COIN-M Two-Day Idiosyncratic Down-Shock Rebound, BTC-Hedged: long a Binance COIN-M inverse alt perp + short equal-dollar BTCUSD_PERP after its 2-day move lags BTC by <=-2.5 sigma, fixed 3-day clock, max 4 new pairs per date (point-in-time 44-name BINANCE_CM USD_PERP panel, 1-DAY, market-neutral event panel, 20% intrabar alt-leg stop)

Original Research Lead hypothesis (not seeded). It tests the frozen Bybit two-day idiosyncratic down-shock rule (promoted iteration 1: 185 trades, PF 1.67, 1.75% per trade, DD 10.8%) on a fourth venue, Binance COIN-M inverse perps, with the rule unchanged. Lesson L266 says this template is worth testing; this test does not cite the Bybit result as evidence that it will work here. Why this venue: BINANCE_CM holds 2.8% of the corpus against a >=5% target. The design is long-short and market-neutral (long_short is 25%, against a <=55% long-only target), multi-instrument, and event-driven (the best-surviving tag family, 16%). No COIN-M hypothesis so far uses a hedged, residual-defined long-side trigger. The COIN-M relief-rally fade in the pipeline is the SHORT side on a range-location trigger. The COIN-M deleveraging flush is unhedged, uses a raw return and needs an OI gate. The COIN-M compression traps are NR7 patterns. MECHANISM: when one coin falls far more than BTC over two days without a market-wide move, the excess drop is mostly forced or impatient coin-specific selling: a liquidation cluster in that coin's perp, a large holder exiting, or a rumour. On COIN-M the collateral IS the coin, so a falling coin both cuts a long's margin and its collateral value, and liquidations there are convex (inverse-contract PnL). This makes forced selling overshoot. Liquidity providers who absorb it are paid as the coin partly recovers relative to BTC over the next few days. Shorting BTC removes the market factor. RULES (signal on each completed 1-DAY bar; bar timestamp = close, 00:00 UTC): W_alt = ln(close_t/close_{t-2}), W_btc the same for BTCUSD_PERP.BINANCE_CM, X = W_alt - W_btc. sd = std of the alt's daily log returns over vol_lookback days (baseline 60), ending at the bar two days before the signal (the shock days are excluded). zx = X / (sd*sqrt(2)). DATA-QUALITY RULE (pre-registered; it came out of the pre-study): skip the signal if the alt printed any zero-change daily close in the last 10 bars, or if sd < 1%. Stale prints on thin COIN-M contracts (VET, GALA, ZIL) otherwise produce about 900 false z extremes. UNIVERSE (point in time, using universe_at()/is_listed()): the 44 BINANCE_CM USD_PERP alts in `instruments`. A name is eligible once it has been listed >=90 days at the signal. Names that later delisted (LUNA) stay in the panel and are closed at delisting by the base template. EXECUTION: market orders at the signal bar's close. On this 24/7 venue that price is the next bar's open, the platform's faithful equivalent of a next-open fill (capability next_bar_open_fill: use_equivalent). The 20% stop is an intrabar protective stop (reduce-only stop-market, platform-matched; capability intrabar_stop: supported), not an exchange-resting order. SIZING: inverse contracts are sized with contracts_for_notional(). PRE-STUDY (Research Lead, decoded BINANCE_CM 1-DAY catalog closes, 2020-08 to 2026-10, 43 alts vs BTCUSD_PERP, one event per name per 3-day window, entry at the signal close, staleness filter applied, gross of costs): zx<=-2.5 fired 280 alt legs on 105 distinct event dates. Mean hedged (alt minus BTC) 3-day log return was +1.41% per leg and median +2.44%. Excluding the LUNA collapse leg (-531% log, i.e. about -99%), the mean was about +3.3%. With the same-date cap of 4 legs (most negative zx first): 179 legs on 105 dates, mean +1.94% after clipping the LUNA leg at the 20% stop, +1.91% excluding LUNA (t 2.51), median +0.97%. By year (capped, stop-clipped, legs/mean): 2020 5/+1.5%, 2021 9/-5.3%, 2022 25/+5.7%, 2023 42/-1.9%, 2024 59/+3.3%, 2025 19/+5.9%, 2026 20/+0.8%. 2021 and 2023 lost money and 2026 is flat, so the edge is concentrated in 2022, 2024 and 2025. That concentration is the main honest risk. The per-DATE averaged net return was +0.70% (t 0.81, 54% of dates positive), much weaker than the per-leg figure. The 2023 legs are the most numerous and lost money. COST ARITHMETIC: COIN-M taker is about 0.06% round trip per leg, so about 0.12% per pair on the alt notional, plus impact on thin alt contracts. A capped, stop-clipped gross of about +1.9% per pair is about 16x that cost; even the per-date figure of +0.82% gross is about 7x. PRE-REGISTERED COUNTS (lessons L238/L242/L259): about 105 distinct event dates over 6 years (about 17 per year); about 180 pairs, which is about 360 position legs plus BTC hedge resizes. EXPOSURE (L267): 10% of equity per alt leg, at most 4 new pairs per date, at most 6 open pairs. Maximum gross is therefore 60% long alts + 60% short BTC = 1.2x, at config.leverage 2.0, inside the 5x conservative cap. A 1x alt position contributes roughly 80-130% annualised vol, so a 10% leg contributes about 8-13%, well inside the 120% per-position budget. Expected strategy vol is about 10-15% a year, against the 60% ceiling.