#1LIVE
DOWN-CAP MORNING FADE (close exit, Q0-Q3)
The flagship. +1.054%/trade net, 61.5% win, 9,548 trades, capacity 5-6x via the Q2/Q3 bands.
How it works: Buy the tiny illiquid stocks that overshoot DOWN on a gap-open morning at 9:15, then hold to the CLOSE (not lunch); the overshoot leaks back out all day, and exiting at the close beats the 13:30 exit by about +0.72%/trade.
| Period | net %/trade |
|---|
| 2021 | +1.22 |
| 2022 | +0.64 |
| 2023 | +0.95 |
| 2024 | +1.10 |
| 2025 | +1.51 |
| 2026 | +1.02 |
2026 evidence and deployment state
Live debut 8 Jul 2026: +1.52%/trade, 33 wins of 38. The band tilt (W2) sizes the 2-6% overshoot up to +1.68% / +2.22% per trade. The Q2/Q3 capacity extension (T1) is green in 2026 at +0.95% / +0.74% and lifts single-day deployable size to about Rs 5cr. House rule: flat open-to-close, full size, no second tranche, no intraday stop (X1), keep the |index gap|>=0.3% gate (W1) and index-relative excess (X3).
Receipt: RESEARCH-N.md (N3), RESEARCH-W.md (W2), RESEARCH-T1.md (Q2/Q3)
#2PAPER
DOWN-CAP HIGH-YIELD DIVIDEND RUN-UP (long)
New edge. +1.298%/trade net, t=7.1, 6 of 6 years green, about 150 events a year, uncrowded.
How it works: Buy a small non-F&O stock five sessions before its ex-dividend date and sell into the cum-dividend close the day before ex; high-yield names drift up into the record date. Yield ladder is monotone: 1-2% pays +1.08%, 2-3% pays +1.59%, over 3% pays +2.07%.
| Period | net %/trade |
|---|
| 2021 | +1.98 |
| 2022 | +0.77 |
| 2023 | +1.25 |
| 2024 | +1.12 |
| 2025 | +1.43 |
| 2026 | +1.36 |
2026 evidence and deployment state
2026 +1.36% raw (abnormal +0.45%). Point-in-time restricted (dividend announced by T-5) keeps 95% of the edge at +1.273%, still 6/6 green. Gate: yield>=1% and 20-day turnover>=2cr, cash/delivery. Ladder the size by yield. Paper now, calendar-armed for live once the run-up entry is wired to the corporate-actions feed.
Receipt: RESEARCH-Z4.md (leg a)
#3CALENDAR-ARMED
MIDCAP-150 RECONSTITUTION ADD PRINT-POP
Cleanest mechanical calendar trade: per-year 0 of 6 red on the abnormal print, monotone flow/ADV.
How it works: Buy each stock added to the Nifty Midcap 150 at the ED-2 close and sell the whole basket at the ED-1 rebalance print, into the passive funds' forced buying. One-day hold, so naked is effectively hedged.
| Period | the print |
|---|
| print abnormal | +0.63 |
| print net | +0.53 |
| naked net | +0.86 |
| win rate | +67% (naked) |
| per-year abnormal | 6 of 6 green |
| 2026 naked | -1.52 |
2026 evidence and deployment state
Naked print-pop +0.96% mean / +0.86% net, 67% win, 13-19 names twice a year. Abnormal print is green in all 6 years; only the naked leg went red in 2026 (-1.52%, weak tape). Flow/ADV is monotone: thin-print adds pay +0.83% at 72% win. Tilt toward thinner-ADV adds, hedge with a Midcap-150 future if the tape is risky. Next actionable cycle: the Sept 2026 review. Per-name basket log lives on W:.
Receipt: RESEARCH-Z2.md (Z2-c), parent RESEARCH-H10.md
#4PAPER
BAN-ENTRY INTRADAY SHORT
Ship-grade signal: +0.79% net/trade, 66% win, n=705, positive vs controls all 6 years.
How it works: On the day a stock ENTERS the F&O ban list, short the cash name at 9:15 and cover at 15:25. Fresh futures shorts are blocked, so the name bleeds all session: it opens +0.27% above the prior close and drains to -0.62% by 15:25 without reverting.
| Period | net %/trade |
|---|
| 2021 | +1.44 |
| 2022 | +0.75 |
| 2023 | +0.77 |
| 2024 | +0.73 |
| 2025 | +0.11 |
| 2026 | -0.19 |
2026 evidence and deployment state
2026 n=8, -0.19% (thin sample; the naked raw leg decayed to +0.11% by 2025, so the stable form is relative/index-hedged). Fill-robust at a 9:16 entry (+0.61% net, 64% win). About 120 trades a year, episodic. Blocked from live by borrow: needs SLB or an intraday-only cash short. The overnight leg of the same mechanism is the LIVE ban-spell sleeve below.
Receipt: RESEARCH-H9.md (test a)
#5PAPER
EXPIRY-DAY KILL-ZONE STRADDLE SELL (NIFTY weekly)
The one option-selling edge that survived honest fixed-strike accounting: +18.9%/trade of credit, n=101, 69% win.
How it works: On NIFTY weekly expiry afternoon, sell the at-the-money straddle at 13:55 and buy it back at 15:15 with a 1.6x-credit stop, harvesting the 0-DTE theta crush. The 14:00-15:00 hour bleeds about 21% of the opening premium, every year (H1a).
| Period | % of credit / trade |
|---|
| 2024 | +30.3 |
| 2025 | +10.2 |
| 2026 | +13.9 |
2026 evidence and deployment state
2026 n=20, +13.9%. Only 2.5 years of expired-option minute data exist, so this is a backtest, not live. Worst day -84.6% (4.5x mean) passes the tail line. SENSEX weekly is green every year (+13.1%) but fails the tail line (-74.3% = 5.7x mean), so it stays a watchlist. Monthlies are excluded everywhere (red years). Expiry detection MUST use the min-afternoon-straddle<0.25% rule, not a naive ratio.
Receipt: RESEARCH-TWIN.md (C11), RESEARCH-H1.md (H1a)
#6PAPER
DOWN-CAP NIGHT BOOK (dc_night)
Small, uncrowded, decaying. +0.652%/night, 59.2% win, green all 6 years.
How it works: Buy the top-6 momentum strong-closers (close position>=0.7) in the non-F&O small-cap universe at the close and sell at the next open. The overnight risk premium is uncrowded down here (only 1.5% F&O overlap).
| Period | net %/night |
|---|
| 2021 | +1.447 |
| 2022 | +0.353 |
| 2023 | +0.684 |
| 2024 | +0.955 |
| 2025 | +0.079 |
| 2026 | +0.276 |
2026 evidence and deployment state
2026 +0.276/night, but the 2025-26 stretch is only +0.08 / +0.28 at 54-58% up-nights, so this is decaying. Concentration passes (drop-top-50 still +0.411). Capacity about Rs 19L across the 6-name book. Deploy small, size to the recent regime, cut if it goes red.
Receipt: RESEARCH-W.md (W4)
#7PAPER
SMALLCAP OVERNIGHT VOLUME POP (dc_pop)
The sharper, rarer volume-surge cousin of dc_night. +1.487%/night (Q0-Q2), 69.3% win, green all 6 years.
How it works: Buy small-cap non-F&O names that gapped UP at least 3% AND closed strong (close position>=0.75) on at least 3x median turnover, at the close; sell at the next open. Rare: median 2 picks a night.
| Period | net %/night |
|---|
| 2021 | +2.419 |
| 2022 | +1.137 |
| 2023 | +1.528 |
| 2024 | +1.664 |
| 2025 | +0.143 |
| 2026 | +0.806 |
2026 evidence and deployment state
2026 +0.806/night, recovered from 2025's +0.143 (which was a 47.5%-win coin flip at the cost margin). Concentration passes across 815 names. Cap about Rs 2.5L a name. Additive beside dc_night. Deploy small, cut if red.
Receipt: RESEARCH-X.md (X4)
#8LIVE
BAN-SPELL LONG (nights only)
Older live regulatory sleeve. Index-hedged +1.07%/spell, +1.39% nights-only, 61% win, green all 6 years.
How it works: When a stock sits in the F&O ban list, fresh shorts are forbidden, so there is net upward pressure. Hold ONLY the overnight legs across the spell: buy at the close, sell at the next open, repeat each night. The daytime hours lose -0.2%/day, so the entire return is the overnight gap (+0.23%/night, 65% win, green 6/6).
| Period | the edge |
|---|
| hedged / spell | +1.07% |
| nights-only | +1.39% |
| per night | +0.23% |
| win | +61-65% |
| years green | 6 of 6 |
2026 evidence and deployment state
Reconfirmed independently by H9 test a(iv): buy 15:25 to next open reproduces the nights-only spell edge. About 80 spells a year, strengthening. Biggest risk: the name gaps down on the same bad news that caused the ban; the index hedge only partly covers a stock-specific drop.
Receipt: strategies_page.py (mech_trades banspell), RESEARCH-H9.md (test c)
#9LIVE
NIGHT BOOK: TOP MOMENTUM OVERNIGHT
Older live sleeve. +0.39%/night base, +0.44% with the strong-close filter, 68-70% nights green.
How it works: Hold only the strongest 20-day-momentum F&O names overnight through stock futures, keeping only the ones that also closed strong; buy at the close, sell at the next open. The overnight premium concentrates at the very top of the momentum list.
| Period | the edge |
|---|
| base / night | +0.39% |
| strong-close | +0.44% |
| best soft tilt | +0.448% (idx_ret>0 and cp>0.85) |
| 2026 | +0.09% |
2026 evidence and deployment state
2026 so far only +0.09%/night, the weakest year, so the whole book is decaying (2025 near flat). N1 killed the 3x-lever idea (no regime earns it) but kept a soft ranking tilt. Guarded by the results-calendar veto (Z3) below. Per-night log lives on the VM grader, not a static repo file.
Receipt: strategies_page.py, RESEARCH-N.md (N1 residue)
#10LIVE
EX-DIVIDEND DRIFT-DOWN SHORT (E5)
Older live mechanical sleeve, now reframed. +0.42%/trade historically, +0.25% net close in the clean re-test, 5 of 6 years green.
How it works: On a stock's ex-dividend morning the price should open lower by the whole dividend but usually opens down LESS; short at the open and cover by 11:00 (captures 69% of the drift) or the close.
| Period | abnormal %/trade |
|---|
| 2021 | -0.060 |
| 2022 | +0.041 |
| 2023 | -0.158 |
| 2024 | +0.222 |
| 2025 | +0.050 |
| 2026 | +0.027 |
2026 evidence and deployment state
Reframe from Z4: almost none of this is dividend-specific (abnormal near zero) -- it is the generic NSE open-fade harvest, simply SCHEDULED by the dividend calendar. So do NOT yield-scale (the highest-yield bucket has the worst win rate; >=2% buckets go negative at the 11:00 exit). Keep flat notional across yields 0.3-3.0 and drop the dead overnight-bounce tail.
Receipt: strategies_page.py, RESEARCH-Z4.md (leg b)
#11PAPER
SPECIAL-DIVIDEND DOWN-CAP EX-DAY SHORT
New small edge. +0.925%/trade net, 64% win, t=4.37, 6 of 6 years positive.
How it works: On the ex-day of a SPECIAL dividend (yield>=3%) in a small non-F&O name with 20-day turnover>=2cr, short from the open to the close. Pairs with the high-yield run-up long on the same names for roughly a +3% event cycle.
| Period | net %/trade |
|---|
| 2021 | +0.35 |
| 2024 | +2.90 |
| 2025 | +1.80 |
| 2026 | +1.28 |
| n | 103 (~17/yr) |
2026 evidence and deployment state
2026 +1.28%. Abnormal +0.368% (t=2.0). Size small: the top-5 days are 37.7% of the P&L. Regular-dividend down-cap ex-day shorts are DEAD (-0.005% after cost); only the special-dividend, high-yield slice works.
Receipt: RESEARCH-Z4.md (leg d)
#12LIVE
PANIC-DAY BOUNCE BASKET (event basket)
Older live sleeve. +0.70%/event, 81% of days green, 32 events in 10 years.
How it works: When 4 or more F&O stocks crash more than 2% intraday but fight back to close near their highs, the panic overshot; buy the basket at 15:25 and sell at the next morning's open.
| Period | the record |
|---|
| 10-year | +0.70%/event |
| days green | +81% |
| 2026 | +~1.5%/event (7 of 7 wins) |
2026 evidence and deployment state
7 events, 7 wins in 2026. H3 confirmed the clock: enter on the 14:30 nowcast plus 15:25, and sell at the next day's 09:15 open (holding longer gives back about 0.30%). On aftershock days the dispersion fade is an opening-print-only trade (the excess gap reverts inside the first 15 minutes). Stands down after a count>=25 mega-crash day (the 2020 lesson).
Receipt: strategies_page.py (abasket_10y), RESEARCH-H3.md
#13LIVE
EXCESS-GAP FADE (F6, the morning engine)
Older live sleeve. +0.43%/day traded, 17x over 5.5 years, green every year.
How it works: On mornings the index itself gaps more than 0.4%, fade only the EXCESS: short the names that gapped up too far beyond the index, buy the ones that gapped down too far (top 8, same-direction), 9:15 to the close, with a 1-ATR stop. Never fade the index part, that move is real information.
| Period | net %/day traded |
|---|
| worst years | 2023 +0.22, 2025 +0.27 |
| drawdown | -5.0% |
| years green | 6 of 6 |
| 2026 | +0.20 (37 days) |
2026 evidence and deployment state
2026 +0.20%/day over 37 days. Sized by the Y2 IV rule below. The 1-ATR stop caps the worst single trade at -10.2%. Biggest risk: a day the whole market trends one way and gaps keep extending past the stop.
Receipt: strategies_page.py (fable_lab f6_champion), RESEARCH-Y.md
#14RULE
IV-PERCENTILE FADE SIZER (Y2)
Protective alpha. A live sizing rule, not a trade: downweight the fade on low-IV mornings.
How it works: At 9:20 on any gap morning, read the NIFTY ATM-IV trailing-1-year percentile. Size the down-cap / excess-gap fade to 0.5x if the percentile is in its bottom third, else 1.0x. Low-IV mornings pay noticeably less; there is no upsize on high-IV mornings (the top tercile is not reliable enough to lever).
| Period | high-minus-low IV, %/day |
|---|
| 2021 | +0.22 |
| 2022 | +0.22 |
| 2023 | +0.34 |
| 2024 | -0.07 |
| 2025 | +0.12 |
| 2026 | +0.04 |
2026 evidence and deployment state
high>low in 5 of 6 years (2024 the thin-n exception). On the real fade book, the high-IV tercile pays +0.955%/day vs +0.450% for the low tercile. Log iv_0920, its percentile, and the size multiplier every gap morning whether or not a trade fires.
Receipt: RESEARCH-Y.md (Y2)
#15CALENDAR-ARMED
RESULTS-CALENDAR VETO (the TRENT rule)
Protective alpha. Variance hygiene for the overnight book, calendar-armed.
How it works: Drop any overnight night-book pick whose results print lands inside its close-to-next-open window and backfill with the next-ranked name. Earnings overnight is variance, not edge.
| Period | reporting picks vs rest |
|---|
| reporting picks | -0.218% mean |
| not reporting | +0.338% mean |
| variance | ~2x wider |
| p5 tail | 5x fatter (-6.32% vs -1.17%) |
| win | 52.7% vs 65.1% |
2026 evidence and deployment state
Only 0.7% of picks (55 of 7,704) report inside the window, but they are worse in 5 of 6 years and carry the tail (ANGELONE -9.69%, HINDZINC -8.62%). Fires at book formation (~15:25) off the NSE board-meeting-intimation calendar feed. Mirror studies (pre-results bias, post-results PEAD drift) were both killed.
Receipt: RESEARCH-Z3.md
#16RULE
BONUS / SPLIT EX-DATE VETO
Protective alpha. A do-not-hold risk rule.
How it works: Never initiate or hold a small-cap long through a bonus or split ex-date, and never buy the post-split 'cheap share'. The cheap share bleeds about -4% to -6% (control-adjusted) over the next 5-10 sessions, and the effect is strengthening 2024-26.
| Period | cheap-share long, ex+10 |
|---|
| raw | -6.04% |
| control-adj | -5.64% |
| win | 24% |
| years | red every year |
| trend | worsening (2021 -5.41 to 2026 -11.86) |
2026 evidence and deployment state
The mirror smallcap short earns +5.6% but is not tradeable (no borrow in micro/small caps); the F&O-shortable subset is red in 3 of 6 years, so only the veto ships. Source list flows from the same corporate-actions endpoint pm_ops.py already uses.
Receipt: RESEARCH-T5.md
#17RULE
EXIT-BY-RECORD-DATE RULE
Protective alpha. An operational exit rule salvaged from a killed edge.
How it works: If long any name into a buyback or corporate-action record date for any reason, exit BY the record date. The post-record unwind costs about 1.5% over two sessions more than 70% of the time.
| Period | post-record unwind |
|---|
| cost | ~-1.5% in 2 sessions |
| frequency | 70%+ of events |
| the trade itself | killed (-0.54%/event) |
2026 evidence and deployment state
The run-up-into-record-date trade was a look-ahead artifact (the honest announcement+1 rule is -0.54%/event, red-median 5 of 6 years), because NSE corp-actions carries no announcement timestamp. Method lesson: any corp-action study must join corporate-announcements per symbol for point-in-time discipline.
Receipt: RESEARCH-Z1.md
#18RULE
THETA KILL-ZONE TIMING (H1a)
Protective alpha. A clock, not a trade: where premium decays fastest.
How it works: If the machine ever sells option premium, sell into the 14:00-15:00 expiry-day hour and never in the morning. A 0-DTE straddle loses about 84% of its opening premium by the close, and the single fastest hour is 14:00-15:00, about 21% of opening value, every year. This is the window C11 harvests.
| Period | % of opening premium lost 14:00-15:00 |
|---|
| 2021 | 17.2 |
| 2022 | 22.3 |
| 2023 | 21.6 |
| 2024 | 18.5 |
| 2025 | 22.3 |
2026 evidence and deployment state
A tight 17-22% band, the peak-decay hour in every year across 191 expiry days. There is no comparable intraday window on 1-DTE or 2+DTE (that theta is overnight). The naked morning short-premium version was killed (see graveyard).
Receipt: RESEARCH-H1.md (H1a)
#19RULE
FADE HOUSE RULES (bundle)
Protective alpha. Six load-bearing rules distilled from the fade kill-tests.
How it works: Keep the fade flat open-to-close at full size: no second tranche and no intraday stop (X1, the trough IS the open). Keep the |index gap|>=0.3% gate (W1) and index-relative, not peer-relative, excess (X3). Do not size the fade up on panic or event days (X2, the sign is unstable). Never hold the fade overnight (W3, it is a strictly one-day event). Prefer low-delivery night names (N2 inverse). Never expect a T2T gap-down bounce to survive to the next close (X9).
| Period | what each rule saved |
|---|
| ladder + stop | -0.55% / -0.10% per trade |
| drop the gate | +0.28 vs +1.05 (ratio 0.27) |
| peer excess | +0.36 vs +1.03 |
| panic size-up | red 2021 and 2024 |
| day-2 hold | red 5 of 6 years |
2026 evidence and deployment state
These are the negative results that keep the flagship honest: each was a plausible improvement that testing rejected, and skipping any one of them would bleed the edge.
Receipt: RESEARCH-X.md (X1/X2/X3), RESEARCH-W.md (W1/W3), RESEARCH-N.md (N2)
#20LIVE
Rights-issue long veto (U2)
protective alpha
How it works: Never hold a long through a rights ex-date window: parents bleed ~2.5% over ex+1..+5 (71% of the time, event-specific, 6/6y) as renunciation arbitrage sellers press the stock. Flows through the same evening veto pipeline as bonus/split.
| Period | result |
|---|
| post-ex short (the mirror), all years | +1.8 to +2.5%/event green 6/6 |
2026 evidence and deployment state
veto active from 13 Jul; F&O short overlay = forward-watch satellite (~1.5 events/yr)
Receipt: RESEARCH-U2.md
#21CALENDAR-ARMED (late March 2027)
FY-end tax-loss rebound (U6)
calendar sleeve, 1x/year, big
How it works: Buy the worst-performing small-cap decile at the last March close (India's tax year-end forces loss-harvest selling that then vanishes), hold 10-20 sessions through April.
| Period | result |
|---|
| 2022 | +10.9% |
| 2023 | +11.6% |
| 2024 | +7.0% |
| 2025 | +5.2% |
| 2026 | +27.5% |
| spread vs controls | +3.8% every year; other-month boundaries ~0; December mirror absent |
2026 evidence and deployment state
n/a until Mar 2027; rule text in RESEARCH-U6.md
Receipt: RESEARCH-U6.md
#22LIVE
Pharma Signals site
distribution, not alpha
How it works: Separate public property: pharma-sector signals from the machine, sized for a $30M book (3% cap, liquidity-bound), anonymized signal families, permanent issue-log + after-the-fact forward-test ledger.
| Period | result |
|---|
| URL | pharma-signals.pages.dev |
2026 evidence and deployment state
forward test begins Mon 13 Jul
Receipt: pharma_page.py
#23LIVE (paper)
Claude-BTST synthesis sleeve
judgment layer, forward-test
How it works: Daily 15:18: the machine hands its full mechanical candidate slate (night book, pops, ban/ex-div shorts, event names, all vetoes) to Claude, which picks the 3 best overnight positions with conviction + basis tags. Published on ops, graded at next open, gate-protected like every sleeve.
| Period | result |
|---|
| forward test | begins 13 Jul 2026; no backtest by design: judgment is graded live |
2026 evidence and deployment state
first picks Mon 13 Jul 15:18
Receipt: claude_btst.py
#24LIVE
THE ALLOCATOR (W22C): risk-parity capital layer
rate-raiser on the whole book, no new signal
How it works: Inverse-volatility weights (20% cap) across all sleeves, walk-forward verified: the flat config had 34% of risk in the most volatile sleeve. Deployed: F6 cut hard, freed capital spread across the low-vol mechanical sleeves; monthly recompute rule armed for when live history matures.
| Period | result |
|---|
| vol-matched growth vs flat | +23% (plain) to +148% (capped form) |
| max drawdown | -27% to -37% lower |
2026 evidence and deployment state
applied to the live config 11 Jul
Receipt: RESEARCH-W22C.md
#25LIVE from tonight (laptop feeders nse_push 17:30 + claude_btst 15:18)
Buyback record-date veto (X12B-S exhaust)
How it works: ca_veto now catches buyback record dates: night book, dc picks, pops and Claude-BTST never hold long into one (-3.25% T-1..T+5 unwind, red 6/6y, t=-7.7 avoided).
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-X12B.md
#26LIVE on VM from Mon 13 Jul (suggest mode)
Go-live blocker fixes B1-B5 (exec engine hardening)
How it works: Idempotency ledger, latching P&L kill switch, gross-cap backstop + cross-sleeve dedup, uniform gate enforcement, atomic f6 entry+stop with unwind. Two review-caught bugs fixed + unit-tested. Deployed to VM, exercised daily by suggest mode.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: REDTEAM-GOLIVE.md
#27PAPER from Mon 13 Jul, recorder to build
W24A results-night IV crush (PAPER, young-edge track)
How it works: Short ATM straddle T-1 15:20 -> T 10:00 on top-60 reporters. n=193 with imputed tails: +8.2 to +11.5%/premium, t 4.3-6.7, all years green both bounds, top-1 3.4%. Pessimistic worst 6.5x vs 6x line + 2.5y sample = NO MONEY: paper-forward, auto-promote at n>=20 clean.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-W24A.md
#28PAPER-FIRST (young-edge gate), sleeve build commissioned
IPO 30d post-unlock long (W26A)
How it works: LONG IPO names T+2 -> T+10 after the 30-day anchor unlock passes. +1.18% abnormal vs turnover+momentum-matched controls, t=2.56, n=390, green 5/5 years, top-1 2%. Overhang-lift mechanism; 90d variant dies under the same control.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-W26A.md
#29LIVE (suggest) from Mon 13 Jul
NB3 night-book regime sizing (W27 improver)
How it works: x1.5 on idx-up/strong-close nights, x0.886 otherwise (avg 1.0x, mean-preserving). Book +0.248 -> +0.271%/night, delta t=3.55 green 6/6y, worst-5 tail SHALLOWER. Deployed pm_ops + exec_engine.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-W27-IMPROVER.md
#30LIVE on VM cron
RISKGUARD continuous risk layer
How it works: Hourly + 15:10 deep pass on VM: exposure, worst-night tail rupees, regime flags, protection states; 320k overnight cap advisory; GTT disaster stops at -12% staged (real Dhan v2 forever API) behind live_enabled.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RISKGUARD.md
#31PAPER half-size from next expiry, recorder build in progress
S3-A expiry-morning straddle drain (PAPER)
How it works: SELL NIFTY 0DTE ATM straddle 09:25 -> 11:00 on expiry days. n=131: +6.98%/premium net, t=2.66, all years green, tail 4.3x, r=0.07 vs C11, sequential same-day capital. The C11 clock's morning half.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-S3B.md
#32DORMANT (entry condition = the regime dial)
W34B listing-winner runners (DORMANT-ARMED)
How it works: Mainboard IPOs closing day-1 at cp>=0.8 and +15%: buy day-2, asymmetric trail. Passes all lottery lines (n=65, +1.57% mean, p99 +71%, worst yr -3.1% of budget) BUT the fuel is regime-gated: zero 2026 qualifiers. Self-arming mania tripwire, zero standing cost, paper-first when it wakes.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-W34B.md
#33PAPER (young-edge gate), sleeve build commissioned
W36R monthly-low deep-pullback bounce (Manu's thesis, PAPER)
How it works: Top-100 F&O in >=20% pullback: touch of a confirmed monthly LOW, entry next open after the touch day closes back above, target prior swing high. 2025-26: +2.06%/trade vs +0.83% placebo (t=2.85, n=167); held-out 2021-24 confirms (t=2.41), no red year. Owner-sourced; his data audit recovered the buried touches.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-W36R.md
#34LIVE on VM cron
EXEC-FMEA guards (metaguard + holiday calendar + fail-closed)
How it works: 31 failure modes catalogued, 22 guarded: metaguard watchdog (30-min process/artifact/token/disk checks), NSE holiday awareness, fail-closed on corrupt gates, corruption-tolerant signal reads, laptop heartbeats, log rotation. Drilled live: killed daemon detected, corrupt gates suppressed all entries.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: EXEC-FMEA.md
#35Playbook in RESEARCH-W40.md section 6
W40 intraday anatomy playbook (knowledge asset)
How it works: Minute-level anatomy of the funded level edge: separation decisive 12:30-14:00 (t~6), at-touch entry = free not better (+1.47 vs +1.51), the FLUSH-and-reclaim beats gentle defense (+2.49 vs +0.98): terror at a level is fuel. No exit rule cleared the ship bar: live sleeve unchanged; the playbook is the ship.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-W40.md
#36DEPLOYED to VM 11 Jul night
The Great Verification (5-audit battery + 7 fixes)
How it works: PIT: zero look-ahead, 3 fixes (inert results-veto restored, close-grading honesty, fresh liq_hist). Drift: nightbook returned to verified spec, banentry grading day fixed. Robustness: no cliffs in top-5. Rehearsal: Monday simulated clean, NB3 exit-sizing fixed. W42: 1% risk law + min-stop floor shipped.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: PIT-AUDIT.md / DRIFT-AUDIT.md / ROBUSTNESS-AUDIT.md / EXEC-REHEARSAL.md / RESEARCH-W42.md
#37DEPLOYED; promotion back to full size via forward counter
event_A re-adjudication + tier separation
How it works: The edge STANDS under its verified count>=4 gate (green 6/6 incl 2024 +0.23%, pooled +1.58% n=233, 2025-26 strongest). Red-2024/decay were the UNGATED form's artifacts. Fixes: relaxed nearA/wide tiers now ride paper with named tiers; event alloc halved to 80k until 20 forward trades.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-EVENTA-READJ.md
#38DEPLOYED (paper-only Fridays)
W45B Friday event veto (curiosity find: the weekend is not paid)
How it works: Friday overnight entries earn one night's edge for 3 nights of risk: event_A Fridays are NET NEGATIVE (difT -2.02), nightbook Friday per-night edge collapses to 0.071% with 7/15 worst tails. Ship: Friday event baskets ride paper-only (measured, unfunded). Queued for daylight review: nightbook Friday 0.58x + the Monday size-up seed.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-W45B.md
#39Knowledge asset + dial spec; quarterly recompute queued
W45D Naivety Index (the graveyard as an instrument)
How it works: Built from our own killed pattern-edges: the retail-chase index fades monotonically z +0.97 (2021) -> -0.91 (2026H1: the series floor). Empirical proof the forced-flows-only architecture matches the CURRENT regime, plus a quarterly dial spec (3 cheap chase-edges) that says when patterns become tradeable again.
Detailed receipt is in the named research document.
2026 evidence and deployment state
Receipt: RESEARCH-W45D.md