategies' 2026 mean, and ≥90th percentile vs random. PAIR_BEST turned out to be P3+P4, so it is the same test as PAIR_P3P4. BH over the 10 distinct tests gives the same answer (min q 0.16). Train pair ranking: all 28 pairs. Only P3+P4 was above 0 on train (+0.7, n=34). Every other pair lost.

Reading it. Requiring 2 or 3 strategies to agree lowered train expectancy compared with the single strategies (K3_30 −11.0 vs −5.9). By the time 3 signals agree, much of the move is already used up and the stop is wider. In 2026 the k-of-n combos were negative for Jan–Sep and positive only in October. P3+P4 (PDH/PDL breakout on a narrow-CPR day) is the only clean combo with a positive train and 2026 (+0.7 → +19.8, Jan–Sep 2026 +14.6, n=23). But it has 25 trades, q=0.13, a median train trade of −13, and depends on a few big days. It goes on the forward-watch list next to NR7.

C3. By quarter (options)

combo	2024Q4	2025Q1	2025Q2	2025Q3	2025Q4	2026Q1	2026Q2	2026Q3	2026Q4
K2_15	-12.5	-10.1	-7.7	-0.6	-3.5	8.1	-14.4	-4.5	29.7
K2_D	-12.2	-7.0	-5.8	-1.7	-2.9	5.6	-13.0	-5.3	24.6
K3_30	-22.9	-11.0	-15.0	2.1	-13.3	1.5	-11.1	1.8	48.5
N2	0.7	10.2	-1.0	1.6	-1.2	60.3	2.8	17.0	155.7
N234	30.4	-7.7	34.4	-24.6	-7.9	19.0	-25.0	4.3	155.7
N3	22.1	-17.1	35.6	8.6	-21.6	39.1	-13.6	2.1	155.7
N4	59.5	-11.4	35.8	-26.5	-9.2	34.4	2.8	4.1	149.4
PAIR_BEST	-7.8	-19.2	8.5	12.8	-5.0	49.1	2.0	-11.5	79.8
PAIR_P1P5	-23.2	-13.1	-28.6	1.8	-4.6	1.2	-7.9	10.7	-2.9
PAIR_P2P5	-19.3	-10.5	-26.5	9.8	-4.3	-11.3	-11.2	9.3	32.3
PAIR_P3P4	-7.8	-19.2	8.5	12.8	-5.0	49.1	2.0	-11.5	79.8

C4. Spot-level check 2022–2026 (spot net pts, 1 pt cost)

combo	2022	2023	2024	2025	2026
K2_15	-0.2	-6.2	7.3	-4.1	4.0
K2_D	4.2	-5.8	5.1	-1.5	2.0
K3_30	-2.4	-13.2	5.8	-10.2	9.6
N2	14.7	-1.7	-18.0	10.5	62.7
N234	-29.2	9.6	11.6	0.4	22.4
N3	-12.6	5.5	20.2	8.2	46.2
N4	–	–	96.8	-7.2	58.8
PAIR_BEST	–	–	27.6	11.2	48.0
PAIR_P1P5	–	–	-12.7	-9.8	12.9
PAIR_P2P5	–	–	-10.8	-3.6	7.0
PAIR_P3P4	–	–	27.6	11.2	48.0

On spot, k-of-n agreement flips sign from year to year (K2_15: −0.2, −6.2, +7.3, −4.1, +4.0). P3+P4 needs VWAP for P4, so it only starts Oct-2024.

Premium support (“black box” / premium reversal): buy the option when its own premium bounces off support

Added after Satya's 21:24 IST question: “what about option premium support strategy”. Pre-registered at 21:26 IST in PREREG.md before any result. Real 1m ATM weekly option bars 21-Oct-2024 to 23-Oct-2026. The chart is the current ATM strike's own 5m premium chart. Costs ≈0.5 pt per side plus statutory charges (STT 0.15% of sell premium from 1-Apr-2026). Train Oct-2024–Dec-2025, test 2026. Random benchmark = same minute, same CE/PE, ATM on random days of the same split, same exit.

PS1. What the earlier (early-Oct 2026) work found

Study (nifty-setups-backtest)	What it was	Out-of-sample result
atm-daily-sr	Prior-day ATM CE/PE premiums mapped to spot S/R rails (half-scale), bounce/retest entries; index book SL20, 40%@+20, 30%@+40, trail rest	275 trades, −10.3 pts per trade net (−0.3 gross); old futures-style costs ~10 pts per trade
atm-daily-sr-optbuyer	Same levels, booked as option buyer (SL10, 40%@+10, 30%@+20, trail)	476 trades, −1.5 option pts per trade net, −0.5 gross, 47% win
confluence-v1	Retailer trap + AMD fake-break + ATM premium black-box support bounce, all same direction	20 trades, +7.2 gross but −4.2 net (cost ~11 pts per trade), p=0.48
confluence-v1-scaleout	Same entry, SL20 / 40%@+20 / 30%@+40 / trail exit	30 trades, −0.4 gross, −12.2 net
trader-filters-r126	Black box or round-50 + rejection wick (R1), plus R2/R6 filters, option buyer	All 8 parameter sets net negative (≈ −1.5 to −2 pts per trade, n≈1,000–2,000)
trader log (hand-logged stream calls)	11 “Black Box” trades May–Jul 2026	10 of 11 marked WIN. These are discretionary, hand-logged calls with no fixed rule and no losers' log, so they cannot be reproduced or tested.

The confluence idea was the only one with a positive gross result, and that was on 20 trades. Under corrected costs (~2 pts per round trip instead of ~11), +7.2 gross would still be only about +5 net on n=20, which is not testable evidence.

PS2. The 4 pre-registered variants (option net pts per trade, corrected costs)

Variant	Exit	Train n / mean / win	2026 n / mean / win	2026 mean before charges	2026 random mean (pctile)	p	BH q (8)	Jan–Sep 2026	Oct 2026	2026 without best 3	Counts?
V1 Prior-day premium low bounce	old: SL20, 40%@+20, 30%@+40, trail rest	74 / -3.5 / 49%	35 / -10.4 / 26%	-9.2	-3.4 (2)	1.00	1.00	-11.1	-4.8 (n=4)	-14.2	no
V1 Prior-day premium low bounce	structural stop, trail 50% after 2R	74 / -8.8 / 24%	35 / -15.1 / 14%	-14.1	-0.4 (0)	1.00	1.00	-13.6	-26.9 (n=4)	-18.7	no
V2 Intraday premium swing-low bounce ("black box")	old: SL20, 40%@+20, 30%@+40, trail rest	254 / -2.2 / 45%	175 / -4.0 / 43%	-2.7	-2.8 (20)	0.99	1.00	-4.0	-4.0 (n=13)	-4.7	no
V2 Intraday premium swing-low bounce ("black box")	structural stop, trail 50% after 2R	254 / -3.3 / 29%	175 / -2.8 / 29%	-1.8	-0.9 (21)	0.91	1.00	-3.0	-0.5 (n=13)	-5.2	no
V3 Round premium level (×25) bounce	old: SL20, 40%@+20, 30%@+40, trail rest	271 / -1.7 / 47%	189 / -4.3 / 43%	-3.0	-3.2 (21)	1.00	1.00	-4.6	-1.0 (n=15)	-5.0	no
V3 Round premium level (×25) bounce	structural stop, trail 50% after 2R	271 / -2.8 / 30%	189 / -7.0 / 24%	-6.0	-1.4 (1)	1.00	1.00	-8.3	+7.4 (n=15)	-9.6	no
V4 Premium day-high breakout (resistance break)	old: SL20, 40%@+20, 30%@+40, trail rest	223 / -2.2 / 47%	156 / -4.8 / 42%	-3.5	-3.4 (21)	1.00	1.00	-5.2	-0.7 (n=15)	-5.5	no
V4 Premium day-high breakout (resistance break)	structural stop, trail 50% after 2R	223 / -3.8 / 31%	156 / -1.8 / 28%	-0.7	-3.0 (71)	0.73	1.00	-1.0	-8.6 (n=15)	-5.3	no

V1 level = prior-day low of the same contract. V2 = most recent confirmed 5m swing low after 09:30. V3 = multiple of 25 (premium 40–400). V1–V3 need a touch within max(2 pts, 3%) approached from above, then the next 5m close above the touch-bar high. V4 = 5m close above the premium's high of the day so far. Entry is at the next 1m option open. First event per day. “Counts” needs q<0.10, ≥90th percentile vs random, and still positive without the best 3 trades. Pre-charge means include the 0.5 pt per side slippage.

Reading it. Every variant and both exits lost money in both periods, and no BH q is below 1.00. Several are worse than random entries at the same minute (V1, and V3 with the trail exit, sit at the 0–2nd percentile). A premium bouncing off its support does not predict the next move any better than chance. The prior-day premium low is the worst level, probably because overnight theta decay pulls the premium down to it mechanically rather than real buyers defending it. Satya's old scale-out exit and the trail-50% exit give the same answer. October 2026 does not rescue it either (Oct 2026: V3 trail +7.4 on 15 trades; the others are negative).

PS3. Regime filters chosen on train, applied unchanged to 2026 (BH over 8)

var	ex	label	buckets	train_n	train_mean	test_n	test_mean	always_on_test	random_same_bucket_mean	subset_pctile	t_p	q_bh	counts
V1	X_OLD	none	–	0	–	0	–	-10.37	–	–	1.00	1.00	no
V1	X_T50	none	–	0	–	0	–	-15.13	–	–	1.00	1.00	no
V2	X_OLD	L1_VIX	>16	43	2.87	52	-8.97	-4.00	-3.33	1.55	1.00	1.00	no
V2	X_T50	L3_ER20	mid0.2-0.4	80	1.82	44	-6.92	-2.84	-1.18	14.15	0.97	1.00	no
V3	X_OLD	L5_NR7	NR7	40	6.72	27	-7.19	-4.33	-4.40	20.55	0.97	1.00	no
V3	X_T50	L4_PDR	normal	98	0.74	67	-4.81	-7.04	1.07	77.30	0.87	1.00	no
V4	X_OLD	L2_DMA_trend	up	94	1.23	41	-5.53	-4.75	-4.14	40.20	0.95	1.00	no
V4	X_T50	L2_DMA_trend	mixed	46	0.01	33	-2.73	-1.76	-2.47	45.95	0.69	1.00	no

The train-chosen regime (VIX>16, NR7, uptrend, …) lost money in 2026 in all 6 cases where a filter qualified, and was worse than always-on in 5 of those 6 (V1 had no qualifying filter). 0 of 8 count.

PS4. By quarter (trail-50% exit)

var	2024Q4	2025Q1	2025Q2	2025Q3	2025Q4	2026Q1	2026Q2	2026Q3	2026Q4
V1	-24.3	-7.3	-12.1	-0.7	-2.3	-23.4	-15.6	-4.4	-26.9
V2	-6.2	2.7	-4.9	-3.5	-5.1	-4.2	-0.0	-4.6	-0.5
V3	-2.4	-6.2	-0.0	-6.5	1.3	-6.9	-10.3	-7.6	7.4
V4	-8.6	1.2	-7.1	-4.4	-1.5	3.5	-3.0	-3.3	-8.6

By quarter (old scale-out exit)

var	2024Q4	2025Q1	2025Q2	2025Q3	2025Q4	2026Q1	2026Q2	2026Q3	2026Q4
V1	-1.6	-1.9	-3.6	-6.3	-3.8	-17.3	-11.7	-5.9	-4.8
V2	-4.0	1.0	-1.5	-6.6	-0.4	-4.9	-7.3	-0.1	-4.0
V3	-4.9	-0.6	-1.3	-6.8	3.7	-1.7	-6.2	-5.8	-1.0
V4	-7.2	1.2	-3.5	-4.1	1.3	-5.7	-4.0	-5.8	-0.7

Monthly and regime-bucket tables: prem_by_period.csv · prem_regime_buckets.csv · trades prem_trades.csv.

Trader discretion: can the log's choices be turned into rules?

Added after Satya's 21:34 IST question: “why can't you find those discretion rules”. Design pre-registered at 21:37 IST in PREREG.md (TRADER-DISCRETION) before any taken/skipped or outcome number was seen. Real 1m Nifty ATM weekly option bars. Corrected costs (≈0.5 pt per side plus statutory charges, STT 0.15% of sell premium from 1-Apr-2026).

TD1. The data

The log is trader-log/trades_timestamped.csv: 225 rows, 06-May to 05-Oct-2026, built from stream screenshots and the fixed-date Excel files with entry premium and start/stop times. 218 rows are usable (one NO_TRADE row and rows with no time or strike removed): 110 Nifty and 108 Sensex. Sensex rows are mapped to the Nifty ATM equivalent at the same minute. Logged result: 169 WIN, 47 LOSS, 2 BE (78% wins).

How well the log matches real option bars (earlier fill-match work): only 22 of 218 (10%) match the real premium at the logged time (tiers A/B/B2/C: exact, within ±5 or ±15 minutes, or within 5% of the premium). 148 (68%) match only on strike and time, because the log has no usable premium. 47 (22%) disagree with the real bars (wrong premium, or the strike did not trade). When the earlier loose premium matcher was tried at shifted placebo times, it matched about as often as at the logged times (Nifty 29% vs 31% placebo). So the logged premiums do not reliably identify a real fill.

TD2. Entries by setup family, and what real bars say happened

family	entries	log_win_pct	replay_win_pct	xold_nifty_atm_pts	covered_by_any_rule_pct	covered_by_V_pct
BREAK	57	82	25	-1	33	21
CONT	41	56	40	1	29	17
SUPPORT	40	65	23	-1	35	25
PM3	24	88	45	1	8	8
SWEEP	17	88	18	-6	41	24
BLACKBOX	15	93	7	-6	33	33
OTHER	12	92	20	-2	25	8
BTST	9	100	12	-12	22	0
NEWS	3	100	33	6	67	33

Replay win = the earlier fill-match simulation with the logged stop and target on real 1m bars. X_OLD = Nifty ATM option bought at the next 1m open after the logged time, with the SL20 / 40%@+20 / 30%@+40 / trail exit. Covered = an entry of ours (V1–V4 premium-support or P1–P8) fired the same day, same direction, within ±10 minutes.

What this shows. On real bars the trader's calls are ordinary. The overall replay win rate is 27% versus 78% logged, and the Black Box calls go from 14 of 15 logged wins to 1 of 15 in the replay. Some of the gap is how the replay works: it uses the Nifty ATM and the next 1m open, while he may have used a different strike, a better fill or a manual exit. That gap can't be closed, because 68% of rows have no premium to check against. But a 78% → 27% drop is far too big to be explained by fills alone.

TD3. Which of our rule entries did he take?

Across the 53 days with log entries our rules made 662 candidate entries; 103 (15.6%) line up with one of his entries. Only 30% of his entries are matched by any rule of ours, and 19% by a premium-support rule. Placebo check: shift his entry times by ±30 or ±60 minutes and the match rate is 8.8–13.1%, or 11% if the direction is flipped. So his entries line up with our setups only slightly more often than chance. Most of what he does is not one of these setups.

src	candidates	taken	take_rate_pct	xold_mean	xt50_mean
P1	47	8	17.0	-9.4	-10.4
P2	49	14	28.6	-5.5	-9.3
P3	14	5	35.7	-5.0	-12.5
P4	10	3	30.0	-8.1	-24.4
P5	41	7	17.1	-2.4	1.5
P6	48	7	14.6	-0.7	-4.3
P7	42	2	4.8	0.7	8.9
P8	15	4	26.7	0.4	9.4
V1	14	5	35.7	-0.4	-9.0
V2	124	21	16.9	-5.0	0.8
V3	170	18	10.6	-5.6	-4.4
V4	88	9	10.2	-5.9	-6.4

Outcome of candidates he took vs skipped (our exits, real bars)

taken	n	xold_mean	xold_win_pct	xt50_mean	xt50_win_pct
skipped	559	-5.06	42	-3.05	28
taken	103	-2.12	46	-6.04	29

The taken candidates did no better than the skipped ones. They were better with the old exit (−2.1 vs −5.1) and worse with the trail-50% exit (−6.0 vs −3.0). Both groups lose money. Skipping those candidates did not avoid losers.

TD4. Taken vs skipped: entry-time features (effect size = standardised mean difference; Mann-Whitney p, BH over 17)

feature	taken_mean	skipped_mean	smd	mw_p	n_taken	n_skipped	q_bh
tod	100.796	137.596	-0.407	0.000	103	559	0.001
oi_chg15	13.190	6.956	0.299	0.001	93	546	0.012
prior_trades	2.282	1.832	0.286	0.004	103	559	0.022
vix	15.373	14.907	0.185	0.082	103	559	0.348
basis_chg15	1.698	1.325	0.010	0.228	93	546	0.589
prem_chg5	8.092	7.828	0.034	0.257	103	559	0.589
dist_round50	12.275	13.162	-0.120	0.258	103	559	0.589
pcr_chg15	-0.005	-0.012	0.032	0.304	93	546	0.589
prem_dd	13.202	15.073	-0.137	0.312	103	559	0.589
dist_round100	26.583	25.282	0.092	0.397	103	559	0.675
basis	48.162	48.878	-0.016	0.469	103	559	0.725
spot_mv15	27.906	25.207	0.098	0.548	93	546	0.748
gap	-0.104	-0.140	0.070	0.572	103	559	0.748
spot_pos	0.733	0.721	0.044	0.726	103	559	0.881
expiry_day	0.184	0.193	-0.022	0.837	103	559	0.903
prem_chg15	11.045	10.637	0.029	0.850	93	546	0.903
dte	4.699	4.628	0.034	0.919	103	559	0.919

Only three features differ after BH. Time of day: taken entries come ~37 minutes earlier on average (SMD −0.41). Prior trades that day: more trades already that day (+0.29). ATM OI change over 15 minutes: OI rising faster (+0.30). Premium drawdown, speed of fall, bounce speed, distance from the day high/low and from round levels, gap, basis, PCR change and expiry day show no difference (|SMD| ≤ 0.14). “Prior trades” is a property of his day, not of the market, so it cannot become a rule.

TD5. Models with leave-one-day-out CV (fit May–Jul 2026, holdout Aug–Oct 2026)

Model	Leave-one-day-out AUC (May–Jul)	Holdout AUC (Aug–Oct)
Depth-2 tree (min leaf 10)	0.49	0.46
L2 logistic, 17 features	0.61	0.55

An AUC of 0.5 means the model does no better than a coin flip at telling taken from skipped. The logistic's 0.61 in CV comes almost entirely from time of day and prior trades (coefficients −1.2 and +0.8). It drops to 0.55 on the held-out months. This repeats the early-Oct remine (trader-filter-mine-timed), where the holdout AUC was ≈0.5 and the best broad rule (near the day extreme and a calm 3-minute move) had lift 1.24 with a CI of 0.84–1.61.

TD6. The pre-registered filter and its tests

Extraction rule (fixed in advance): take the tree leaf with the highest take rate among leaves with ≥10 takes, provided it is at least 1.2× the base rate. Result: tod <= 36.50 AND basis_chg15 > -8.25. tod is minutes after 09:15, so 36.5 means before about 09:52 IST. On May–Jul this leaf took 34 of 65 candidates (52%) against a base rate of 16%.

Held-out log months (Aug–Oct 2026): the filter fired 17 times and he took 1. Lift 0.41 (base 14%), day-block bootstrap 90% CI 0.00–1.35. Failed: the CI needed to be above 1.

As a strategy on 2024–26 history (May–Jul 2026 excluded; first filtered candidate per day; random = same candidate minute and type on random days, 30 repeats; BH over 4):

period	exit	n	mean	win	total	p	rand_mean	rand_pctile	drop_top3	q_bh	counted
Oct24-Dec25	xold	229	-2.91	46.29	-665.79	0.98	-2.53	53.33	-3.42	0.98	False
Oct24-Dec25	xt50	229	-3.76	32.75	-861.98	0.91	-3.72	56.67	-6.32	0.98	False
2026 (Jan-Apr+Aug-Oct)	xold	104	-2.30	48.08	-238.74	0.87	-2.26	46.67	-3.42	0.98	False
2026 (Jan-Apr+Aug-Oct)	xt50	104	4.54	36.54	472.48	0.21	-4.07	96.67	-1.58	0.82	False

0 of 4 count. Three of the four are negative and no better than random. The 2026 trail-50% row is +4.5 per trade at the 97th percentile vs random, but q=0.82 and −1.6 without its best 3 trades, the same October-heavy pattern as the rest of this page.

TD7. Why the discretion rules can't be found (bias and small-n disclosure)

Selection bias. The log is self-reported from stream calls. Losers may be missing or relabelled. The real-bar replay (27% wins against 78% logged) points that way. We can't learn rules that separate winners from losers when the losers are not reliably in the log.
Fills can't be checked. Only 10% of rows match a real premium. For the rest we don't know the strike, fill or exit he actually used.
Small n. 103 taken candidates on 53 log days. With 17 features a depth-2 tree is the most the data supports, and even that overfits (CV AUC 0.49).
Most of his trades aren't our setups. 70% of his entries match no rule we have. Whatever he sees (tape, order flow, the chat, news) is not in the 1m OHLC/OI/VIX/basis data we have.
What does hold: he trades early (09:16–10:00 is over-represented 1.8–2.9× in both studies). As a strategy, early entries alone do not make money on real bars.

Files: disc_trader_entries.csv · disc_candidates.csv · disc_feature_contrast.csv · disc_model.json · disc_filter_fullhist.csv. Scripts: s9_discretion.py, s9b_fullhist.py, s10_disc_model.py (sklearn), s11_disc_fullhist.py.

YouTube live calls: a verifiable, pre-outcome log tested on real bars

Added after Satya's 23:15 IST (26-Oct) request. Source: the trader's public live streams at youtube.com/@KarolSEBIRegisteredIITian/streams. We read stream metadata and Hindi auto-captions with yt-dlp, with no login, cookies or credentials. 360p two-second clips were used only to read the strike label on screen, then deleted. No video is embedded or redistributed; the links below go to the original streams. Design pre-registered at 23:40 IST in PREREG.md (YOUTUBE-CALLS), before any call was scored. Research only, not advice.

Verdict. We pulled the calls straight from the stream, so the log is written at the moment of each call, before the outcome. On that log, 29 of 58 fresh calls in 31 sessions (25-Aug to 23-Oct-2026) could be scored on real 1m option bars. They made +3.1 net pts per trade (median -13.7, win 41%, p=0.31). That is percentile 91 against random entries at the same minute, and drops to -4.7 without the best 3 trades. By the pre-registered rule this does not count as an edge. Of the scored calls he said "target done" on 12; real bars hit the target on 6. His self-reported log does not line up with the streams. Of its 34 rows in this window, 1 matches a stream call within 10 minutes, and 5 rows (+620 pts) are on 25-Sep, a day with no stream. The taken-vs-skipped analysis found no filter that passes the pre-registered extraction rule, so nothing could be tested on 2024–26 history.

YT1. Inventory

105 past live streams, 2026-05-06T08:57 to 2026-10-23T09:03 IST (554 hours). All have Hindi auto-captions. Over the 118 trading days from 06-May to 23-Oct-2026: 74 days have full market-hours coverage, 24 partial (<80% of 09:15–15:30), and 20 have no stream (06-30, 07-06, 07-15, 07-17, 07-20, 07-21, 07-22, 07-23, 07-24, 08-13, 08-24, 09-02, 09-04, 09-18, 09-22, 09-24, 09-25, 09-28, 09-30, 10-12). Some may have been deleted or never streamed; we cannot tell which. Files: yt_stream_inventory.csv · yt_trading_day_coverage.csv.

YT2. Pilot (22-Oct, 21-Oct and 19-Oct streams) and feasibility

Timing is reliable. Stream start + caption offset matched the on-screen IST clock within −9 to +3 s on all 4 checked frames.
Strike is readable from the hand-written chart labels (e.g. “73000CE / 73900PE”) by OCR in 50 of 58 call frames. Expiry is never stated; we assume the nearest weekly.
Entry premium is almost never spoken. Calls are like “entry here”, “above this candle”, “SL N points, target 50”. So entry = next 1m open after the call; SL/target come from his spoken points where given.
Extraction accuracy. On 22-Oct we hand-read 12 calls. The pipeline (caption rules + analyst review) found 8 of them (recall about 67%). A raw caption-rule hit is a real fresh call only about 27% of the time (random sample of 30). The rest are retrospective (“we took this, target done”), promos (Telegram/VIP, broker, contests), or conditional plans that never say whether they triggered. That is why an analyst review step was added; it is a deviation from the purely mechanical pre-registered extractor, and it was done without looking at bar outcomes.
Pilot scorability: 0 of 11. The 21/22-Oct calls are Sensex, and the box has no Sensex option bars after 05-Oct (backfill needs an Upstox token, which we did not use). The two 19-Oct Nifty calls lacked a readable PE strike or a stated side.

YT3. All 33 streams (25-Aug..23-Oct): what was extracted

289 caption-rule candidates → 110 dropped as promo/gold/crypto talk → 179 reviewed → 58 fresh calls. The other 121 were retrospective, commentary, vague, or conditional plans with no activation. Of the 58: 29 scored, 13 not scorable (Sensex after 05-Oct, no real bars), 9 strike unreadable, 7 side not stated. That is about 1.9 extractable calls per session, against the ~10 per day in his self-reported log.

Stream time (IST, links to the original stream at that second)	Contract	SL / target pts (*=fallback)	Entry bar @ price	Exit	Net pts	His stated outcome	Setup
2026-08-25 09:23	NIFTY 24300 PE	5.5 / 10	09:24 @ 115.3	T 09:25	+8.5	SL	SCALP
2026-08-27 09:31	SENSEX 77700 PE	20* / 50	09:32 @ 332.8	SL 09:32	-23.6	T	OTHER
2026-08-28 09:17	SENSEX 77000 CE	22 / 30	09:18 @ 634.2	SL 09:19	-26.3	T	OTHER
2026-08-28 09:43	SENSEX 77000 CE	20* / 30	09:44 @ 734.1	SL 09:47	-24.6	SL	OTHER
2026-09-03 12:03	SENSEX 77100 PE	20* / 50	12:04 @ 437.7	SL 12:05	-23.9	no_claim	OTHER
2026-09-03 12:57	SENSEX 76400 CE	20* / 37	12:58 @ 330.2	SL 12:58	-23.6	T	BREAK/TRENDLINE
2026-09-03 15:01	SENSEX 76400 CE	13.5 / 20	15:02 @ 253.8	SL 15:02	-16.9	no_claim	OTHER
2026-09-07 11:13	NIFTY 24100 PE	2.5 / 10	11:14 @ 303.5	SL 11:14	-4.4	no_claim	OTHER
2026-09-10 11:12	SENSEX 74500 CE	20* / 40*	11:13 @ 350.6	T 11:15	+36.2	T	SUPPORT/REJECTION
2026-09-10 11:19	SENSEX 75000 PE	20* / 40*	11:20 @ 283.1	T 11:42	+36.4	T	SUPPORT/REJECTION
2026-09-11 10:16	NIFTY 23150 CE	12* / 10	10:17 @ 194.2	T 10:21	+8.3	T	SCALP
2026-09-15 10:29	NIFTY 23700 PE	12* / 24*	10:30 @ 332.1	T 10:46	+21.9	no_claim	OTHER
2026-09-15 14:09	NIFTY 23400 PE	12* / 101	14:10 @ 149.4	time 15:15	+58.8	no_claim	SCALP
2026-09-17 10:11	SENSEX 74000 CE	20* / 25	10:12 @ 614.8	SL 10:13	-24.3	T	BREAK/TRENDLINE
2026-09-17 11:45	SENSEX 74000 CE	20 / 40*	11:46 @ 438.9	SL 11:46	-23.9	SL	OTHER
2026-09-17 14:28	SENSEX 74000 CE	20* / 40*	14:29 @ 422.1	T 14:30	+36.1	T	BREAK/TRENDLINE
2026-09-23 09:18	SENSEX 74300 CE	20* / 40*	09:19 @ 546.9	T 09:19	+35.8	T	OTHER
2026-09-23 09:25	SENSEX 74300 CE	18 / 36*	09:26 @ 532.9	SL 09:26	-22.1	no_claim	SCALP
2026-09-23 09:54	SENSEX 74800 PE	20* / 21	09:55 @ 327.9	SL 09:56	-23.6	no_claim	OTHER
2026-09-29 10:26	NIFTY 22500 CE	4 / 15	10:27 @ 132.1	SL 10:28	-5.5	no_claim	SCALP
2026-10-01 10:01	SENSEX 72000 CE	20* / 40*	10:02 @ 559.6	SL 10:03	-24.1	no_claim	BREAK/TRENDLINE
2026-10-01 12:46	SENSEX 72300 PE	60 / 120*	12:47 @ 436.0	T 12:50	+115.9	no_claim	BREAK/TRENDLINE
2026-10-01 15:11	SENSEX 71600 CE	20 / 50	15:12 @ 364.4	SL 15:12	-23.7	T	OTHER
2026-10-05 10:26	NIFTY 22400 CE	12* / 15	10:27 @ 169.8	T 10:28	+13.3	no_claim	SCALP
2026-10-05 12:06	NIFTY 22650 PE	12* / 24*	12:07 @ 232.5	SL 12:18	-13.8	no_claim	OTHER
2026-10-05 12:54	NIFTY 22650 PE	12* / 36	12:55 @ 214.5	SL 12:57	-13.7	T	OTHER
2026-10-05 14:09	NIFTY 22400 CE	15 / 30	14:10 @ 182.5	T 14:21	+28.3	T	BREAK/TRENDLINE
2026-10-09 12:17	NIFTY 22200 CE	12* / 24*	12:18 @ 321.7	T 12:49	+22.0	no_claim	OTHER
2026-10-16 09:18	NIFTY 23150 PE	12* / 20	09:19 @ 258.6	SL 09:20	-13.8	no_claim	BREAK/TRENDLINE

Entry = open of the next 1m bar after the call minute, +0.5 slippage. SL/target are from his spoken points (fallback SL 12 Nifty / 20 Sensex, target 2×SL, marked *). Same-bar SL+target counts as SL. Time exit 15:15. −0.5 slippage at exit plus Rs20/order, STT 0.15%, exchange, SEBI, stamp and GST (Nifty lot 65, Sensex 20). Nifty and Sensex points are different sizes (Sensex premiums are roughly 3× Nifty), so the pooled mean is only indicative. Before charges: +5.6 per trade. With target forced to 2×SL: +2.7.

YT4. Stated vs real, by setup and index

group	n	win	pts	total	cut
BREAK/TRENDLINE	7.0	42.9	13.5	94.4	setup
OTHER	14.0	21.4	-9.9	-138.6	setup
SCALP	6.0	66.7	10.2	61.3	setup
SUPPORT/REJECTION	2.0	100.0	36.3	72.6	setup
NIFTY	12.0	58.3	9.2	109.9	index
SENSEX	17.0	29.4	-1.2	-20.1	index
SL	3.0	33.3	-13.3	-39.9	stated outcome
T	12.0	50.0	3.8	45.9	stated outcome
no_claim	14.0	35.7	6.0	83.8	stated outcome

His stated outcomes are the first “target done/touched” or “SL hit” phrase within 40 minutes of the call. They agree with real bars on 53% of the 15 scored calls where he stated one. A stated “target done” was a real target hit 6 of 12 times. By stated outcome: “target” calls +3.8 per trade (n=12), “SL” calls -13.3 (n=3), no claim +6.0 (n=14).

YT5. Against the self-reported log (overlapping dates)

date	log_rows	log_wins	log_pts	matched_yt_10min	stream_that_day	yt_fresh_calls
2026-08-26	4	3	199	1	yes	1
2026-08-31	2	2	115	0	yes	0
2026-09-01	13	10	304	0	yes	0
2026-09-11	7	6	459	0	yes	1
2026-09-25	5	4	620	0	no stream	0
2026-10-05	3	2	71	0	yes	4

The log covers only 6 dates in this window. On 25-Sep there was no stream, yet the log shows 5 trades for +620 pts. On 05-Oct the log's strikes (Nifty 24800CE, Sensex 79500) are 2,200 Nifty and 7,300 Sensex points away from that day's market (spot ≈22,580 / ≈72,200). On 01-Sep the log has 13 trades (10 wins), but the stream that day gives no fresh call with an activation. Overall, 1 of 34 log rows match a stream call (same side, ±10 min). The self-reported log is not a record of the live calls.

YT6. Taken vs skipped (redo of the discretion analysis with stream calls)

Same method as the Trader discretion section (s9/s10 code unchanged; YouTube calls with known side as “taken”; fit 25-Aug..30-Sep, holdout 1–23-Oct). 51 calls with a known side → 50 with entry-time features; 315 candidates; fit base take rate 12%. 48% of his calls have one of our V1–V4/P1–P8 candidates within ±10 min, same direction. The strongest contrasts: he takes trades after a strong move in his direction (side-aligned spot move over 15 min, standardised difference ≈1.0; 15-min premium change 19.5 vs 9.5 for skipped; BH q<0.01). In plain words, he chases momentum.

feature	taken_mean	skipped_mean	smd	mw_p	q_bh
spot_mv15	44.850	19.613	1.019	0.000	0.000
prem_chg15	19.458	9.456	0.823	0.000	0.000
prem_chg5	12.511	8.153	0.618	0.000	0.001
spot_pos	0.865	0.692	0.555	0.002	0.007
gap	-0.243	-0.046	-0.534	0.002	0.007
prem_dd	8.716	16.103	-0.486	0.002	0.007

Tree CV AUC 0.57 / holdout 0.62; logistic 0.54 / 0.47. No tree leaf had ≥10 takes at ≥1.2× the base rate, so the pre-registered extraction gives no filter. There is nothing to test on held-out data or 2024–26 history, and we did not go looking for one after the fact. Candidates he took: X_OLD −0.8 pts per trade (n=46). Skipped: −4.2 (n=269). Both are ≤0.

YT7. Limits

Recall is about 67%, so silent or off-camera trades (VIP Telegram) are missed. Those channels need sign-up and were not accessed.
Sensex calls after 05-Oct and all Bank Nifty calls can't be scored (no real bars on the box; no token used, no gap-filling).
Nearest-weekly expiry is assumed. Strike OCR can read the wrong panel. n=29 is small.

Files: yt_calls_scored.csv · yt_review_adjudicated.csv · yt_random_bench.csv · yt_by_group.csv · yt_vs_selflog.csv · yt_disc_candidates.csv · yt_disc_model.json · scripts scripts/yt_*.py, s12_yt_discretion.py, s13_yt_disc_model.py.

Expiry day: do cheap (~20) counter-trend options jump 2×–3×?

Added after Satya's 19:55 IST (27-Oct) question: “on expiries opposite to trend from around 20 premium there is 2x or 3x jumps”. Design pre-registered at 20:05 IST in PREREG.md (EXPIRY-CHEAP-COUNTER) before any number was computed. Real 1m bars of the expiring contract:

109 Nifty expiry days (108 with at least one signal; 26-Dec-2024 had none; 03-Oct-2024..27-Oct-2026; 10-06/13/19 from option panels without volume, 27-Oct from today's EOD file with ATM±4 strikes only).
104 Sensex expiry days (04-Oct-2024..01-Oct-2026).

Spot 15:15–15:29 is treated as missing. Settlement uses the official close. Option buyer only.

Verdict. The jumps are real, but they are not a counter-trend effect and they are not tradable.

How often: after the first counter-trend touch of 15–25, the option later printed ≥2× on 40% of signals (Nifty) and 38% (Sensex). It printed ≥3× on 21% / 19% and ≥5× on 12% / 7%. The median peak came 32 / 28 minutes after the signal.
Not counter-trend specific: a ~20 option picked at a random minute does about the same (2×: 35% / 32%), and so does the with-trend option (43% / 36%).
Most of them die: 83% (Nifty) and 95% (Sensex) settle at ≈0.
As a buyer trade: all 6 pre-registered test rows (2 indices × 3 exits) lose money in 2026 (-11.8 to -5.1 pts per trade), and all 6 train rows lose too. 0 of 6 count. 1 of 39 sensitivity test rows are positive (bands, Sensex scaled band, DECAY, time windows), and none of them is significant. Picking a regime on train does not rescue it.

EC1. Definitions (pre-registered)

Trend at minute t: UP if spot is above the day open AND above its level 20 minutes earlier; DOWN if both are below; otherwise none. The index has no volume, so VWAP is not used.
The trade: buy the counter-trend OTM option (UP → PE, DOWN → CE) at the first minute from 09:35 to 15:00 when its 1m close is in band (15–25; sensitivity 10–30; Sensex also ×≈3.3 Nifty-equivalent). One per side per day.
Variants and controls: DECAY = the option had traded ≥50 earlier that day. Controls are the with-trend option in the same band, and a random minute in the same window (30 repeats).
Exits: (a) half at 2× fill, half at 3× or the 15:29 close; (b) stop at 50% of fill, then trail 50% of the peak gain once the gain reaches 2× the risk (price ≥ 2× fill); (c) hold to settlement intrinsic.
Fill and costs: fill = next 1m open +0.5 (Sensex +1.5). Costs are Rs20/order, STT 0.1% / 0.15% from Apr-2026 on sells and settlement, plus exchange, SEBI, stamp and GST.
Split: train Oct-2024..Dec-2025, test 2026.

EC2. Frequency facts

ix	signal set	expiry_days	signals	reached 2× %	reached 3× %	reached 5× %	median min to peak	settled ≈0 %
NIFTY	counter 15–25	108	170.0	40.0	21.2	11.8	32.5	82.9
NIFTY	with-trend 15–25	108	170.0	42.9	21.2	10.6	26.0	79.4
NIFTY	counter 10–30	108	171.0	39.8	21.6	11.7	30.0	83.0
NIFTY	counter DECAY 15–25	108	156.0	39.7	19.2	7.7	52.5	78.8
NIFTY	counter W1 09:35–11	108	149.0	41.6	22.8	12.8	36.0	82.6
NIFTY	counter W2 11–13	108	133.0	38.3	21.8	10.5	46.0	75.9
NIFTY	counter W3 13–14:30	108	114.0	45.6	21.1	7.9	34.0	71.9
NIFTY	counter W4 14:30–15	108	56.0	30.4	16.1	5.4	8.0	71.4
NIFTY	random minute 15–25 (30 reps)	108	5,093.0	35.0	20.9	9.2	27.0	79.4
SENSEX	counter 15–25	104	166.0	38.0	18.7	7.2	28.0	95.2
SENSEX	with-trend 15–25	104	167.0	35.9	21.0	13.2	20.0	93.4
SENSEX	counter 10–30	104	167.0	37.7	18.6	7.2	28.0	95.2
SENSEX	counter DECAY 15–25	104	157.0	32.5	19.1	8.9	26.0	96.2
SENSEX	counter scaled band	104	167.0	43.7	25.1	11.4	42.0	80.2
SENSEX	counter W1 09:35–11	104	151.0	39.1	19.2	7.3	28.0	94.7
SENSEX	counter W2 11–13	104	138.0	32.6	21.0	13.0	32.0	96.4
SENSEX	counter W3 13–14:30	104	131.0	31.3	21.4	12.2	15.0	90.8
SENSEX	counter W4 14:30–15	104	103.0	36.9	25.2	10.7	9.0	87.4
SENSEX	random minute 15–25 (30 reps)	104	4,980.0	32.2	18.9	9.0	17.0	94.4

“Reached N×” = a later 1m high ≥ N × the signal-minute close, at any time up to 15:29. Measured from the fill (next open +slippage), ≥2× drops slightly. The median signal minute is 09:36, because a 15–25 option on the counter side usually already exists right after the trend check starts. Sensex points are about 3.3× Nifty points, so a raw 15–25 Sensex option is much further OTM; the “scaled band” row fixes that.

Share of expiry days where a counter-trend first signal (either side) reached 2× / 3× / 5×, by year:

ix	year	days_with_signal	any2x	any3x	any5x
NIFTY	2024	12	41.7	25.0	25.0
NIFTY	2025	53	45.3	28.3	18.9
NIFTY	2026	43	55.8	32.6	16.3
SENSEX	2024	12	50.0	33.3	8.3
SENSEX	2025	52	40.4	25.0	7.7
SENSEX	2026	40	60.0	32.5	17.5

Quarterly: ec_freq_quarter.csv.

EC3. Tradable test (primary, pre-registered; BH over the 6 test rows)

ix	exit	split	n	mean net pts	median	win %	p10	p90	best	mean w/o top 3	with-trend control	random-minute mean	pctile vs random	p	BH q (6)	counts
NIFTY	half 2x / half 3x or close	train	99	-7.37	-18.75	23.23	-23.66	27.90	35.93	-8.70	-6.79	-8.39	83.33	0.00	–	False
NIFTY	half 2x / half 3x or close	test	70	-6.08	-18.17	24.29	-23.89	30.14	36.58	-7.90	-5.43	-5.96	50.00	0.02	0.03	False
NIFTY	SL 50% + trail after 2x	train	99	-4.56	-10.53	26.26	-12.84	12.61	56.71	-5.88	-4.87	-4.39	60.00	0.00	–	False
NIFTY	SL 50% + trail after 2x	test	70	-5.15	-10.65	25.71	-12.93	12.97	26.57	-6.46	-0.58	-3.70	23.33	0.00	0.00	False
NIFTY	hold to settlement	train	99	-0.72	-19.21	13.13	-23.38	49.15	426.77	-9.99	-5.94	-5.96	100.00	0.91	–	False
NIFTY	hold to settlement	test	70	-8.14	-20.45	14.29	-24.23	26.42	207.18	-14.85	-2.81	-8.79	60.00	0.08	0.10	False
SENSEX	half 2x / half 3x or close	train	98	-13.99	-24.09	14.29	-27.98	25.57	30.72	-15.39	-15.08	-14.78	66.67	0.00	–	False
SENSEX	half 2x / half 3x or close	test	68	-11.80	-22.95	17.65	-26.81	28.49	32.89	-13.85	-10.29	-11.22	30.00	0.00	0.00	False
SENSEX	SL 50% + trail after 2x	train	98	-5.51	-14.34	24.49	-16.76	7.33	351.31	-10.05	-10.75	-9.38	96.67	0.16	–	False
SENSEX	SL 50% + trail after 2x	test	68	-8.87	-14.05	22.06	-15.70	11.13	18.35	-10.08	-5.08	-6.62	13.33	0.00	0.00	False
SENSEX	hold to settlement	train	98	-16.14	-22.59	3.06	-25.05	-20.52	352.30	-22.72	-18.86	-15.76	50.00	0.00	–	False
SENSEX	hold to settlement	test	68	-6.38	-22.31	5.88	-25.18	-18.87	591.71	-21.71	-0.81	-10.63	66.67	0.53	0.53	False

Net option points per unit (Sensex rows are in Sensex points; divide by ≈3.3 for Nifty-equivalent). A row counts only if the test mean > 0, p < 0.05, q < 0.10, it beats random (≥90th percentile) and the with-trend control, the train mean > 0, and the mean without the top 3 is > 0.

Why “2×–3× happens 20–40% of the time” still loses money: the median trade loses most of the premium. Exit (a) wins only 1 trade in 4 to 6, and most of the rest expire near zero, so the 2× and 3× winners don't pay for them. Hold-to-settlement has rare huge winners (best +592 pts) but a win rate of 3–14%, and it is negative without the top 3. The p < 0.05 rows are significantly negative.

EC4. Sensitivities (test 2026)

set	ix	exit	n	mean	median	win	no_top3	t_p	q_bh
band 10–30	NIFTY	half 2x / half 3x or close	70	-6.20	-17.65	22.86	-8.02	0.01	0.03
band 10–30	NIFTY	SL 50% + trail after 2x	70	-5.37	-10.35	24.29	-6.69	0.00	0.00
band 10–30	NIFTY	hold to settlement	70	-8.19	-20.20	12.86	-14.91	0.08	0.13
band 10–30	SENSEX	half 2x / half 3x or close	68	-11.80	-22.95	17.65	-13.85	0.00	0.00
band 10–30	SENSEX	SL 50% + trail after 2x	68	-8.89	-14.05	22.06	-10.09	0.00	0.00
band 10–30	SENSEX	hold to settlement	68	-6.40	-22.31	5.88	-21.73	0.53	0.63
Sensex scaled band	SENSEX	half 2x / half 3x or close	68	-14.33	-60.74	29.41	-19.94	0.10	0.16
Sensex scaled band	SENSEX	SL 50% + trail after 2x	68	-9.53	-34.54	26.47	-18.83	0.16	0.23
Sensex scaled band	SENSEX	hold to settlement	68	-16.59	-63.21	17.65	-43.13	0.36	0.44
DECAY 15–25	NIFTY	half 2x / half 3x or close	64	-5.74	-17.65	29.69	-7.80	0.04	0.07
DECAY 15–25	NIFTY	SL 50% + trail after 2x	64	-1.57	-12.11	29.69	-6.30	0.71	0.77
DECAY 15–25	NIFTY	hold to settlement	64	-5.69	-22.80	17.19	-15.09	0.35	0.44
DECAY 15–25	SENSEX	half 2x / half 3x or close	63	-11.50	-25.75	20.63	-13.85	0.00	0.00
DECAY 15–25	SENSEX	SL 50% + trail after 2x	63	-9.62	-15.64	22.22	-11.40	0.00	0.00
DECAY 15–25	SENSEX	hold to settlement	63	-11.91	-25.84	3.17	-25.35	0.36	0.44
W1 09:35–11	NIFTY	half 2x / half 3x or close	60	-6.39	-18.55	23.33	-8.54	0.02	0.04
W1 09:35–11	NIFTY	SL 50% + trail after 2x	60	-5.35	-10.65	25.00	-6.90	0.00	0.00
W1 09:35–11	NIFTY	hold to settlement	60	-9.24	-20.30	13.33	-15.76	0.05	0.09
W1 09:35–11	SENSEX	half 2x / half 3x or close	61	-11.05	-22.70	18.03	-13.31	0.00	0.00
W1 09:35–11	SENSEX	SL 50% + trail after 2x	61	-8.70	-13.91	22.95	-9.96	0.00	0.00
W1 09:35–11	SENSEX	hold to settlement	61	-4.40	-22.19	6.56	-21.48	0.70	0.77
W2 11–13	NIFTY	half 2x / half 3x or close	55	-3.74	-17.25	34.55	-6.00	0.23	0.31
W2 11–13	NIFTY	SL 50% + trail after 2x	55	1.20	-10.58	30.91	-4.85	0.81	0.85
W2 11–13	NIFTY	hold to settlement	55	-2.34	-19.02	21.82	-11.22	0.71	0.77
W2 11–13	SENSEX	half 2x / half 3x or close	55	-10.11	-23.55	25.45	-12.50	0.00	0.01
W2 11–13	SENSEX	SL 50% + trail after 2x	55	-4.65	-13.93	27.27	-8.65	0.15	0.22
W2 11–13	SENSEX	hold to settlement	55	-3.27	-22.34	5.45	-22.44	0.85	0.87
W3 13–14:30	NIFTY	half 2x / half 3x or close	50	-6.20	-17.57	28.00	-8.82	0.03	0.06
W3 13–14:30	NIFTY	SL 50% + trail after 2x	50	-5.19	-10.69	26.00	-6.70	0.00	0.00
W3 13–14:30	NIFTY	hold to settlement	50	-6.91	-19.42	20.00	-15.10	0.20	0.27
W3 13–14:30	SENSEX	half 2x / half 3x or close	55	-11.15	-23.25	21.82	-13.76	0.00	0.00
W3 13–14:30	SENSEX	SL 50% + trail after 2x	55	-6.91	-13.96	21.82	-9.89	0.00	0.01
W3 13–14:30	SENSEX	hold to settlement	55	-1.25	-22.04	7.27	-19.31	0.91	0.91
W4 14:30–15	NIFTY	half 2x / half 3x or close	32	-10.65	-18.42	18.75	-14.76	0.00	0.00
W4 14:30–15	NIFTY	SL 50% + trail after 2x	32	-6.18	-10.04	18.75	-8.57	0.00	0.00
W4 14:30–15	NIFTY	hold to settlement	32	-10.25	-17.92	15.62	-16.06	0.01	0.02
W4 14:30–15	SENSEX	half 2x / half 3x or close	42	-10.50	-21.57	23.81	-13.48	0.00	0.01
W4 14:30–15	SENSEX	SL 50% + trail after 2x	42	-7.00	-13.53	21.43	-10.35	0.00	0.01
W4 14:30–15	SENSEX	hold to settlement	42	-13.94	-21.74	7.14	-22.51	0.02	0.04

EC5. Regime chosen on train only, then tested on 2026

ix	exit	label	bucket	train_n	train_mean	test_n	test_mean	test_win	test_p	test w/o top 3	q_bh
NIFTY	half 2x / half 3x or close	L3_ER20	mid0.2-0.4	25	-4.81	17	-13.24	11.76	0.00	-19.76	0.01
NIFTY	SL 50% + trail after 2x	L2_DMA_trend	up	42	-1.00	21	-7.02	19.05	0.00	-10.26	0.01
NIFTY	hold to settlement	L2_DMA_trend	up	42	8.20	21	-7.81	9.52	0.48	-20.27	0.58
SENSEX	half 2x / half 3x or close	L1_VIX	>16	16	-8.91	23	-9.08	21.74	0.06	-15.04	0.09
SENSEX	SL 50% + trail after 2x	L3_ER20	mid0.2-0.4	34	0.03	19	-9.08	21.05	0.00	-13.47	0.01
SENSEX	hold to settlement	L3_ER20	mid0.2-0.4	34	-4.64	19	9.25	5.26	0.78	-23.46	0.78

The best train buckets were mostly still negative on train. The two that were positive on train both failed in 2026: Nifty DMA-trend up, hold to settlement (+8.2 → −7.8), and Sensex ER20 mid, trail (+0.0 → −9.1). The only positive test bucket (Sensex hold, +9.3, n=19) had a train mean of −4.6, wins on 1 of 19 trades, and is −23.5 without its top 3.

EC6. Fill realism

ix	src	n	vol_known	sig_vol_zero_pct	fill_vol_zero_pct	fill_vol_median	fill_vol_p10
NIFTY	by_expiry	164	1	0	0	3,004,992	1,117,084
NIFTY	eod_atm	2	0	0	0	5,257,915	5,257,915
NIFTY	opt_panels	4	0	–	–	–	–
SENSEX	by_expiry	166	1	0	0	372,550	123,720

Every signal and fill bar had traded volume (median about 3.0M Nifty and 0.37M Sensex units at the fill minute), so the prints are real trades, not stale quotes. Bid/ask spreads are unknown (no history). On a 20-premium option a 0.5–1 pt spread is 2.5–5% per side, so real results would be somewhat worse than shown.

Files: ec_signals.csv (every signal with outcome) · ec_random.csv · ec_freq.csv · ec_primary.csv · ec_sens.csv · ec_regime.csv · ec_fills.csv · scripts s15_ec_cache.py, s16_ec.py, s17_ec_stats.py.

Why 20? Is the ~20 premium level itself special on expiry days?

Added after Satya's 20:14 IST (27-Oct) follow-up: “but there is some thing with this 20 i think”. Design pre-registered at 20:20 IST in PREREG.md (WHY-20) before any number was computed. Same data and rules as the expiry section above: expiring contract, real 1m bars, spot 15:15–15:29 missing, official-close settlement, buyer only, next-open fill + slippage, full costs.

Verdict: 20 is not special.

Level scan: across 10 levels (5–100), the chance of a later 2× or 3×, of settling at ≈0, and the net expectancy all move smoothly with level. 20 falls on the curve for every metric and index: 0 of 18 pre-registered “is 20 an outlier” checks pass. Cheaper options are simply worse, and every level loses money for a buyer.
Round-number support: 20 does not hold better than 17, 18, 22 or 23. The small round-number advantage under a fixed −2 tolerance comes from level 10, where −2 is a wide cushion; with a % tolerance it disappears.
“Holds at 20, buy the base breakout”: on Nifty it is positive in 2026 (best: N=20, SL 50% + trail, +9.3 pts per trade, n=51). But it lost −8.5 on train, has p=0.29, and is −3.8 without the best 3 trades. 20 was the worst level on train and the best on test: a flip like that is noise, not structure. 0 of 12 pre-registered rows count.

W1. Level scan: first touch of each level from above (every OTM option, every expiry day)

ix	set	L	n	days	P(≥2×) %	P(≥3×) %	settle≈0 %	net/premium (a)	net/premium (b)	net/premium (c)	(a) train	(a) test
NIFTY	raw	5	1519	108	30.28	17.84	93.88	-0.67	-0.44	-0.58	-0.70	-0.63
NIFTY	raw	10	1274	108	37.28	22.61	86.97	-0.41	-0.22	-0.31	-0.43	-0.37
NIFTY	raw	15	1134	108	37.65	23.54	82.10	-0.33	-0.22	-0.30	-0.35	-0.30
NIFTY	raw	20	1046	108	39.29	22.47	78.01	-0.29	-0.19	-0.29	-0.28	-0.31
NIFTY	raw	25	987	108	38.80	23.51	73.45	-0.25	-0.18	-0.21	-0.27	-0.23
NIFTY	raw	30	907	108	39.80	23.04	69.57	-0.22	-0.14	-0.19	-0.23	-0.20
NIFTY	raw	40	745	108	39.60	22.42	65.10	-0.20	-0.15	-0.19	-0.20	-0.19
NIFTY	raw	50	567	96	38.10	19.58	63.14	-0.21	-0.10	-0.24	-0.21	-0.21
NIFTY	raw	75	232	48	37.50	19.40	61.21	-0.18	-0.07	-0.27	-0.19	-0.17
NIFTY	raw	100	96	19	34.38	15.62	59.38	-0.25	-0.13	-0.30	-0.42	-0.08
SENSEX	raw	5	3260	104	22.02	11.72	98.99	-1.30	-0.95	-1.05	-1.36	-1.19
SENSEX	raw	10	2669	104	29.60	17.38	96.82	-0.85	-0.57	-0.73	-0.91	-0.74
SENSEX	raw	15	2440	104	32.99	19.43	94.84	-0.67	-0.42	-0.63	-0.72	-0.55
SENSEX	raw	20	2309	104	33.82	21.09	92.59	-0.57	-0.36	-0.55	-0.62	-0.45
SENSEX	raw	25	2214	104	36.40	22.76	90.06	-0.47	-0.28	-0.44	-0.53	-0.35
SENSEX	raw	30	2126	104	36.97	22.77	88.76	-0.43	-0.26	-0.42	-0.50	-0.29
SENSEX	raw	40	2003	104	37.74	23.91	85.07	-0.36	-0.21	-0.33	-0.41	-0.26
SENSEX	raw	50	1889	104	38.01	24.35	81.79	-0.33	-0.19	-0.26	-0.38	-0.23
SENSEX	raw	75	1697	104	40.01	24.81	74.96	-0.24	-0.14	-0.20	-0.28	-0.16
SENSEX	raw	100	1540	104	40.39	24.48	69.03	-0.18	-0.11	-0.16	-0.21	-0.14
SENSEX	scaled	5	2414	104	33.31	20.13	94.24	-0.63	-0.40	-0.62	-0.70	-0.50
SENSEX	scaled	10	2084	104	38.00	23.08	87.52	-0.41	-0.23	-0.38	-0.46	-0.30
SENSEX	scaled	15	1892	104	37.90	24.26	82.14	-0.33	-0.21	-0.26	-0.39	-0.23
SENSEX	scaled	20	1787	104	40.01	24.73	76.89	-0.26	-0.15	-0.20	-0.29	-0.21
SENSEX	scaled	25	1644	104	39.78	24.88	73.36	-0.23	-0.14	-0.20	-0.26	-0.17
SENSEX	scaled	30	1555	104	40.26	25.02	69.45	-0.18	-0.11	-0.16	-0.21	-0.14
SENSEX	scaled	40	1319	104	40.49	23.65	64.59	-0.15	-0.12	-0.17	-0.17	-0.13
SENSEX	scaled	50	1054	100	40.80	23.15	62.62	-0.14	-0.10	-0.18	-0.13	-0.15
SENSEX	scaled	75	469	54	41.58	22.60	58.00	-0.10	-0.13	-0.15	-0.10	-0.13
SENSEX	scaled	100	226	22	40.71	19.91	58.85	-0.13	-0.02	-0.15	-0.09	-0.40

Band ±20% around each level (20 = 16–24). Net/premium = net P&L ÷ fill; −0.29 means you lose on average 29% of the premium paid. Error bars and CIs are a 95% bootstrap over days. Sensex “scaled” = level × (Sensex open / Nifty open) ≈ ×3.3.

Is 20 an outlier vs a smooth curve fitted to the other 9 levels?

ix	set	metric	at15	at20	at20_ci	at25	smooth_fit_at20	special
NIFTY	raw	P(≥2×) %	37.654	39.293	35.2..43.6	38.804	39.159	False
NIFTY	raw	P(≥3×) %	23.545	22.467	18.6..26.3	23.506	23.605	False
NIFTY	raw	P(settle≈0) %	82.099	78.011	75.5..80.8	73.455	75.775	False
NIFTY	raw	ret/premium exit a	-0.332	-0.291	-0.387..-0.196	-0.249	-0.271	False
NIFTY	raw	ret/premium exit b	-0.218	-0.194	-0.244..-0.14	-0.180	-0.175	False
NIFTY	raw	ret/premium exit c	-0.296	-0.287	-0.474..-0.072	-0.213	-0.228	False
SENSEX	raw	P(≥2×) %	32.992	33.824	30.6..36.8	36.405	34.661	False
SENSEX	raw	P(≥3×) %	19.426	21.091	17.9..24	22.764	21.366	False
SENSEX	raw	P(settle≈0) %	94.836	92.594	91.1..93.9	90.063	92.631	False
SENSEX	raw	ret/premium exit a	-0.667	-0.566	-0.639..-0.492	-0.473	-0.563	False
SENSEX	raw	ret/premium exit b	-0.425	-0.360	-0.404..-0.306	-0.280	-0.350	False
SENSEX	raw	ret/premium exit c	-0.630	-0.551	-0.774..-0.271	-0.441	-0.514	False
SENSEX	scaled	P(≥2×) %	37.896	40.011	36.7..43.4	39.781	39.344	False
SENSEX	scaled	P(≥3×) %	24.260	24.734	21.3..28.3	24.878	24.681	False
SENSEX	scaled	P(settle≈0) %	82.135	76.889	74.8..79	73.358	75.754	False
SENSEX	scaled	ret/premium exit a	-0.333	-0.261	-0.341..-0.176	-0.228	-0.258	False
SENSEX	scaled	ret/premium exit b	-0.207	-0.154	-0.195..-0.108	-0.141	-0.168	False
SENSEX	scaled	ret/premium exit c	-0.260	-0.201	-0.402..0.0123	-0.204	-0.231	False

W2. Round-number support: does the premium bounce at 20 more than at 17/18/22/23?

ix	set	L	round	n	low≥L−2 next 5m %	low≥L−2 next 15m %	low≥0.9L next 15m %	close>L after 15m %
NIFTY	raw	10	True	1397	61.3	36.6	17.5	33.0
NIFTY	raw	17	False	1226	35.6	17.6	15.3	34.0
NIFTY	raw	18	False	1203	35.2	17.1	15.2	34.7
NIFTY	raw	20	True	1182	32.5	17.5	17.5	36.0
NIFTY	raw	22	False	1157	30.5	16.3	16.9	34.6
NIFTY	raw	23	False	1139	28.1	14.3	16.4	34.3
NIFTY	raw	25	True	1104	25.5	11.7	14.4	34.1
NIFTY	raw	30	True	1071	23.6	13.1	16.6	35.2
NIFTY	raw	50	True	864	12.5	6.0	12.4	29.3
SENSEX	raw	10	True	2999	62.4	38.8	18.6	32.7
SENSEX	raw	17	False	2642	37.7	19.4	16.2	33.7
SENSEX	raw	18	False	2602	37.5	18.9	16.9	34.4
SENSEX	raw	20	True	2559	31.8	15.3	15.3	33.6
SENSEX	raw	22	False	2502	29.8	15.2	15.9	33.3
SENSEX	raw	23	False	2470	28.5	13.4	15.0	34.4
SENSEX	raw	25	True	2423	27.5	13.2	15.5	35.3
SENSEX	raw	30	True	2340	22.4	10.7	14.1	33.7
SENSEX	raw	50	True	2089	17.7	9.7	16.4	36.8
SENSEX	scaled	10	True	2291	56.4	30.2	14.4	34.8
SENSEX	scaled	17	False	2035	32.7	17.1	14.6	35.5
SENSEX	scaled	18	False	2006	32.9	17.1	15.8	35.6
SENSEX	scaled	20	True	1950	30.8	14.7	14.7	36.0
SENSEX	scaled	22	False	1923	29.4	15.5	17.2	37.0
SENSEX	scaled	23	False	1899	25.6	13.9	15.9	36.3
SENSEX	scaled	25	True	1848	24.8	13.6	15.9	37.3
SENSEX	scaled	30	True	1767	23.5	12.2	16.5	35.9
SENSEX	scaled	50	True	1487	12.6	7.3	14.7	31.9

Round {10,20,25,30,50} minus non-round {17,18,22,23}, and 20 minus mean(17,18,22,23) (percentage points, 95% day-bootstrap CI):

ix	set	metric	round_minus_nonround	ci	l20_minus_17_18_22_23	ci20
NIFTY	raw	hold5	1.11	-0.6..2.7	0.06	-2.5..2.7
NIFTY	raw	hold15	2.11	0.7..3.6	1.13	-1.2..3.5
NIFTY	raw	hold15_pct	-0.03	-1.4..1.2	1.56	-0.9..3.8
NIFTY	raw	up15	-0.70	-1.9..0.4	1.63	-0.5..3.7
SENSEX	raw	hold5	0.73	-0.7..2.0	-1.65	-3.8..0.4
SENSEX	raw	hold15	1.94	1.0..3.0	-1.49	-3.0..0.1
SENSEX	raw	hold15_pct	0.08	-0.9..1.1	-0.71	-2.3..0.7
SENSEX	raw	up15	0.31	-0.6..1.2	-0.32	-1.9..1.1
SENSEX	scaled	hold5	1.39	0.2..2.6	0.55	-1.6..2.8
SENSEX	scaled	hold15	0.67	-0.4..1.8	-1.28	-3.2..0.5
SENSEX	scaled	hold15_pct	-0.64	-1.7..0.4	-1.17	-2.9..0.7
SENSEX	scaled	up15	-0.79	-2.0..0.5	-0.09	-1.9..1.8

The fixed “low ≥ L−2” measure falls steadily with level: 61% at 10 and 33% at 20 (Nifty, 5 min). The round-number edge in that measure therefore comes from level 10, where −2 is a 20% cushion. With a fair tolerance (≥0.9L) every level is at 12–18% (Nifty), with no bump at 20. 20 vs its non-round neighbours: Nifty +1.1 points (CI −1.2..3.5), Sensex −1.5 (−3.0..0.1).

W3. “Holds at 20” variant (zone 18–22, floor 17, base N minutes, buy the break of the base high)

ix	N	exit	split	n	mean	median	win	no_top3	random_mean	random_pctile	p	q_bh_primary	counts
NIFTY	10	half 2× / half 3× or close	train	81	-10.11	-26.26	28.40	-13.47	-6.91	0.00	0.01	–	False
NIFTY	10	half 2× / half 3× or close	test	71	5.09	-1.14	42.25	1.27	-0.05	100.00	0.26	0.85	False
NIFTY	10	SL 50% + trail after 2×	train	81	-6.64	-15.17	27.16	-9.61	-3.74	3.33	0.02	–	False
NIFTY	10	SL 50% + trail after 2×	test	71	3.29	-14.09	35.21	-5.74	1.03	80.00	0.57	0.85	False
NIFTY	10	hold to settlement	train	81	-11.54	-27.68	22.22	-19.52	-6.87	0.00	0.07	–	False
NIFTY	10	hold to settlement	test	71	4.43	-24.62	28.17	-5.20	2.11	76.67	0.56	0.85	False
NIFTY	20	half 2× / half 3× or close	train	66	-11.10	-19.79	31.82	-14.97	-2.99	0.00	0.02	–	False
NIFTY	20	half 2× / half 3× or close	test	51	-0.28	-1.17	43.14	-4.47	2.71	3.33	0.96	0.96	False
NIFTY	20	SL 50% + trail after 2×	train	66	-8.49	-16.56	30.30	-11.13	-1.62	0.00	0.01	–	False
NIFTY	20	SL 50% + trail after 2×	test	51	9.25	-15.99	41.18	-3.77	3.31	96.67	0.29	0.85	False
NIFTY	20	hold to settlement	train	66	-11.73	-29.55	25.76	-21.05	-2.91	0.00	0.12	–	False
NIFTY	20	hold to settlement	test	51	6.83	-21.02	33.33	-6.39	4.95	80.00	0.51	0.85	False
SENSEX	10	half 2× / half 3× or close	train	114	-12.13	-29.70	24.56	-14.88	-12.15	50.00	0.00	–	False
SENSEX	10	half 2× / half 3× or close	test	76	-6.43	-27.70	30.26	-10.61	-7.51	80.00	0.17	0.85	False
SENSEX	10	SL 50% + trail after 2×	train	114	-4.20	-17.70	28.07	-10.66	-7.49	93.33	0.37	–	False
SENSEX	10	SL 50% + trail after 2×	test	76	-7.25	-17.07	28.95	-10.67	-5.19	13.33	0.02	0.22	False
SENSEX	10	hold to settlement	train	114	-11.62	-31.32	7.89	-26.40	-10.67	36.67	0.23	–	False
SENSEX	10	hold to settlement	test	76	0.84	-29.99	9.21	-24.26	0.18	56.67	0.96	0.96	False
SENSEX	20	half 2× / half 3× or close	train	91	-13.68	-31.85	23.08	-19.15	-9.76	3.33	0.03	–	False
SENSEX	20	half 2× / half 3× or close	test	67	-4.41	-28.05	29.85	-11.45	-5.31	76.67	0.49	0.85	False
SENSEX	20	SL 50% + trail after 2×	train	91	-10.74	-20.80	26.37	-15.09	-4.82	0.00	0.01	–	False
SENSEX	20	SL 50% + trail after 2×	test	67	-2.94	-18.60	25.37	-11.87	-2.45	46.67	0.63	0.85	False
SENSEX	20	hold to settlement	train	91	-9.10	-36.64	14.29	-28.36	-7.82	43.33	0.49	–	False
SENSEX	20	hold to settlement	test	67	3.53	-31.84	10.45	-27.64	5.23	36.67	0.86	0.96	False

A row counts only if the test mean > 0, p < 0.05, q < 0.10, it is ≥ 90th percentile vs random, the train mean > 0, the mean without the top 3 is > 0, and it beats the same construct at 15 and 25. Random = same window and side, option nearest the level, 30 repeats.

Same construct at every level (mean net pts per trade; Nifty and Sensex raw). 20 is the worst Nifty level on train and the best on test:

ix	N	exit	split	10	15	20	25	30	40	50
NIFTY	10	SL 50% + trail after 2×	test	-4.70	-7.57	3.29	-0.97	-3.41	-3.88	14.73
NIFTY	10	SL 50% + trail after 2×	train	-3.61	0.55	-6.64	4.76	-7.32	3.09	6.38
NIFTY	10	half 2× / half 3× or close	test	-2.37	-9.43	5.09	-7.57	-7.50	-4.02	-0.90
NIFTY	10	half 2× / half 3× or close	train	-5.87	-6.09	-10.11	-0.09	-5.86	-8.30	-1.55
NIFTY	10	hold to settlement	test	-6.61	-9.20	4.43	-9.03	-9.88	-6.65	-7.46
NIFTY	10	hold to settlement	train	1.55	5.74	-11.54	12.25	-4.14	-1.86	-1.54
NIFTY	20	SL 50% + trail after 2×	test	-6.52	-9.01	9.25	-7.61	-3.00	-11.22	7.71
NIFTY	20	SL 50% + trail after 2×	train	-4.13	2.91	-8.49	5.97	-0.91	-3.41	1.61
NIFTY	20	half 2× / half 3× or close	test	-8.66	-9.89	-0.28	-4.89	-10.25	1.13	-5.47
NIFTY	20	half 2× / half 3× or close	train	-6.67	-4.42	-11.10	3.63	-2.84	-10.50	1.95
NIFTY	20	hold to settlement	test	-7.08	-8.47	6.83	-5.35	-8.39	3.62	-3.22
NIFTY	20	hold to settlement	train	2.60	10.36	-11.73	13.78	-1.44	-1.84	3.50
SENSEX	10	SL 50% + trail after 2×	test	-5.03	-4.77	-7.25	-7.30	-8.58	3.21	3.36
SENSEX	10	SL 50% + trail after 2×	train	-7.43	-6.78	-4.20	-3.82	-3.29	-3.38	-14.72
SENSEX	10	half 2× / half 3× or close	test	-9.81	-10.98	-6.43	-8.77	-11.75	-4.08	-12.30
SENSEX	10	half 2× / half 3× or close	train	-11.62	-11.93	-12.13	-13.75	-12.54	-6.63	-13.82
SENSEX	10	hold to settlement	test	-1.62	-1.77	0.84	-25.25	-6.24	-7.70	-1.28
SENSEX	10	hold to settlement	train	-13.91	-11.84	-11.62	-16.10	-6.22	-3.86	-6.67
SENSEX	20	SL 50% + trail after 2×	test	-3.62	-6.15	-2.94	-10.79	8.35	-4.25	13.29
SENSEX	20	SL 50% + trail after 2×	train	-7.21	-12.11	-10.74	-11.48	-9.33	9.37	-4.07
SENSEX	20	half 2× / half 3× or close	test	-8.19	-7.91	-4.41	-8.89	-4.32	-1.46	-12.99
SENSEX	20	half 2× / half 3× or close	train	-14.03	-17.30	-13.68	-14.16	-12.68	10.86	-3.62
SENSEX	20	hold to settlement	test	5.55	-2.97	3.53	-25.97	9.07	-1.51	8.77
SENSEX	20	hold to settlement	train	-14.91	-15.62	-9.10	-14.77	-4.62	6.54	4.87

Time of day (Nifty, holds-at-20): most signals come before 11:00. That window loses on train (−10 to −15) and is about zero on test (−4.2 to +3.1). The 11:00–13:00 and 13:00–14:30 rows have 3–25 trades each and swing sign between train and test. See w20_holds_windows.csv. Sensitivity table (all levels, BH within the table): 20 of 114 test rows are positive and none is significant; the significant rows are negative.

Files: w20_scan_levels.csv · w20_scan_special.csv · w20_touch_levels.csv · w20_touch_round_vs_nonround.csv · w20_holds_summary.csv · w20_holds_by_level.csv · event files w20_scan / w20_touch / w20_holds(_random).csv · scripts s18_why20.py, s19_why20_stats.py, ec_lib.py.

“20 without time value”: an ITM ~20 premium that is almost all intrinsic

Satya's follow-up at 20:29 IST (27-Oct): “20 without timevalue can you add that condition”. The design was pre-registered at 20:31 IST in PREREG.md (TWENTY-NO-TV). State: the option is ITM by 18–22 index points (intrinsic from the 1m spot close), its premium is 16–24, and its time value (premium − intrinsic) is between −2 and +2. Sensitivities: |TV| ≤ 1 and ≤ 10% of premium. Sensex is shown raw and scaled (levels × Sensex/Nifty open, ≈3.3). Delta ≈ 1, so the option roughly doubles on a ~20-point spot move. Signals run 9:35–15:00 only, because spot from 15:15 onward is missing and is not filled.

Verdict: no edge, and the state is rare.

Rare and late: only 41 Nifty events on 37 of 109 expiry days, at a median 14:35. Time value only drains to ≤2 in the last hour. Sensex raw had just 2 events, because a Sensex option with no time value is never as cheap as 20. In 2026 there were only 5 trades per index, so the test period cannot confirm anything.
Fewer total losses, about the same chance of doubling: Nifty P(≥2×) is 39% vs 40% for the OTM-20. P(settle≈0) is 34% vs 76%. P(≥3×) is 5% vs 23%, because a 3× needs a 40-point spot move in about 30 minutes. The average loss per rupee of premium halves (−0.13 vs −0.26), but it is still a loss.
It is a spot bet: the same-minute direction on spot, before any cost, also loses on train (Nifty −7.4 / −5.3 / −8.2 points for exits a'/b'/c'). Futures pay about 7 points round trip in STT and charges (about 12 from Apr-2026), against about 1.2 for the option. So the delta-1 option is a cheaper wrapper for the same losing direction bet; neither has an edge. 0 of 18 pre-registered primary rows count.

N1. Frequencies: ITM-20 with no time value vs OTM-20 (from the Why-20 scan, signal ≥ 9:35)

what	grp	n	days	P(≥2×) %	P(≥3×) %	settle≈0 %	median min to peak	median signal minute (0=9:15)	net/premium (a)	net/premium (c)
ITM-20 no time value (tv2)	NIFTY_raw	41.00	37.00	39.02	4.88	34.15	8.00	320.00	-0.13	-0.09
OTM-20 all day	NIFTY_raw	732.00	108.00	40.44	23.36	75.82	30.00	118.00	-0.26	-0.23
OTM-20 from 13:00 (t>=225)	NIFTY_raw	196.00	75.00	41.33	20.92	75.00	16.00	271.50	-0.25	-0.38
ITM-20 no time value (tv2)	SENSEX_raw	2.00	2.00	50.00	50.00	50.00	22.00	333.00	0.08	1.29
OTM-20 all day	SENSEX_raw	1,786.00	104.00	33.09	20.27	91.88	16.00	163.00	-0.59	-0.54
OTM-20 from 13:00 (t>=225)	SENSEX_raw	633.00	104.00	33.81	19.59	88.94	9.00	292.00	-0.59	-0.67
ITM-20 no time value (tv2)	SENSEX_scaled	57.00	43.00	33.33	19.30	45.61	7.00	323.00	-0.19	-0.01
OTM-20 all day	SENSEX_scaled	1,278.00	104.00	38.42	23.79	75.12	28.00	129.00	-0.29	-0.19
OTM-20 from 13:00 (t>=225)	SENSEX_scaled	371.00	82.00	35.58	21.02	71.70	15.00	279.00	-0.34	-0.35

By time-value definition:

grp	tvdef	n	days	x2	x3	zero	peak_min_med	sig_time_med	tv_med
NIFTY_raw	tv1	26.0	24.0	50.0	11.5	38.5	14.0	305.0	0.1
NIFTY_raw	tv10pct	43.0	38.0	34.9	4.7	34.9	5.0	319.0	0.8
NIFTY_raw	tv2	41.0	37.0	39.0	4.9	34.1	8.0	320.0	0.6
SENSEX_raw	tv1	2.0	2.0	50.0	50.0	50.0	22.0	333.0	-0.1
SENSEX_raw	tv10pct	2.0	2.0	50.0	50.0	50.0	22.0	333.0	-0.1
SENSEX_raw	tv2	2.0	2.0	50.0	50.0	50.0	22.0	333.0	-0.1
SENSEX_scaled	tv1	40.0	32.0	42.5	20.0	37.5	11.5	322.0	1.0
SENSEX_scaled	tv10pct	58.0	43.0	32.8	19.0	44.8	9.0	322.0	2.2
SENSEX_scaled	tv2	57.0	43.0	33.3	19.3	45.6	7.0	323.0	2.2

By trend (same trend label as s16; counter = buying against the 20-minute move and the open):

grp	mode	n	days	x2	x3	zero	peak_min_med	sig_time_med	tv_med
NIFTY_raw	counter	13.0	13.0	46.2	0.0	46.2	15.0	300.0	0.8
NIFTY_raw	neutral	14.0	14.0	21.4	7.1	28.6	12.0	304.5	0.5
NIFTY_raw	with	14.0	13.0	50.0	7.1	28.6	3.0	322.0	0.4
SENSEX_raw	neutral	1.0	1.0	100.0	100.0	0.0	41.0	327.0	0.6
SENSEX_raw	with	1.0	1.0	0.0	0.0	100.0	3.0	339.0	-0.7
SENSEX_scaled	counter	11.0	10.0	18.2	0.0	36.4	7.0	323.0	3.0
SENSEX_scaled	neutral	22.0	19.0	22.7	13.6	59.1	4.0	329.0	2.9
SENSEX_scaled	with	24.0	22.0	50.0	33.3	37.5	11.5	312.5	0.4

By window (W2 11:00–13:00, W3 13:00–14:30, W4 14:30–15:00; there were no W1 events):

grp	win	n	days	x2	x3	zero	peak_min_med	sig_time_med	tv_med
NIFTY_raw	W2	7.0	7.0	28.6	0.0	57.1	29.0	189.0	0.4
NIFTY_raw	W3	12.0	12.0	41.7	16.7	33.3	23.5	277.5	1.3
NIFTY_raw	W4	22.0	21.0	40.9	0.0	27.3	3.0	337.5	0.5
SENSEX_raw	W4	2.0	2.0	50.0	50.0	50.0	22.0	333.0	-0.1
SENSEX_scaled	W2	6.0	4.0	33.3	16.7	50.0	32.5	164.5	1.4
SENSEX_scaled	W3	20.0	18.0	25.0	10.0	50.0	8.5	274.0	2.2
SENSEX_scaled	W4	31.0	28.0	38.7	25.8	41.9	6.0	334.0	2.1

N2. Tradable buyer test (earliest event per side per day; pts per unit after all costs)

Exits: (a) half at 2×, rest at 3× or the 15:29 close. (b) Stop at fill − 10 (≈ spot 10 against), then trail 50% of peak once +20. (c) Hold to settlement intrinsic (official close). Spot control at the same minute and direction: (a') stop −20, half +20, rest +40 or the 15:14 spot close. (b') stop −10, trail 50% once +20. (c') stop −20, else the official close. Spot-control figures are after futures costs. Coin-flip = random direction at the same minute (1000 sims). The R1 random control (random minute in the same window, contract in the state) almost equals the signal, because the state usually lasts only 1–2 minutes, so it is not informative here.

grp	mode	split	exit	n	mean	median	win	no_top3	p	q_bh_primary	random_mean	spot_same_dir	spot_coinflip_mean	counts
NIFTY_raw	all	train	a	35	-3.23	-13.92	40.00	-6.44	0.40	–	-2.70	-14.70	-7.94	False
NIFTY_raw	all	train	b	35	-1.30	-11.15	40.00	-3.81	0.53	–	-1.09	-12.53	-9.14	False
NIFTY_raw	all	train	c	35	-1.30	-13.20	37.14	-6.23	0.80	–	-0.93	-15.48	-10.50	False
NIFTY_raw	all	test	a	5	-2.76	1.32	60.00	-16.43	0.66	1.00	-2.76	-14.22	-14.51	False
NIFTY_raw	all	test	b	5	0.86	1.68	60.00	-11.25	0.88	1.00	0.86	-22.13	-14.81	False
NIFTY_raw	all	test	c	5	-6.19	-5.71	40.00	-15.64	0.22	1.00	-6.19	-23.82	-18.40	False
NIFTY_raw	counter	train	a	11	-2.25	-2.85	45.45	-12.59	0.75	–	-1.51	-10.36	-8.32	False
NIFTY_raw	counter	train	b	11	2.19	4.22	63.64	-2.25	0.54	–	-0.43	-10.60	-8.13	False
NIFTY_raw	counter	train	c	11	-1.85	-21.95	45.45	-14.42	0.81	–	-1.94	-12.41	-9.05	False
NIFTY_raw	counter	test	a	2	3.58	3.58	100.00	–	–	1.00	3.58	3.15	-14.17	False
NIFTY_raw	counter	test	b	2	9.21	9.21	100.00	–	–	1.00	9.21	-21.37	-21.37	False
NIFTY_raw	counter	test	c	2	-1.77	-1.77	50.00	–	–	1.00	-1.77	-10.59	-20.89	False
NIFTY_raw	with	train	a	11	-0.54	-6.88	45.45	-10.40	0.93	–	-0.06	-11.60	-7.22	False
NIFTY_raw	with	train	b	11	-1.31	-11.15	36.36	-9.91	0.79	–	-1.27	-14.50	-11.91	False
NIFTY_raw	with	train	c	11	2.00	-6.06	36.36	-10.91	0.80	–	2.53	-15.63	-11.96	False
NIFTY_raw	with	test	a	2	-3.31	-3.31	50.00	–	–	1.00	-3.31	-21.33	-13.99	False
NIFTY_raw	with	test	b	2	-1.42	-1.42	50.00	–	–	1.00	-1.42	-21.58	-7.89	False
NIFTY_raw	with	test	c	2	-6.98	-6.98	50.00	–	–	1.00	-6.98	-31.58	-15.12	False
SENSEX_scaled	all	train	a	41	-22.58	-62.66	36.59	-32.05	0.05	–	-20.86	-49.16	-18.71	False
SENSEX_scaled	all	train	b	41	-9.05	-36.70	34.15	-19.77	0.31	–	-4.16	-47.61	-26.48	False
SENSEX_scaled	all	train	c	41	-10.39	-60.35	36.59	-28.60	0.58	–	-9.41	-43.73	-15.34	False
SENSEX_scaled	all	test	a	5	28.65	95.72	60.00	-74.64	0.53	1.00	29.49	-3.66	-21.39	False
SENSEX_scaled	all	test	b	5	37.16	26.32	60.00	-36.26	0.36	1.00	37.16	-68.92	-66.62	False
SENSEX_scaled	all	test	c	5	69.37	86.40	60.00	-72.33	0.34	1.00	70.21	38.86	9.05	False
SENSEX_scaled	counter	train	a	8	-27.57	-36.36	37.50	-60.19	0.16	–	-27.47	-53.98	-23.04	False
SENSEX_scaled	counter	train	b	8	-24.85	-36.78	25.00	-36.86	0.02	–	-24.60	-52.90	-43.52	False
SENSEX_scaled	counter	train	c	8	-26.87	-33.58	37.50	-57.88	0.15	–	-26.57	-63.52	-32.20	False
SENSEX_scaled	counter	test	a	1	-75.62	-75.62	0.00	–	–	1.00	-70.49	-109.19	-33.68	False
SENSEX_scaled	counter	test	b	1	-36.00	-36.00	0.00	–	–	1.00	-35.99	-77.15	-77.15	False
SENSEX_scaled	counter	test	c	1	-73.31	-73.31	0.00	–	–	1.00	-68.18	-109.19	8.81	False
SENSEX_scaled	with	train	a	18	-3.87	-47.48	44.44	-24.10	0.83	–	-3.51	-38.55	-18.77	False
SENSEX_scaled	with	train	b	18	5.20	-36.59	38.89	-19.11	0.74	–	6.05	-43.35	-25.66	False
SENSEX_scaled	with	train	c	18	10.76	-44.89	44.44	-22.01	0.65	–	11.26	-26.90	-6.99	False
SENSEX_scaled	with	test	a	4	54.72	95.72	75.00	-73.67	0.29	1.00	54.48	22.72	-18.95	False
SENSEX_scaled	with	test	b	4	55.45	55.40	75.00	-36.51	0.25	1.00	55.45	-66.86	-63.93	False
SENSEX_scaled	with	test	c	4	105.05	118.20	75.00	-71.36	0.22	1.00	104.81	75.88	7.75	False

Spot/futures control before costs, and the futures cost (pts):

grp	split	futures cost pts (round trip)	spot (a') before costs	spot (b') before costs	spot (c') before costs
NIFTY_raw	test	11.63	-2.59	-10.50	-12.19
NIFTY_raw	train	7.26	-7.43	-5.27	-8.22
SENSEX_raw	train	19.65	4.25	-11.50	28.12
SENSEX_scaled	test	35.18	31.51	-33.74	74.04
SENSEX_scaled	train	21.59	-27.57	-26.02	-22.14

All 2026 (test) trades. Sensex-scaled test means are positive only because of 3 winners out of 5; without the top 3 they are −36 to −75:

grp	date	typ	t	mode	spot	K	ref	tv	fill	max_ratio	pnl_a	pnl_b	pnl_c	sa	sb	sc
NIFTY_raw	2026-01-06	PE	323	with	26,131.1	26,150.0	17.2	-1.7	17.4	1.1	-18.5	-11.2	-17.8	-28.1	-18.1	-28.1
NIFTY_raw	2026-02-17	PE	345	counter	25,730.7	25,750.0	21.1	1.9	21.6	1.7	1.3	1.7	2.2	4.9	-18.0	-2.7
NIFTY_raw	2026-04-21	CE	342	with	24,570.0	24,550.0	21.6	1.5	21.9	2.1	11.9	8.4	3.9	-14.6	-25.1	-35.1
NIFTY_raw	2026-06-16	PE	345	neutral	23,978.8	24,000.0	22.9	1.8	23.6	1.1	-14.4	-11.3	-13.5	-34.7	-24.7	-34.7
NIFTY_raw	2026-07-28	PE	337	counter	23,981.6	24,000.0	19.2	0.8	19.6	3.0	5.9	16.7	-5.7	1.4	-24.7	-18.5
SENSEX_scaled	2026-01-01	CE	334	with	85,270.2	85,200.0	69.7	-0.5	70.2	1.2	-73.7	-36.5	-71.4	-89.0	-56.4	-89.0
SENSEX_scaled	2026-01-08	PE	326	with	84,333.5	84,400.0	63.5	-2.9	66.3	4.2	95.7	147.5	150.0	75.3	-56.1	129.0
SENSEX_scaled	2026-05-07	PE	343	with	77,940.0	78,000.0	63.4	3.4	66.4	3.5	95.7	84.5	86.4	52.3	-77.6	50.0
SENSEX_scaled	2026-06-18	PE	342	counter	77,229.8	77,300.0	71.7	1.4	72.1	1.1	-75.6	-36.0	-73.3	-109.2	-77.2	-109.2
SENSEX_scaled	2026-07-30	CE	326	with	77,669.3	77,600.0	69.1	-0.2	70.0	5.0	101.1	26.3	255.1	52.2	-77.4	213.5

Regime filter chosen on train only (best single label with n ≥ 15), applied to test unchanged. With 3–5 test trades this proves nothing:

grp	exit	filter	train_n	train_mean	test_n	test_mean
NIFTY_raw	a	L3_ER20=range<0.2	20	0.48	5	-2.76
NIFTY_raw	b	L3_ER20=range<0.2	20	2.20	5	0.86
NIFTY_raw	c	L3_ER20=range<0.2	20	2.42	5	-6.19
SENSEX_scaled	a	L1_VIX=<13	17	-6.08	3	41.05
SENSEX_scaled	b	L1_VIX=<13	17	5.65	3	45.77
SENSEX_scaled	c	L2_DMA_trend=up	16	14.16	5	69.37

N3. Fill realism

grp	n	sig_vol_med	fill_vol_med	negTV_contracts
NIFTY_raw	41	9,451,650.0	9,161,800.0	12
SENSEX_raw	2	3,777,300.0	4,822,240.0	3
SENSEX_scaled	57	1,909,970.0	1,916,970.0	20

All signal and fill minutes had traded volume. The median volume at the signal minute is very high (about 9.5M units on Nifty), because these are near-ATM strikes in the last hour of expiry. Bid-ask spreads are unknown (there is no quote data). ITM spreads on expiry afternoons can be wider than the 0.5 / 1.5 slippage assumed. Time value is computed from the spot close against the option's last trade in the same minute, so it is noisy. 35 contracts were excluded at least once for printing more than 2 points below intrinsic.

Files: ntv_events.csv · ntv_trades.csv · ntv_summary.csv · ntv_vs_otm20.csv · ntv_freq*.csv · ntv_random.csv · ntv_regime.csv · ntv_fill.csv · ntv_spot_gross.csv · ntv_negtv.csv · scripts s20_notv.py, s21_notv_stats.py.

4. The 8 rules (as found in popular Indian trading guides)

#	Strategy	Rule as coded
P1	ORB15 breakout	First 5m close beyond 09:15–09:29 range (09:30–12:00). Stop = other side of the range.
P2	ORB5 first 5-min candle	First 5m close beyond the 09:15–09:19 candle (to 11:00). Stop = other side.
P3	Prior-day high/low breakout	Day opens inside the prior-day range; first 5m close above PDH / below PDL (to 14:00). Stop = breakout bar extreme ±5.
P4	Narrow-CPR breakout + VWAP	Narrow-CPR day only (width < 30th pct of last 250 days). First 5m close beyond CPR and on the same side of VWAP (09:30–12:00). Stop = other CPR edge.
P5	Supertrend(10,3) flip + VWAP	5m Supertrend(10,3) flips, confirmed by VWAP side (09:45–14:30). Stop = Supertrend line.
P6	EMA 9/21 cross + VWAP	5m EMA9 crosses EMA21, confirmed by VWAP side (09:45–14:30). Stop = 6-bar swing ±5.
P7	RSI(14) 30/70 reversal	5m RSI(14) crosses back above 30 (CE) / below 70 (PE) (09:45–14:30). Stop = 6-bar swing ±5.
P8	Gap fade (gap-fill)	Gap 0.3–1.5%: first 5m close back through the 15-min range toward the gap fill (to 11:00) → buy against the gap. Stop = other side of range.

Sources: stoxgo.in CPR guide, algoram.in “best NIFTY options buying strategy 2026”, tradingzenith.net “7 setups tested”, Ojha CPR playbook. The VWAP-pullback and with-gap ORB variants were already tested on the buyer-method audit.

5. Method, data and limits

Spot Nifty 1m 2022-01-03 to 2026-10-23 (Upstox historical). Spot 15:15–15:29 is never used (frozen since the 2026-08-03 closing auction). Options: real 1m ATM weekly bars from 21-Oct-2024 to 23-Oct-2026. 28 of 4,772 option rows had no bar at entry and were skipped, never filled.
Entry at the next minute's option open after a 5m spot close. Exits are decided on spot 1m closes and filled at the next option open. One trade per strategy per day.
Costs: 0.5 pt slippage per side + brokerage, STT (0.15% from 1-Apr-2026), exchange charges, stamp and GST. Lot 75 (65 from 2026).
VIX “at open” = prior VIX close (NSE VIX opens at the prior close). FII long % = NSE participant-wise OI (1,209 days fetched from the public archive), using the previous day's file. FII long % sat below 30% for all of 2026 versus about 44% median in 2022–24, so the >45% bucket has no 2026 data. That label partly just encodes time.
VWAP needs Nifty futures volume, which only exists from Oct-2024. P4–P6 therefore have one spot OOS year only.
The 2017–2021 walk-forward the brief asked for is impossible: no intraday data before 2022 is available.
Option selling and futures are out of scope.

Files: results · scripts · charts · pre-registration.

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	100	89	36	21	27	34	21	17
P2	87	100	34	21	24	33	24	16
P3	82	78	100	29	21	24	8	5
P4	75	77	46	100	15	21	16	2
P5	43	39	14	6	100	84	54	19
P6	45	44	14	8	70	100	50	17
P7	25	29	4	5	41	46	100	12
P8	90	86	13	3	64	72	56	100

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	0	9	7	5	32	36	57	2
P2	9	0	9	5	35	38	54	3
P3	16	20	0	5	35	41	76	3
P4	18	20	8	0	45	52	66	9
P5	51	56	24	20	0	10	21	2
P6	48	51	23	19	8	0	23	3
P7	68	66	40	22	16	21	0	3
P8	10	14	8	14	8	14	15	0

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	100	86	36	20	26	32	20	17
P2	86	100	34	20	24	32	23	16
P3	36	34	100	22	11	11	4	4
P4	20	20	22	100	6	7	5	2
P5	26	24	11	6	100	66	34	17
P6	32	32	11	7	66	100	37	17
P7	20	23	4	5	34	37	100	12
P8	17	16	4	2	17	17	12	100

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	100	89	36	21	27	34	21	17
P2	87	100	34	21	24	33	24	16
P3	82	78	100	29	21	24	8	5
P4	75	77	46	100	15	21	16	2
P5	43	39	14	6	100	84	54	19
P6	45	44	14	8	70	100	50	17
P7	25	29	4	5	41	46	100	12
P8	90	86	13	3	64	72	56	100

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	0	9	7	5	32	36	57	2
P2	9	0	9	5	35	38	54	3
P3	16	20	0	5	35	41	76	3
P4	18	20	8	0	45	52	66	9
P5	51	56	24	20	0	10	21	2
P6	48	51	23	19	8	0	23	3
P7	68	66	40	22	16	21	0	3
P8	10	14	8	14	8	14	15	0

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	100	86	36	20	26	32	20	17
P2	86	100	34	20	24	32	23	16
P3	36	34	100	22	11	11	4	4
P4	20	20	22	100	6	7	5	2
P5	26	24	11	6	100	66	34	17
P6	32	32	11	7	66	100	37	17
P7	20	23	4	5	34	37	100	12
P8	17	16	4	2	17	17	12	100

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	100	89	36	21	27	34	21	17
P2	87	100	34	21	24	33	24	16
P3	82	78	100	29	21	24	8	5
P4	75	77	46	100	15	21	16	2
P5	43	39	14	6	100	84	54	19
P6	45	44	14	8	70	100	50	17
P7	25	29	4	5	41	46	100	12
P8	90	86	13	3	64	72	56	100

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	0	9	7	5	32	36	57	2
P2	9	0	9	5	35	38	54	3
P3	16	20	0	5	35	41	76	3
P4	18	20	8	0	45	52	66	9
P5	51	56	24	20	0	10	21	2
P6	48	51	23	19	8	0	23	3
P7	68	66	40	22	16	21	0	3
P8	10	14	8	14	8	14	15	0

if ↓ fires	P1	P2	P3	P4	P5	P6	P7	P8
P1	100	86	36	20	26	32	20	17
P2	86	100	34	20	24	32	23	16
P3	36	34	100	22	11	11	4	4
P4	20	20	22	100	6	7	5	2
P5	26	24	11	6	100	66	34	17
P6	32	32	11	7	66	100	37	17
P7	20	23	4	5	34	37	100	12
P8	17	16	4	2	17	17	12	100