TSE Prime Idiosyncratic Down-Shock Rebound, TOPIX-ETF Hedged: buy a Japanese large cap at the next Tokyo open after its close-to-close return lags the equal-weight JPEQ panel by <=-2.5 sigma, short yen-matched 1306.JPEQ (TOPIX ETF), fixed 5-session clock (JPEQ 1-DAY, ~100 names, market-neutral)

Original Research Lead hypothesis (not seeded). It applies lesson 266's reference-relative down-shock template (promoted on Bybit vs BTC, Hyperliquid vs BTC and ASX gold miners vs a peer basket) to a new venue and asset class: Tokyo Prime large caps on JPEQ. JPEQ has 6 experiments in the 5,007-experiment corpus, and no stock-level shock-rebound test has been run there. The idea also fills the under-represented long-short / hedged bucket (25% vs a 45%+ target). MECHANISM: Japanese large caps have a well-documented short-horizon reversal, which is stronger than in the US and is usually attributed to retail-dominated, liquidity-demanding selling (margin-call 追証 liquidation by 制度信用 longs, NISA retail stop-outs) that market makers absorb at a discount. The trigger measures the shock RELATIVE to the stock's own market, not as a raw return, so market-wide down days do not fire it. The hedge removes TOPIX beta, so the trade earns only the idiosyncratic rebound. SIGNAL: for each panel name i on each session t, residual r_i,t = log(C_i,t/C_i,t-1) - mean_j log(C_j,t/C_j,t-1), where the mean is the equal-weight average over all panel names with a bar that session. z_i,t = r_i,t / std(r_i, t-60..t-1), computed point in time with the std from the 60 PRIOR sessions only. EXECUTION: decide on the session-t daily close and fill at the session t+1 OPEN (the platform's native next_open_fill on session venues, supported). Exit with a market order decided at the close of the 4th session after entry, filling at the open of the 5th session after entry, for 5 sessions of exposure. HEDGE: whenever the long book changes, the strategy re-targets a single netted SHORT in 1306.JPEQ (NEXT FUNDS TOPIX ETF) to minus the total yen notional of open longs, rounded to the nearest 100-unit board lot. JPEQ supports shorts at 1x with 1.10%/yr borrow modelled. PRE-STUDY (Research Lead, frozen rule, daily JPEQ bars 2018-01-04 to 2026-10-30, 100 names): the <=-2.5 sigma trigger fired 2,733 times on 1,115 distinct sessions. The mean next-open to +5-session-open return was +1.11% raw and +0.24% relative to the equal-weight panel. The date-clustered t on the panel-relative return was 1.77. The panel-relative return was positive in 2018, 2019, 2020, 2022, 2023, 2024 and 2025, and negative in 2021 (-0.59%) and 2026 YTD (-0.29%). The +1.11% raw figure is mostly the panel's own drift: the panel is today's top-100 market caps, so it carries survivorship. The TOPIX hedge exists so that the analyst can judge the idiosyncratic leg, not that drift. COST ARITHMETIC: JPEQ charges zero commission and a 3 bp half-spread modelled per fill, so about 0.06% RT per leg and about 0.12% for the hedged pair per unit of long notional (5-day borrow on the hedge is about 0.02%). Expected capture is about +0.24% panel-relative gross, roughly 2x the hedged cost. Against TOPIX it should be somewhat higher, because the survivor panel's residual drift over TOPIX lands in the long arm. CAPS: 8% of equity per long name, at most 6 concurrent longs, at most 3 new entries per session (taking the most negative z first). Maximum same-date gross long exposure is 24% and maximum total long exposure 48%, matched by an equal yen short in 1306 (gross at most 0.96x, net about 0). Names whose single 100-share lot exceeds the 8% budget are skipped (e.g. high-priced 9983), not over-sized. Risk control route (b): a fixed time exit with capped sizing, every position at most 8% of equity, unlevered gross below 1x. No stop is invented; the source template has none.

Venue choice: JPEQ is almost unused (6 experiments, 0.1%), while BINANCE is at 48%. The idea also adds a hedged long-short design (direction quota under-filled) and a non-crypto asset class. Mechanism choice: the reference-relative down-shock with a fixed clock is the template that has transferred across venues (lesson 266). Lesson 266 asks that equity versions define the shock against a peer basket, which this does, and lesson 256/267 favour a hedge leg to keep drawdown near 10%. Differences from failures: unlike the JPEQ chip-diffusion failure (premise_falsified on survivorship in its short arm), this has no unhedged directional short arm, and the hedge is a TOPIX ETF, not stock picks. Unlike the USEQ NR7 and reversal-basket failures, the trigger is peer-relative, not a raw pattern. I also pre-studied the same rule on the Swiss CHEQ panel (44 domestic names), where panel-relative returns were about 0% (-0.08% at 2.5 sigma, 5 sessions), so CHEQ was not used. Pre-registered counts (lessons 238/266): 1,115 distinct event sessions and 2,733 events over 2018-01 to 2026-10 at the baseline threshold. The concurrency caps will reduce traded events to an estimated ~1,200-1,600.

JPEQ 1-DAY panel of all 100 names in config/jp_equities_universe.txt (primary 7203.JPEQ). Each session close it computes every name's log return minus the equal-weight panel mean. It z-scores that residual against the std of the 60 prior sessions' residuals, requiring all 60. Names at z <= -z_entry (2.5) are bought most-negative first: at most 3 new per session and at most 6 longs held after the next open, each sized at floor(8% of MTM equity / close / lot) board lots, and names whose single lot exceeds 8% are skipped. Market orders decided on the close fill at the next Tokyo open (native next_open_fill). Each long is sold by a market order decided at its hold_sessions-th close (5), so it fills at the open of the 6th session: 5 sessions open to open. Every close, one netted short in 1306.JPEQ (TOPIX ETF) is re-targeted to -hedge_ratio x the yen value (at close_t) of the longs held after the next open, rounded to the nearest 1306 lot. The difference is sent as one market order at the same open, and the short is bought back in full when no longs remain. With no 1306 bar there are no new entries. legs_balanced() is checked every session, and the hedge is always sized from the long book actually held. Risk route (b): fixed time exit, at most 8% of equity per name, gross below 1x, no stop or target.

Implements the hypothesis rules as written: peer mean over all panel names with a bar (1306 excluded), point-in-time 60-session residual std, z_entry/hold_sessions tunable within the plan's bounds (declared in _param_bounds), and fixed hedge_ratio 1.0, position_pct 0.08, vol_lookback 60, max_concurrent 6, max_new_per_session 3, hedge 1306.JPEQ. The full 100-name universe is used. point_in_time_universe=True keeps the late-listed 285A (Kioxia, Dec 2024) from cutting the window short; it joins the panel once it has 60 sessions. One forced implementation detail: JPEQ margin is 1x, and initial margin is checked at submission against everything open plus working orders. Exits decided on the same close have not filled yet at that point. So on a full book (6 longs = 48% plus 48% hedge) a same-open rotation would exceed equity, and the platform would cap or refuse it. Instead, the code shrinks entries itself, in whole lots, so that open gross + pending + new longs + added hedge stay within margin_headroom (0.98, fixed) x MTM equity. That avoids refused hedge legs. Expect rotation days on a full book to take smaller or no new entries, so traded events may fall below the pre-registered ~1,200-1,600; the analyst should look at entry_diag and the trade count. 1306.JPEQ is not yet in the catalog and must be collected on demand (1306.T). Without it the run should fail as a data error, never trade unhedged. A price-limit refusal of an exit is retried each close because the time exit is re-evaluated every session.