Every analysis / research page published on aitradeanalytics.tech (57 live pages, each checked for HTTP 200 at build time). Newest first within each category. Verdicts are taken from each page; * = derived from the page's out-of-sample table.
Old studies charged 3 pts slippage/side; real Upstox ATM bid-ask median 0.25 pt -> realistic cost ~2.0 pts/trade (Rs133/lot) not ~7; fixing it adds ~+4.9 pts/trade. Buyer-style exits, with-trend, first-hour, strike choice change results by <1 pt. Re-run (train Oct-24..Dec-25, test 2026, pre-declared, BH over 6): no candidate profitable OOS (best +0.9 pts, p=0.41). ORB with-gap has a real spot edge (+6 to +8 pts/trade 2022-26) but the long option keeps only 25-50% of it.
Live read-only Nifty bias dashboard: fusion factor A–I scores, ATM chain, 1m chart with VWAP and S/R, news and research rule cards; refreshes every 30s.
Old studies charged 3 pts slippage/side; real Upstox ATM bid-ask median 0.25 pt -> realistic cost ~2.0 pts/trade (Rs133/lot) not ~7; fixing it adds ~+4.9 pts/trade. Buyer-style exits, with-trend, first-hour, strike choice change results by <1 pt. Re-run (train Oct-24..Dec-25, test 2026, pre-declared, BH over 6): no candidate profitable OOS (best +0.9 pts, p=0.41). ORB with-gap has a real spot edge (+6 to +8 pts/trade 2022-26) but the long option keeps only 25-50% of it.
1-min spot/futures/options/VIX/BN 2022–26-Oct-2026. ≥200-pt recovery days: 26% of days in 2024–25, 8/16 in Oct-2026 (p=0.007) — a volatility regime (13/16 big-range days), not a new bottom-catching pattern. Put writing/call cover/basis/BN-lead/heavy-put-strike did NOT separate recoveries. Price rule (low within 30 min of open, bounce ≤5 min, VIX≥14.46) hit 28.8% vs 14.5% base OOS 2026 (p=0.003, BH q=0.024); ATM-CE trades after costs ≈ breakeven only with a spot-based runner exit (+2 pts/trade, p=0.39). Today 26-Oct case study included.
Fusion bias flipped a median 76 min after 80+ pt moves began (or never); of 14 leading signals tested out-of-sample 2025-26, 3 passed (Bank Nifty divergence and fast-market as either-way warnings, day high/low rejection modestly directional) and are live on /live/; none profitable alone for an option buyer.
Hub of descriptive "if this happens early in the session, how often does that follow?" rules on Nifty 1m data with Wilson CIs; in-sample, no costs, not an edge by itself.
38 trigger days (2022 – Oct 2026) with a >150-pt first-30-min move: 55% held ±150 at the close, 21% never re-crossed, 79% closed beyond the open in the same direction.
Buying ATM CE/PE/straddle at 15:25 and selling at the next 09:15/09:16 loses every day after costs (CE −14.7, PE −5.4, straddle −20.1 pts/trade, 2024-08 to 2026-09); day-direction and last30 filters do not fix it; 0 of 1,440 variants pass both halves + random test — FAIL.
All 1,167 sessions 2022–2026: last-30-min direction has ~zero link to next-day gap/first hour (ρ −0.01); late sell-offs gap up more (73% vs 60%) but the bounce is gone by 09:20; 12 of 960 ATM option-buy variants positive after costs, none beat random — FAIL.
Hub for intraday rule backtests (ORB, gap fill/go, stop-hunt, power hour, regimes) with walk-forward OOS after costs; all OOS net-negative except one regime filter the page flags as not a tradable edge.
Buying ATM CE at day-low / PE at day-high retests lost money in all 8 variants; slow, decelerating approaches do predict holds (PASS as a predictor) but the option buys still lose.
Overnight/swing and weekly option-premium setups with walk-forward OOS after costs; no result is statistically significant (best OOS p≈0.12, overnight) and the options sample is short.
SVI smile fits the Nifty weekly chain best (1.00 pt RMSE; Bates, SABR, Heston behind); none of 11 smile/IV/OI/parity clues gave a buy-side edge after costs in the 2026 test.
Do multi-day build-ups of option mispricing predict Nifty? Direction FAIL; next-day range PASS as a predictor but FAIL as a buy; breakout WATCH — mostly FAIL for an option buyer.
Every-minute option prices vs SVI fair value: only 4,055 tradable mismatches (median 1.6 pts), clustered on expiry days and the first 30 min; buying "cheap" ones lost money after costs.
Component-level breakdown of 9 sessions (Upstox 1-min + NSE delivery/participant OI): rallies on 13/15/16-Oct were FII index-futures short covering (OI down, FII shorts −3k/−13k/−5k) with clients selling; 14/19/22/23-Oct sell days saw FIIs re-short; falls led by HDFCBANK/ICICIBANK/RELIANCE/BHARTIARTL on mostly below-normal delivery; Jio-IPO days hit mid/small caps far harder than Nifty. Descriptive, n=9.
Bank results surprise vs next open (293 events, 2012–2026): the bank's own stock reacts (ρ 0.23) but the Bank Nifty gap does not (ρ 0.04); net-beat Saturday weekends median BN gap +0.26% (n=10, WATCH).
Over 15 years, high-impact news days averaged a 1.47% absolute Nifty move vs 0.69% on quiet days, but direction correlation is weak (ρ −0.04): news tags track volatility, not direction.
How Nifty responds to sectors, oil, the Dow, rupee and rates (2012–2026): +0.29% open gap per 1% overnight Dow (R² 0.23); the Hang Seng open is the top feature; oil barely matters.
First PPO reinforcement-learning run on Nifty (80k steps, CPU): −6.40% total return over 554 trades (42.8% win rate), vs buy-and-hold −11.69%. Research only.
Plain-English explainer of the v2 RL trader: an agent looks at Nifty hourly, chooses flat/long/short plus its own stop and target, and learns by trial and error.
Rebuild at exact entry times (225 takes; holdout Jul–Sep 2026): no narrow take-filter survives the holdout with time-symmetric controls (holdout AUC 0.49–0.51).
IF/THEN filters mined from the trade log, takes vs skipped similar setups (3,001 rows): logged takes +3.40 pts vs skipped +1.40; GB AUC 0.58 (modest). Superseded by the timed rebuild.
Who bought / who sold — results ← back to report components_2026-10-13_0915.csv (2 KB) components_2026-10-13_1428.csv (2 KB) components_2026-10-23_1336.csv (2 KB) components_report.txt (5 KB) fut_minu