Breadth-Confirmed Absorbed Flush, BTC-Hedged: buy every Binance USD-M old-alt perp that prints an absorbed-flush day (range >=2x its 60-day return std, down close, close in the upper 45% of range) ONLY on dates when >=2 names of a 26-name point-in-time panel print it together, short equal-dollar BTCUSDT.BINANCE, fixed 3-day clock (1-DAY, max 8 legs at 6%, long-short)

Original Research Lead hypothesis (not seeded). It refines the absorbed-flush rule (promoted on Bybit, 200 trades PF 1.72, and on Hyperliquid, 151 trades PF 1.70; the Binance SPOT replication ended underpowered with DD 22.7%) in two ways that the brief's lessons ask for, and both come from a pre-study, not from an assumption. (1) BREADTH CONFIRMATION. The analyst's spot post-mortem suggested the perp results rely on leveraged-liquidation flow. A liquidation flush that is absorbed is a MARKET-WIDE event: many alts get swept and bought on the same UTC day. The pre-study below shows the rule's whole edge sits on such dates. Single-name absorbed flushes (only one panel name that day) lost money: 105 events, hedged mean -2.46%, date t -3.01. Dates with >=2 names: 86 dates, 492 events, hedged mean +2.41% per event, date-clustered t 2.39. So the trigger is: absorbed-flush bar on the name AND at least K=2 panel names (counted among all eligible panel names, whether or not they are already held) print it on the same daily bar. (2) BTC HEDGE + CONCURRENCY CAP (lesson L267/L256): each entry date opens at most 8 alt legs at 6% of equity each (gross long <=48%), ranked by range multiple (range/sd) descending, and a short BTCUSDT.BINANCE leg of equal total dollar notional (hedge ratio 1.0, pre-registered, not tuned). The book is market-neutral in dollars, long alt-specific rebound vs BTC. RULE (frozen from the promoted panels): on each daily bar for each eligible name, sd = std of daily log returns over the 60 bars ending at the PRIOR bar; range = (high-low)/prior close; close_loc = (close-low)/(high-low). Absorbed flush := range >= 2.0*sd AND close < prior close AND close_loc > 0.55. UNIVERSE (fixed before measurement; point in time): 26 older Binance USD-M USDT perps disjoint from the Bybit 13 panel, the Hyperliquid 15 panel, the Binance SPOT 24 panel, the BTC/ETH/SOL/BNB/XRP/ADA/LINK core and the L250 over-mined HL list: AXS, KSM, 1INCH, ZEN, DASH, CELR, COTI, ICX, IOST, KNC, LRC, OGN, RLC, RSR, RVN, SKL, STORJ, SXP, TRB, ANKR, AR, GALA, MATIC, FTM, WAVES, OCEAN. Delisted/renamed perps (MATIC ends 2024-09; SXP, FTM, WAVES, OCEAN, ICX, LRC, STORJ end before 2026-11) are kept: a leg is eligible on a date only when listed then with >=70 daily bars of history (universe_at()/is_listed(), parameters.point_in_time_universe = true) and the base template closes it at its last price on delisting; a delisted name simply stops counting toward the breadth K. WHY BINANCE USD-M although it is 48% of the corpus: it is the only perp venue that serves delisted contracts and history from 2020 for these names (Bybit/OKX/HL do not), and none of these 26 alts is among the corpus's over-used instruments; the mechanism (perp liquidation flush) requires a perp venue. EXECUTION SUBSTITUTION (capability next_bar_open_fill = use_equivalent): orders are market orders sent at the signal bar's close (00:00 UTC); on a 24/7 venue that price is the next bar's open. RISK CONTROL route (b): fixed 3-day time exit with capped sizing (6% per alt leg, at most 8 alt legs per entry date, BTC hedge equal to the date's alt notional; gross <= ~1x equity, no single position near 100% of equity). PRE-REGISTERED MEASUREMENT (Research Lead, catalog BINANCE 1-DAY bars 2020-02 to 2026-11, decoded from parquet; non-overlapping per coin, entry at signal close, exit at close 3 days later, 0.10% RT per leg so 0.20% for the hedged pair, funding NOT deducted): all absorbed-flush events 597 on 191 dates, hedged mean +1.56% but date-mean -0.33% (t -0.52) -> the unconditioned rule does not transfer here. With K>=2, cap 8 at 6%, hedged: 394 alt legs on 86 distinct entry dates, total +74%, max DD 5.5%, worst entry date -3.7% of equity, best +13.2%, rough event-level annualised Sharpe ~1.0. Unhedged equivalent: +111%, DD 9.6%. K>=3: 60 dates, hedged date t 1.81. By year (K>=3, unhedged date mean): 2021 +5.3%, 2022 +7.6%, 2023 +0.8%, 2024 +3.7%, 2025 +3.0%, 2026 -1.5% (one negative year). Caveats stated up front: returns are right-skewed (median hedged event +0.73% vs mean +2.41%), a few dates (2021-05-24, 2024-03-09) carry much of the total, and K was chosen after seeing that single-name events lose - the analyst should treat K=2 as the registered baseline and K in {2,3,4} as the only plausible alternatives. Distinct event dates (86) exceed the ~60 that lesson L271 asks for.