Hypotheses

Original Research Lead hypothesis, after the published 'VRP' timing rule for volatility ETPs (Tony Cooper, 'Easy Volatility Investing', SSRN 2013). Short-volatility ETPs earn the VIX-futures roll-down plus the equity variance risk premium: implied volatility (VIX) usually sits above the volatility that is later realized, and SVXY collects that gap by being short front VIX futures. The premium is not constant. When realized volatility catches up with or exceeds VIX, the insurance seller is no longer paid for the risk, the futures curve tends to flatten or invert, and the roll-down turns into a loss. These are the episodes in which SVXY's large losses happen. The rule holds SVXY only while the premium is positive and stands in cash otherwise. Instruments: SVXY.USEQ (traded), SPY.USEQ (signal only: realized volatility), and the VIX index close from supplementary `global_benchmarks` (key VIX, daily). Signal at the close of session t, using data through session t-1 only (VIX's 16:15 ET settlement is later than SPY's 16:00 close, so same-day VIX would leak 15 minutes; the whole signal is lagged one session): VRP(t-1) = VIX_close(t-1) - 100 x sqrt(252) x stdev of the last rv_window SPY daily log close-to-close returns ending at t-1. Execution is next_open_fill (supported on USEQ): a decision at session t's close fills at session t+1's open. The strategy trades only from 2018-03-01. SVXY switched from -1x to -0.5x daily exposure on 2018-02-28, after the Feb-2018 'Volmageddon' loss of ~90%, so the earlier history comes from a different product. 2011-2018 bars serve only as warm-up. Pre-study (fixed rule, rv_window 10, threshold 0, one-session lag, open-to-open fills, 30% allocation, 2018-03 to 2026-10): 83 entries, average trade +1.76%, in the market 83% of sessions, Sharpe 0.62, max drawdown 13%. Buy-and-hold SVXY at the same 30% allocation over the same period: Sharpe 0.49, max drawdown 25%. So the gate's pre-registered incremental value (L217) is about +0.13 Sharpe and about half the drawdown. It is not a large return improvement. The pre-study varied the window: 5 sessions gave Sharpe 0.37-0.47, 21 gave 0.35-0.54. An added VIX-below-10-day-mean trend gate doubled turnover and cut the average trade to 0.1-0.4%, so it is excluded. Fee arithmetic: USEQ has zero commission and ~0.02-0.05% round trip (spread plus impact; SVXY trades millions of shares a day). The average trade of ~1.5-1.8% is about 30x cost. Worst open-to-open SVXY session since 2018: -27.8% (2024-08-02 to 08-05). At a 40% allocation that costs 11% of equity, inside the 15% max-trade-loss budget. Risk route (b): signal exit with capped, unlevered sizing (40% of equity, no leverage). Long-only on purpose. Shorting SVXY or buying UVXY in the off state would hold the negative-carry side (UVXY decays about 50-80% a year) and is unmodeled for borrow; the off state is cash. Rationale for the buckets: USEQ is 9.8% of the corpus, and the equity volatility-carry mechanism is untested in the factory. The one prior SVXY run was a calendar pre-VIX-settlement window that ended overfit. This is a regime gate on the premium itself, not a calendar window. Long-only is over-represented; it is justified here because the only tradable side with positive carry is long SVXY.

Original Research Lead hypothesis (not seeded). It is a frozen-rule venue/universe transfer (lesson L228: every transfer is a fresh test, reported on its own) of the promoted 'Capitulation-Close Next-Day Reversal on OKX Spot' (472 trades, average trade 1.50%, Sharpe 1.10). A Binance-spot breadth test of the same rule is in the pipeline. This one moves the rule onto HYPERLIQUID perps. HL is 13.3% of the corpus against a >=20% target, and live trading is shifting there. EVENT ('capitulation close'): a completed 1-DAY bar whose high-low range is >= range_mult (1.5) x the mean range of the previous atr_len (20) bars (event bar excluded) AND whose close-location value CLV = (close - low)/(high - low) is <= clv_max (0.15). That is a range-expansion day on which sellers were still in control at the print. Buy at that close and sell at the close hold_days (1) bars later. MECHANISM: a wide day that closes at its low is the footprint of forced, price-insensitive selling (stop cascades and liquidations of levered longs, which on Hyperliquid run through a transparent on-chain liquidation engine and HLP backstop) continuing up to the bar close. That flow is exhausted at the print, liquidity providers who widened during the cascade return, and the next session retraces part of the overshoot. This is a close-location trigger, not a return-sigma or volume trigger. PRE-REGISTERED PRE-STUDY (fixed rule, catalog HL 1-DAY bars, 20 names, 2021-01 to 2026-10, close-to-close, 0.09% round trip deducted): 825 events on 295 DISTINCT entry dates (~51 dates/yr; per lesson L238 the date count, not the name count, is the evidence unit). Mean net next-day return is +1.21% per event (median +1.00%, win rate 60%), and +0.48% when averaged per entry date first (t ~1.4 on dates alone, so the edge is real but clustered on market-wide capitulations). By year, per event: 2021 +1.39%, 2022 +2.42%, 2023 +1.27%, 2024 -0.60%, 2025 +0.45%, 2026 +1.27%. 2024 is the known weak year. INCREMENTAL VS WHAT IS ALREADY PROMOTED ON HL: only 9% of these events are also <=-2.5-sigma / >=1-sigma-volume days, which is the trigger of the promoted 'Hyperliquid Daily Volume-Confirmed Down-Shock Absorption Panel'. The other 91% (751 events) still net +1.04% next day, so this panel trades mostly different days. It also differs from the HL NR7 panel, which requires a prior compression day. A BTC-beta-hedged version was checked and rejected in the pre-study: the hedged next-day net is only +0.39% per event because most of the rebound is market-wide, so the hedge would strip the edge. The hypothesis therefore stays unhedged, as in the source rule. EXECUTION: Hyperliquid is a 24/7 venue, so 'buy at the event close' is a market order at the signal bar's close. That is the registry's use_equivalent for next-bar-open fills, and it is the same price on a continuous market. The exit is a market order at the close of the bar hold_days later. No stop order is needed; risk route (b) applies (time exit plus capped sizing). UNIVERSE (point in time): HL perps ADA, ATOM, AVAX, BCH, DOGE, DOT, ETC, FIL, LINK, LTC, NEAR, SOL, XRP, AAVE, UNI, CRV, COMP, SNX, MKR, TRX. The selection rule is 'liquid USD-quoted HL perps whose catalog daily history starts before 2021-01-01'. A leg becomes eligible only after atr_len+60 bars of its own history, and the developer must gate each leg with is_listed()/universe_at() so a leg is never entered outside its listed span. BTCUSD.HYPERLIQUID is not traded and not needed.

Original Research Lead hypothesis. It is a cross-venue, BTC-hedged event panel built on a signal the corpus has not used: where an alt's leverage is being traded. Each UTC day, for each of 25 alts, compute Hyperliquid's share of the coin's perp quote volume, share = HLvol / (HLvol + BinanceUSDMvol), where quote volume = base volume x close on each venue's own 1-DAY bar. Then compute z = (log share - 60-day mean of log share) / 60-day std, using the prior 60 days only. Mechanism: Hyperliquid is the on-chain, KYC-free, high-leverage venue. Its flow is dominated by crypto-native leverage chasers and visible whale/vault positioning, which copy-traders and liquidation hunters then target. When a coin rallies AND its leverage migrates to HL far beyond normal, the rally is being financed by marginal on-chain longs, not by the broad CEX market. That positioning is crowded and transparent (HL positions and liquidation levels are public), so it unwinds over the following days. The idea is not a price-shape fade: the control in the pre-study (up-days of similar size WITHOUT a share spike) shows almost no drift. PRE-STUDY (Research Lead, catalog 1-DAY bars, HL history from each coin's HL listing in 2023-24 to 2026-10-17, gross of costs, entry at the signal day's close; it was used to shape the rule, not to choose names). Up-day with share z >= 2.5: n = 304 name-days on 222 distinct dates, raw HL forward return -1.19% over 3 days and -3.0% over 5 days, median -1.1% at 3 days. BTC-beta-hedged (alt minus 60-day beta x BTC, beta clipped 0.3-2.5), the 5-day return is -2.64% (median -2.43%), negative in every year: 2023 -1.5%, 2024 -4.9%, 2025 -0.9%, 2026 -2.2%. Unconditional hedged 5-day drift is -0.58% overall and -0.78% after up-days (alts lagged BTC over the period), so the incremental effect of the share spike is about -1.9% per 5 days. The z >= 2.0 threshold gives -1.60% hedged (651 events), z >= 3.0 gives -2.41% (117), so the effect is monotone, not a cliff. Non-overlapping per name (5-day lockout): 222 events (~60/yr), hedged 5-day -2.12%. Dropping the best 5% of short outcomes it is still -0.76%. Control: up-days with return z > 1 and NO share spike (z < 1) drifted -0.06% over 3 days (n = 2,770). Low-share days (z <= -2.5) drifted +0.47%, which is consistent with the mechanism. Caveats stated up front: 2023 sample is small (41 events); per-name results are mixed (AVAX/APT/INJ/SUI strongly negative, DOGE/ARB/DOT slightly positive), and all names stay in. Lessons L230 and L237 are addressed: the short arm is measured on these exact names, and the event count inside the full-overlap span is ~220 non-overlapping, well above 30. COST ARITHMETIC: HL taker round trip 0.09% on the alt leg + 0.09% on the beta-sized BTC leg + ~0.05% impact on mid-caps = ~0.25% per hedged pair. Expected gross hedged capture ~2.1% (non-overlapping) minus ~0.8% alt-underperformance drift that should not be counted as edge = ~1.3%, less costs = ~1.0% net per pair, about 4x the round trip. Funding on the short alt leg is usually received after rallies (positive funding) and is a tailwind, credited by the platform's funding ledger. Execution: a decision on the daily close is a market order at that bar's close (24/7 venue: equivalent to the next bar's open, which is the platform's use_equivalent for next_bar_open_fill). The protective stop is an intrabar stop matched by the platform in backtest and paper, not an exchange-resting order.

Original Research Lead hypothesis. This is a fresh venue test (L228) of the BTC-beta-hedged idiosyncratic-selloff rebound. That family was promoted on Binance USD-M (spot-led variant: 374 trades, avg 0.61%) and on Bybit (retail venue-share variant: 598 trades, avg 0.62%). Here it trades entirely on HYPERLIQUID, which holds 13.3% of the corpus against a 20% target. The design is long-short and market-neutral, so it also adds to the under-represented long_short bucket (24%). Mechanism: when one alt falls far more over 24h than its BTC beta explains, the excess selling is coin-specific forced or impatient flow, for example HL cross-margin liquidations, auto-deleveraging and stop cascades on a thin book. Price-insensitive sellers push the alt below the level where discretionary liquidity clears, and over the next half day it partly recovers RELATIVE TO BTC. Shorting beta x BTC removes the market factor, which is what made the unhedged OKX market-shock version underpowered: it was exposed to the 2022 bear regime. SIGNAL (computed on every closed 4H bar, from past data only): beta = OLS beta of the alt's 4H log returns on BTCUSD.HYPERLIQUID 4H log returns over the last lookback_bars (baseline 180 bars = 30 days), clipped to [0.3, 2.0] and lagged one bar. Residual 24h move: r_res = ln(alt_t / alt_t-6) - beta * ln(btc_t / btc_t-6). sigma_res = std of 4H residual returns (r_alt - beta*r_btc) over lookback_bars, lagged 6 bars so the shock itself is excluded. z_res = r_res / (sigma_res * sqrt(6)). PRE-STUDY (Research Lead, point-in-time, local catalog HL 4-HOUR, 2024-03 to 2026-08, 18 names, one event per coin per hold window, entry at the signal close, net of 0.09% x (1 + beta) round trip on both legs): z_res <= -2.5 with a 12h clock gave 481 events on 164 DISTINCT event days. Mean hedged net was +0.52% per event (median +0.38%, t 3.0). The unhedged alt leg alone averaged +0.80%. By year: 2024 +0.73% (179), 2025 +0.77% (169), 2026 -0.07% (133). 2026 is flat, and that is the main honest risk. The 24h clock gave +0.50% (366 events). z <= -3.0 gave +0.83% (247 events, 76 distinct days). Using the TOTAL (not residual) 24h shock gives higher means but concentrates on only 67 distinct market-crash days, so the residual trigger was chosen for independence (L238: count distinct dates, not per-name trades). A tested and REJECTED filter, named here: requiring Hyperliquid's share of HL+Binance volume to be elevated (Bybit-style venue-share condition) was non-monotone. With share_z >= 1 the result was +0.26% (230 events); with share_z >= 0 it was +0.73%; with share_z < 0 it was 0.00%. That pattern looks like noise, not a mechanism, so the venue-share condition is excluded and no Binance data is used. No short (fade-the-pump) arm: L230, and the Bybit sibling found no upside edge. DATA OVERLAP (L237): every leg (18 alts + BTC) has HL 4H bars from about 2024-01 (HL serves the last ~5000 4H candles, about 2.3-2.7 years). Inside that overlap the pre-study counted 481 events on 164 distinct days, well above the ~30 floor. EXECUTION: decisions are made on the 4H close. On this 24/7 venue the next-bar-open fill (capability next_bar_open_fill = use_equivalent) is a market order at the signal bar's close; this substitution is stated here. Netting accounts: all open events share ONE BTCUSD.HYPERLIQUID position, so the BTC hedge is the NET sum of beta_i x notional_i over the open alt legs. It is re-sized at each alt entry and exit. No hedge mode is needed. Risk route (b): fixed time exit plus capped sizing, no position above 100% of equity.

Original Research Lead hypothesis. This is a frozen-rule venue transfer (lesson L228: a transfer is a fresh test, reported on its own) of the promoted 'Bybit Spot 4H Volume-Confirmed Capitulation Absorption Panel' (run dc5dc587: 405 trades, average trade +1.78%, Sharpe 1.26, max DD 15%). The source rules come over unchanged: 540-bar (90-day) z-scores, shock <= -3 sigma, log-volume >= +1.5 sigma, 6-bar per-coin cooldown, 12-bar (48h) time exit, 10% of equity per leg, at most 5 open, most negative shock filled first. The destination is HYPERLIQUID perps. That venue is 13.3% of the corpus against its >=20% target, and this mechanism has never run there at 4H. The mechanism follows L224/L230: the long side of a shock-absorption event, flat most of the time, fixed-clock exit. A forced or panic seller (on Hyperliquid, often an on-chain liquidation of a levered long) leaves an unusually large 4H drop on unusually high volume. Patient liquidity takes the other side and is paid by a partial rebound over the next 1-2 days. Why Hyperliquid, and why this should differ from the source: HL liquidations are executed on-chain by the liquidator/HLP vault against the book, transparently and at size. The flow is mechanical, with no information behind it, so the reversal premium should exist there too. But L228 shows that transfers often fail (OKX wide-range continuation, OKX upside-jump). The test is therefore registered as a fresh test with its own predictions, not as support carried over from Bybit. Universe, fixed by a data rule before looking at returns: every non-BTC/ETH alt from the promoted OKX/Bybit capitulation and NR7 universes whose HYPERLIQUID 4-HOUR history in the catalog starts on or before 2024-04-30. That gives ADA, AVAX, DOGE, DOT, LINK, LTC, NEAR, ATOM, FIL, SOL, XRP, BCH, ETC, UNI and AAVE. Names weak in the pre-study (SOL, UNI) stay in. Every leg checks is_listed() before entry. HL timeframe limit: 4H is the finest HL interval with more than 2 years of history (catalog HL 4H bars run 2024-01-22 to 2026-10-18, about 6,000 bars per name). After the 540-bar warm-up, signals start 2024-04-21, roughly 2.5 years of tradable history. PRE-STUDY (catalog HL 4H bars, z-scores from prior bars only, entry at the signal close, exactly the frozen rule including the 5-slot cap and cooldown, net of 0.09% round trip + 0.05% impact, funding NOT included). 256 trades, average net +1.99%, median +1.22%, 60% winners, PF 2.04. There were 114 distinct entry days (L238: the honest sample size; the per-day mean was about 0.8% in an uncapped variant). By year, average net per trade: 2024 +3.2% (81 trades), 2025 +2.3% (78), 2026 +0.70% (97). The edge is DECAYING in 2026, and the holdout (last ~20%, about 6 months of 2026) will sit in that weaker regime. The 5 best trades are 27% of P&L. Excluding the two best entry days, the average is +1.53%. The simple-sum equity at 10% per leg returned about +51% with about a 10% drawdown. Lesson L237 check: the overlap where all legs have bars covers the whole window, and the 114 event days are well above the ~30 minimum. Cost arithmetic: HL round trip is 0.09% taker plus about 0.05% impact. A long holding 48h pays about 0.06% funding at a typical +0.00125%/h (more in froth). Total cost is about 0.20% per trade. A conservative 50% haircut on the pre-study's +2.1% gross gives about 1.0% per trade, about 5x cost. Even the 2026-only figure (+0.70% net) is about 3.5x cost. Execution: entry is a market BUY at the signal 4H bar's close. On a 24/7 venue that is the next bar's open (registry next_bar_open_fill -> use_equivalent: market order at the signal bar's close). The exit is a market SELL after 12 closed 4H bars, on a time clock only, with no stop and no target. Risk control route (b): a time exit plus capped sizing at 10% notional per leg and at most 5 legs (50% gross, leverage setting 1x, no single position above 100% of equity). Per-leg volatility contribution is about 10% x ~90% alt vol = ~9% annualized, well inside the 120% cap. Expected strategy vol is about 15-25%. Pre-registered comparisons: (1) lesson L170, the strategy's max drawdown against equal-weight buy-and-hold of the same 15 HL perps over the same window; (2) results per year and with and without the two largest crash days (2024-08-05, 2025-10-10) reported separately; (3) funding paid per trade from the HL funding feed. Known limitation, stated up front: HL history has no 2022-style sustained bear market. The OKX daily down-shock transfer (closed underpowered 2026-10-18) lost 12% in 2022, so this test cannot measure the bear-regime risk.

Original Research Lead hypothesis. It is intraday session momentum on spot gold, a venue with 17 of 4,874 experiments (0.3%). Literature anchor: Gao, Han, Li and Zhou (JFE 2018, 'Market intraday momentum') and Baltussen, Da, Lammers and Martens (JFE 2021, 'Hedging demand and market intraday momentum'). They find that the return accumulated earlier in a trading day predicts the return of the later part of the same day across equity, bond and commodity futures. They attribute it to late-informed and hedging flows that trade in the direction of the day's move, plus gradual information incorporation. In gold the overnight leg is a distinct market: Asian physical and Shanghai-session buying, then London OTC price discovery. The New York session (COMEX, ETF creations and redemptions, macro desks) is where Western flow arrives and extends that move. Rule: at the close of the 1-hour bar ending 08:00 New York, compute the overnight log return R_on from the session open (first bar opening 18:00 NY) to 08:00 NY. Scale it by the standard deviation of the previous N sessions' overnight returns, z = R_on / sd_N. If z >= k, go long; if z <= -k, go short. The fill is at the next bar's open (08:00 NY; session venue next_open_fill, supported). Exit at the open of the bar starting 16:00 NY (decided at the 15:00-16:00 bar close), so the position never crosses the 17:00 NY financing cut and pays no overnight financing. On early-close sessions, exit at the open of the session's last bar (session_calendar helpers). Weekend and holiday gaps are never held. Pre-study on local 1-hour data, US session (08:00->16:00 NY) mean conditional on the overnight z: spot XAUUSD 2003-2008 (1,258 sessions; local spot-gold history currently ends 2008-03, and the Dukascopy backfill to 2026 is collected on demand): z>0.5 +19.1 bp (n=472), z<-0.5 -5.0 bp (n=304), unconditional correlation 0.08. Binance XAUUSDT perp 2025-12 to 2026-10 (310 sessions): z>0.5 +8.5 bp (n=74), z<-0.5 -28.4 bp (n=64), correlation 0.20. Bybit XAUUSDT perp 2026-03 to 2026-10 (224 sessions): z>0.5 +12.5 bp, z<-0.5 -12.8 bp, correlation 0.09. Both arms, separately, had the predicted sign in all three samples, and the short arm was weakest in the 2003-08 bull market (L230: the short arm is pre-registered separately; the developer must tag each trade's side so the analyst can report the arms separately). Fee arithmetic: the expected gross capture is about 10-15 bp per trade against a SPOTMETAL round trip of about 1.5 bp spread and zero financing, so net is about 8-12 bp, above the analyst's 0.05% SPOTMETAL floor. This same mechanism died on crypto (BTC Asia-session carry-over: fee_edge at 0.11% average; ETH COIN-M US-session carry-over: fee_edge) because 0.06-0.10% round trips ate it. Gold's 1.5 bp cost is the reason to test it here, not a cosmetic venue swap. Incremental-signal check (L217): the backtest must also report the unconditional 'always long the US session' arm on the same days (2003-08 about +10 bp, 2025-26 about -6 bp). The claim is that conditioning on the overnight sign adds expectancy over that baseline. Risk control route (b): signal and time exit with capped sizing, at a fixed 1.5x notional. Gold daily std is about 1%, so the position contributes about 25% annualized vol, under the profile's 120% cap. A fixed intrabar protective stop at 1.5% adverse from entry (platform-matched stop, supported) caps a trade at about 2.25% of equity, gaps excepted. The stop is risk control, not a signal.

Original Research Lead hypothesis, not seeded. This is a frozen-rule out-of-universe and out-of-venue breadth test (L227/L228: a transfer is a fresh test) of the promoted 'Capitulation-Close Next-Day Reversal on OKX Spot' panel (472 trades, average trade 1.50%, Sharpe 1.10). That panel is promoted on 12 OKX spot names. This test keeps its rule and clock frozen and runs it on 20 Binance spot names that neither it nor the two in-flight Binance spot panels use, with history back to 2019 that covers the 2019 chop, the 2021 top and the full 2022 bear. EVENT: a completed 1-DAY bar whose high-low range is at least range_mult x the mean range of the previous atr_len bars (event bar excluded), AND whose close-location value CLV = (close - low)/(high - low) is <= clv_max. It is a range-expansion day with sellers still in control at the close. ACTION: buy at that bar's close and sell at the close hold_days bars later. MECHANISM: a wide day that closes on its low is the footprint of forced, price-insensitive selling (stop cascades, liquidations of levered longs on the perp venues spilling into spot, panic market sells) that ran right up to the UTC close. On spot that flow is exhausted at the print, liquidity providers who widened during the cascade come back, and the next session mean-reverts part of the overshoot. HOW IT DIFFERS FROM BINANCE SPOT WORK ALREADY IN FLIGHT: (1) the NR7 failed-downside-break panel (DOT/ATOM/FIL/ETC/BCH/UNI/NEAR/TRX/XLM/AAVE) needs a prior compression day; (2) the 4H capitulation-absorption panel (EOS/TRX/XTZ/VET/ETC/ZEC/XLM/NEO/DASH/XMR/ONT/MATIC/QTUM/ALGO/BCH) and the regime-gated daily down-shock panel trigger on a <=-k-sigma close-to-close RETURN plus volume. On the OKX source only 32% of capitulation-close events were also <=-2.5-sigma return days, so this is a close-location trigger, not a return shock. The abandoned Effort-Without-Result panel needed a SMALL body on high volume; this needs a WIDE range closing at its low, the opposite candle. UNIVERSE (pre-registered rule, not chosen by performance; written before any return was examined on these names): every Binance spot USDT pair whose daily bars start on or before 2020-06-30 and that has at least 150 daily bars in calendar 2020, ranked by total 2020 quote volume (close x volume). Excluded: BTC, ETH, BNB (concentration and exchange token), stablecoins, fiat and wrapped/gold tokens, leveraged UP/DOWN tokens, and every name already used by the OKX source panel (ADA ATOM AVAX BCH DOGE DOT ETC FIL LINK LTC SOL XRP) or by the two Binance spot sibling panels (UNI NEAR TRX XLM AAVE EOS XTZ VET ZEC NEO DASH XMR ONT MATIC QTUM ALGO). The top 20 by 2020 quote volume are: BAND, OMG, ZIL, WAVES, THETA, IOST, KAVA, BAT, LEND, ICX, COMP, REN, BTT, ZRX, WRX, ERD, TFUEL, IOTA, ANKR, KNC. Several later delisted or migrated (LEND and ERD in 2020, BTT in 2022, WAVES, OMG and REN later). They are kept on purpose to limit survivorship bias. Set point_in_time_universe so each leg trades only while is_listed() is true, never opens a position within 3 bars of its delisting, and needs at least atr_len+1 prior bars with volume > 0; the base template closes a position at a delisting at the last price. RESEARCH LEAD PRE-STUDY (Binance spot daily bars, close-to-close, events non-overlapping per name with a 3-bar refractory, net of the 0.20% spot round trip; used to size the prediction, not to select the rule). Frozen rule range_mult 1.5, atr_len 20, clv_max 0.15: n = 955 events on 434 distinct entry dates (L238: ~2.2 names per date, so the evidence is the 434 dates, not 955 trades). Net 1-day forward return mean +1.48%, median +1.07%; 3-day mean +0.81%, so the 1-day clock of the source is kept. Equal-weight per-date mean +0.91%, t = 1.62 over 434 dates. By year (events, net 1-day mean): 2019 63/+0.12%, 2020 116/+5.08%, 2021 103/+2.79%, 2022 225/+1.77%, 2023 134/+0.18%, 2024 84/-0.97%, 2025 117/-0.18%, 2026 104/+1.86%. HONEST CAVEAT: the edge is concentrated in 2020-2022 and was negative in 2024 and slightly negative in 2025, so the walk-forward and the 2025-26 holdout may well come back non-significant. That decay is exactly what this breadth test should expose, and a non-significant result is an informative answer for the promoted OKX panel. Approximate capped book (8% of equity per event, at most 8 concurrent, 1-day hold): 885 trades, total about +121%, max drawdown about 16.8%; yearly 2019 +0.5%, 2020 +38.9%, 2021 +24.3%, 2022 +18.8%, 2023 +0.2%, 2024 -5.7%, 2025 -1.5%, 2026 YTD +13.9%. The equal-weight buy-and-hold of these names drew down well over 80% in the same span (L170/L224 comparison for a long-only design). EXECUTION: daily bars only. next_bar_open_fill has status use_equivalent: entries and exits are market orders at the signal bar's close, which on this 24/7 venue is the next bar's open price. No stop orders. Each name is traded independently, at most one position per instrument; this is an event panel, not a ranked or neutral book. Please report results split by (a) whether the event day was also a <=-2.5-sigma return day and (b) calendar year, so the analyst can see the incremental part and the decay. Parameters are the OKX source's values, frozen as the baseline.

Original Research Lead hypothesis. It is a frozen-rule, disjoint-universe replication (lesson L228: a transfer is a fresh test) of the turn-of-the-month (TOM) flow window. That window has been promoted twice on USEQ: the SPY TOM rule (paper, 401 trades, average trade 0.25%) and the emerging-market country-ETF TOM basket EWZ/EWY/EWT/FXI/INDA (paper, 1,379 trades, average trade 0.44%, Sharpe 0.61). Mechanism (Lakonishok-Smidt 1988; McConnell-Xu 2008, who found the effect in 30 of 35 countries): month-end pension, payroll and 401k contributions, plus index and mandate rebalancing, land in the last session of the month and the first 3 sessions of the next. The new universe is the four DEVELOPED-market country/region ETFs in the platform universe, selected by a rule: every non-US developed-market country or region ETF in config/us_equities_universe.txt. That gives EWG (Germany), EWJ (Japan), EWU (UK) and EFA (EAFE). EWG/EWJ/EWU have daily bars from 1996-03 and EFA from 2001-08. None of them overlaps the SPY or EM runs. These are fixed listed ETFs that all still trade, so there is no point-in-time membership issue. A leg joins the basket from its first bar. Execution: USEQ is a session venue, so a decision made at session N's close fills at session N+1's OPEN (next_open_fill, supported). The rule is: at the close of the second-to-last NYSE session of the month, order the basket; it fills at the open of the month's LAST session. At the close of the 3rd session of the new month, order the exit; it fills at the open of the 4th session. That holds 4 sessions open-to-open. Trading days and month boundaries come from the platform session calendar (src.strategies.helpers.sessions), never hard-coded dates. PRE-REGISTERED PRE-STUDY (fixed rule, catalog open prices, NYSE calendar, 1996-03 to 2026-10). 367 distinct month windows (the independent unit, per L238; the per-name trade count of about 1,400 overstates the evidence). Equal-weight basket gross window return: mean +0.215%, hit rate 0.59, t = 1.72. Per-session drift is 0.054% inside the window vs 0.012% outside it. Sub-periods: 2010+ mean +0.076% over 201 months; 2018+ mean +0.227% over 105 months. The effect is real but weaker than on the EM basket. The non-TOM drift is positive, so a short arm outside the window would lose money (consistent with L230); the design stays long-or-cash. Incremental test (L217): the comparator is the same basket held only on random 4-session blocks, whose expected return is 4 x 0.012% = 0.05%. The window should add about +0.16% per month over that. COSTS: USEQ has zero commission; the spread plus impact is about 0.02-0.05% per round trip on these liquid ETFs. The 0.215% gross per window is about 4-5x that cost.

Original Research Lead hypothesis. It is the weeknight counterpart and economic CONTRAST to the promoted 'Closed-Market Overreaction in OKX Tokenized US-Stock Perps' (weekend drift FADE, 73 trades, avg 0.26%). Bybit lists 24/7 USDT perpetuals on US stocks and ETFs (AAPLUSDT.BYBIT, NVDAUSDT.BYBIT, QQQUSDT.BYBIT, ...). They keep trading while the NYSE is shut. On WEEKENDS nothing in traditional markets is open: perp market makers cannot hedge, flow is thin and crypto-native, and the drift is noise that the Monday open fades. That is the promoted strategy. On WEEKNIGHTS the situation differs in two testable ways. (1) Information arrives: after-hours earnings and guidance, Asian and European sessions, and macro releases before the US open. (2) The drift is hedgeable and anchored: CME Globex index futures (ES/NQ) trade almost round the clock, and US pre-market trading opens at 04:00 ET. So a weeknight perp move should reflect real repricing that the cash session then CONTINUES (gap-and-go after news gaps) instead of fading. Hypothesis: the sign of the weeknight closed-market drift predicts the perp's return over the following cash session. POINT-IN-TIME PRE-STUDY (run 2026-10-19 by the Research Lead on local 1H closes, gross, before costs; EDT-only window with fixed UTC anchors: prior close = 20:00 UTC bar, decision = 13:00 UTC bar, i.e. 30 min before the 13:30 UTC open; Mon-Thu closes only, so no weekends). Define N = log(perp at decision / perp at prior NYSE close). BYBIT, 15 names, 103 weeknights, 1,431 name-nights (Apr 21 - Oct 15 2026), FOLLOWING sign(N): |N|>1.0%: 597 events on 101 nights, +0.25% to the NYSE close (t 1.8). |N|>1.5%: 391 events on 95 nights, +0.17% after 1h, +0.35% after 3h, +0.49% to the close (t 2.6, hit 54%). |N|>2.0%: 266 events on 84 nights, +0.36% after 1h, +0.54% to the close (t 2.3). BINANCE stock perps, as an independent venue check (16 names, 127 weeknights, Mar 10 - Oct 15): same sign, weaker. |N|>1.0%: +0.23% to the close (t 1.9); |N|>3%: +0.47% after 1h (t 2.4). The continuation is mostly market-wide: the cross-sectionally demeaned version is about +0.1-0.3% (t 1-2). So the strategy is a per-name follow and is NOT market-hedged. Honest caveats: (a) I computed both the fade and the follow side before choosing the follow side. The pre-study is in-sample over the SAME ~6 months that the walk-forward will use, so it is a motivation, not independent evidence. The only fresh evidence will be the holdout and paper. (b) About 100 distinct nights and 6 months of history is short. Macro nights move many names together, so the effective sample is closer to the number of nights than the number of trades (L238). (c) The weekend fade and this weeknight follow are opposite rules on the same instrument class. Both being right is exactly the falsifiable claim that hedgeability and information flow, not the closed market itself, decide the sign. Universe rule (fixed before any test, no outcome selection): every Bybit US-stock/ETF USDT perp whose local 1H history starts on or before 2026-05-15. That gives AAPL, AMZN, NVDA, TSLA, MSFT, META, GOOGL, COIN, MSTR, HOOD, PLTR, INTC, QQQ, CRCL, SPY, AVGO, ORCL (17). NFLX (listed 2026-07-06) is excluded by the rule. Set parameters['point_in_time_universe']=true so each leg trades from its own first bar. Funding on Bybit stock perps is charged by the engine where available. Execution (capability registry): only BYBIT is traded (24/7). Decisions are taken on 1H bar closes and filled by a market order at that bar's close; on a 24/7 venue that is the next bar's open (next_bar_open_fill: use_equivalent, substitution named here). The protective stop is an intrabar reduce-only stop-market matched by the platform (intrabar_stop: supported). It is not exchange-resting, and the idea does not need offline exchange protection. NYSE timing comes from src.strategies.helpers.sessions on venue 'USEQ' (session bounds, previous/next trading day, early closes), so EST/EDT shifts, half-days and holidays are handled with nothing hard-coded. USEQ bars are neither traded nor subscribed. Cost arithmetic: Bybit taker round trip is about 0.11% plus modeled impact. The pre-study's gross capture at the 1.5% baseline is +0.49% to the close. I expect roughly half of that out of sample (about +0.25-0.30% gross), or about +0.15-0.20% net per trade, which is about 2x the round trip. This is thin, which is why the prediction below is modest.

Original Research Lead hypothesis. The event is MARKET-WIDE forced selling, not a single-name shock. At each UTC daily close, compute every eligible name's daily log return divided by the std of its previous 30 daily log returns (z). Breadth is the fraction of eligible names with z <= -2. When breadth >= 30% (at least 10 eligible names), buy an equal-weight basket of EVERY eligible name at that close, including names that did not shock. Hold for exactly 2 daily bars. The trade is flat about 95% of the time. Venue: OKX USDT swaps, under-used at 3.6% of experiments. The rebound measured below is a market-level move, so a short hedge would remove it (see rationale). Execution substitution (next_bar_open_fill = use_equivalent): entries and exits are market orders at the signal bar's close, which on 24/7 OKX is the next bar's open. PRE-STUDY (fixed rule, gross of costs, OKX 1-DAY bars 2020-01 to 2026-10, the 35 names below plus BTC). Non-overlapping breadth events at >=30%: 79, or about 12 a year, spread over every year (2020: 4, 2021: 12, 2022: 18, 2023: 13, 2024: 12, 2025: 15, 2026: 5). The equal-weight basket's 2-day forward return averages +1.91% (median +1.61%, hit rate 63%). Mean by year: 2020 +3.7%, 2021 +5.3%, 2022 +0.7%, 2023 +1.2%, 2024 +1.6%, 2025 +1.8%, 2026 -0.4%. Removing the 5 best events leaves +1.11%. At a 20% threshold there are 96 events averaging +1.71%; at 50%, 55 events averaging +1.44%. A 1x-basket equity path net of 0.10% per event compounds to 3.46x with a 21.6% max drawdown, hence 0.7x gross. Incremental expectancy pre-registered (L217 / L238 style): names that did NOT shock on an event day still return +2.09% over 2 days (1,020 name-events), so the signal is the market state rather than the name. Per-name <=-2-sigma shocks on non-breadth days return only +0.38% and are negative in 2022, 2024 and 2025. The breadth condition is therefore what separates an event worth trading from one that is not. This is also the answer to the closed OKX per-name down-shock panel, whose 2022 bear leg drove a 23% drawdown: the breadth version was positive in 2022. A BTC-beta hedge was tested and rejected: hedged 2-day capture was only +0.21% on per-name shocks, because the rebound IS market beta. So the trade is long-only by construction, with a capped gross. Distinct entry dates (L238): about 79 over the history. The per-name trade count (~2,400) overstates the evidence and must not be read as the sample size. Survivorship caveat: OKX keeps no delisted swaps, so the 35 names are survivors listed before 2021. The point-in-time rule below limits but cannot remove this.

Original Research Lead hypothesis. It is a dollar-neutral, two-leg US-listed ETF spread driven by an EXTERNAL signal, the US dollar's trend, measured by the euro-currency ETF FXE. Mechanism, with instruments removed: a weakening dollar loosens financial conditions for emerging markets and a strengthening dollar tightens them. EM corporates and sovereigns carry large dollar-denominated debt, cross-border bank lending and portfolio flows into EM follow the dollar with a lag, and EM equity returns in USD also carry currency translation, which trends (currency momentum). This is the 'dollar as a global risk factor' channel: Bruno and Shin (2015), and Avdjiev, Bruno, Koch and Shin (2019, IMF Economic Review). So EM equities should outperform US equities for weeks after a dollar decline and underperform after a dollar rally. EUR/USD is used as the dollar proxy because the euro is about 58% of DXY, and FXE trades on the same venue and session as the traded legs, so the signal is point-in-time with no publication lag. INSTRUMENTS: EEM.USEQ (iShares MSCI Emerging Markets) and SPY.USEQ are traded. FXE.USEQ (Invesco CurrencyShares Euro Trust) is a signal-only leg and is never traded. All three have 1-DAY bars in the catalog; the joint history starts at FXE's 2005-12-12 listing, about 20.8 years. Primary instrument: EEM.USEQ. extra_instruments: SPY.USEQ, FXE.USEQ. RULE: on the LAST trading session of each week (use the session helpers, never a hard-coded weekday, so holiday weeks work), compute s = ln(FXE_close_t / FXE_close_{t-lookback}), with lookback = 20 sessions by default. If s > +deadband, target LONG EEM / SHORT SPY. If s < -deadband, target LONG SPY / SHORT EEM. Inside the deadband, keep the current book. With the default deadband of 0 the book is always in one state after warm-up. Orders are decided at that session's close and fill at the next session's open, which is the platform's supported next_open_fill on USEQ. Rebalance only when the target state flips. Within a state, no weekly re-sizing, so turnover stays low. SIZING / RISK CONTROL route (b): signal exit plus capped sizing. Each leg is 50% of current equity at entry (whole shares, floor), so gross is about 100% and net about 0. No position exceeds 50% of equity, inside 2x Reg-T margin. The exit is the signal flip. There is no stop. Pre-study volatility of this book is about 8% a year, well inside the conservative profile's 60% band. Leverage is 1.0x gross. RESEARCH LEAD PRE-STUDY (pandas, close-to-close, weekly decision at the week's last session, held from the next session, no costs, 2006-2026). It was used to choose this idea over three others, not as acceptance evidence. Four other macro-impulse spreads were screened and DISCARDED in the same pass: breakeven-inflation momentum to XLE+XLB vs XLU+XLP (Sharpe -0.31 to 0.03), real-yield momentum to QQQ vs XLE+XLF (negative), 2s10s steepening to XLF vs XLU (about 0.1), and VIX change to IWM vs SPY (about 0). A cross-country term-spread rotation was also discarded (t of about 0). The FXE 20-session rule gives Sharpe 0.49 on the full-notional spread from 2006 (15 of 21 years positive) and 0.79 after 2015. The lookback neighbourhood is positive throughout: FXE 15/20/30/40 sessions give 0.20/0.49/0.19/0.27, and FRED EUR/USD gives 0.15 to 0.48 over 15-60 sessions. Since 2008 there are 225 position episodes (111 long-EEM), with a mean spread return of +0.70% per episode before costs and a 56% win rate. At 50/50 notional, annual volatility is 8.1% and max drawdown 19.0%. INCREMENTAL-INFORMATION CONTROL (lesson L217, pre-registered): the same instruments, clock and lookback, signed by the EEM/SPY spread's OWN 20-session momentum instead of FXE, give Sharpe 0.16 over the full history and -0.37 after 2015 (10- and 40-session versions are about 0 or negative). The FXE signal also stays positive (0.29) on the weeks when it DISAGREES with the spread's own momentum. The external dollar signal therefore carries information beyond spread momentum. The developer should report this control beside the result: require_fx_signal=false switches the sign source to the spread's own momentum, as a shadow comparison only. DISTINCT EVIDENCE UNITS (lesson L238): about 1,080 weekly decision dates and about 240 position episodes (about 480 leg trades) over 2006-2026. The two legs of an episode are one bet, so the episode count is the honest sample. That is far above the roughly 30 events lesson L237 asks for, and every leg has bars over the whole span. FEE ARITHMETIC: USEQ has zero commission. Spread plus impact is about 0.02-0.05% round trip per leg, so about 0.1% of one-leg notional per episode for both legs. The expected gross capture is about +0.70% of spread per episode, or about +0.35% per leg, which is over 3x the cost. The analyst floor on USEQ is about 0.05% per trade. DIVERSIFICATION: USEQ is not a crypto venue and adds a macro-FX driver that no deployed strategy uses (the promoted NFCI sector spread uses credit and financial conditions, not the dollar). It is long-short and dollar-neutral (long-short is 24% of the corpus) and is neither BTC nor a momentum/NR7/shock-absorption clone. The abandoned 'Month-End FX Hedge-Rebalancing FXE' hypothesis traded FXE itself on a month-end calendar. This hypothesis uses FXE only as information about the dollar cycle and trades EM vs US equities on its trend.

Original Research Lead hypothesis. It is a fresh, frozen-rule transfer test (L228) of the promoted Bybit strategy 'VolSpike ATR-Band Breakout (FMZ 488520, ianzeng123) on a 6-Name Bybit Alt-Perp Panel' (run b22697a9: 166 trades, avg trade 1.20%, Sharpe 0.70, paper_stage). The rule set comes from ianzeng123's FMZ Pine strategy 'VolSpike ATR RSI Strategy with ATR Bands' (Source URL: https://www.fmz.com/strategy/488520). The rules are kept exactly as the Bybit run wrote them. Only the venue and the names change, and every name is new. RULES, evaluated per instrument on each closed 4H bar: (1) VolSpike: the bar's volume is greater than the SUM of the previous 7 bars' volumes. (2) RSI(14, Wilder) filter: 20 < RSI < 80. (3) Wick filter. LONG requires close > open and upper-wick/body = (high-close)/|close-open| <= 2.0. SHORT requires close < open and lower-wick/body = (close-low)/|close-open| <= 2.0. A zero body is replaced by 0.0001, as in the source. (4) ATR bands from ATR(14, Wilder): upper = close + 2.5*ATR, lower = close - 2.5*ATR. A long takes its stop at the lower band and its take-profit at close + 1.0*(upper-close). A short takes its stop at the upper band and its take-profit at close - 1.0*(close-lower). Pine semantics are preserved. A same-direction signal while in a position re-arms the stop and target from the new bar's close and ATR, and adds no size. An opposite signal closes the position and reverses it. UNIVERSE, fixed by rule before any return was examined: OKX USDT swaps of legacy (pre-2021-listed) liquid alts that have OKX 4H history in the catalog starting before 2021-08, excluding every name in the Bybit panel (HBAR, XLM, ALGO, ETC, ATOM, AAVE) and BTC/ETH/SOL for corpus concentration. That gives DOT, UNI, LTC, BCH, NEAR, FIL, TRX, LINK, ADA, XRP, AVAX, SUSHI, CRV, COMP, SAND and AXS. DOGE, MANA and XTZ were in the candidate list but have only weeks of OKX 4H bars in the catalog, so they were dropped by the history rule, not by their results. Point-in-time: a leg trades only while is_listed() and at least 30 days after listed_since(). OKX keeps no delisted swaps, so residual survivorship bias remains and is stated, not fixed. Each name runs the identical rule independently, with one netted position per instrument. This is an event panel, not a ranked or neutral book. EXECUTION SUBSTITUTIONS (registry): Pine strategy.entry fills at the next bar's open. On this 24/7 venue that is written as a market order at the signal bar's close (next_bar_open_fill: use_equivalent). The stop and target trigger inside the bar via a bracket or a reduce-only stop-market plus limit (intrabar_stop: supported). When one bar touches both, the engine's adaptive OHLC ordering decides (same_bar_policy is not selectable). This is noted, not emulated. RESEARCH LEAD PRE-STUDY: an independent re-implementation on catalog bars, net of 0.10% RT, used only to check feasibility, not to accept the idea. Calibration: on the original Bybit 6-name panel the re-implementation gives 171 trades and +1.29% per trade, against the run's 166 trades and +1.20%, so it reproduces the promoted result. On this OKX panel (2020-2026) it gives about 366 trades on 250 distinct entry dates (at most 13 on one date), mean +1.0% net, median +4.3%, win rate 58%, t about 2.3. The edge is concentrated in 2023-2026 (+1.9%, +2.5%, +1.3%, +0.4% per trade). 2020-2022 were negative or flat on few trades (24, 28 and 33 trades; -0.7%, -1.8%, 0.0%). Per L230 the arms were measured separately. The SHORT arm averaged +1.9% on 167 trades and was positive every year from 2022, but lost in the 2020-21 bull. The LONG arm averaged +0.3% on 200 trades. Both arms are kept because the source trades both. The bull-year weakness of the short arm is the main pre-registered risk. A Hyperliquid 4H transfer was also measured: 20 names, Jan 2024 onward, 630 trades, only +0.35%, t 1.3. It was rejected as the destination because its history is short and its result weak; this is reported here, not hidden. Venue rationale: OKX is 3.6% of the corpus (target >=5%) and long_short is 24% of the corpus. No VolSpike rule has been tested outside Bybit. Risk control route (a): a 2.5-ATR(14) 4H stop is about 5-7% of price, so 10% notional x about 6% is about 0.6% of equity at the stop, inside the 2% cap. At most 16 legs can be open at 10% each (160% gross, about 1.6x on OKX margin, under the profile's 5x). Each leg contributes about 10% x roughly 90% annualized volatility, which is about 9%, far under the 120% per-position cap.

Original Research Lead hypothesis (not seeded). An event-driven, long-only, unlevered spot panel on OKX_SPOT, which is 0.2% of the corpus. EVENT ('capitulation close'): a completed daily bar whose high-low range is at least range_mult x the mean range of the previous atr_len bars (the event bar excluded) AND whose close-location value CLV = (close - low)/(high - low) is <= clv_max. In words, a range-EXPANSION day on which sellers were still in control at the close. Buy that close and sell at the close hold_days bars later. MECHANISM: a wide day that closes at its low is the footprint of forced, price-insensitive selling (stop cascades, liquidations of levered longs on the perp venues, panic market sells) that ran right up to the bar close. On spot, that flow is exhausted at the print. Liquidity providers who widened during the cascade come back, and the next session mean-reverts part of the overshoot. This is a CLOSE-LOCATION trigger, not a return or volume shock. HOW IT DIFFERS FROM WHAT THE FACTORY HAS ALREADY TESTED: (1) The down-shock absorption panels (Hyperliquid daily, OKX swap daily, Bybit spot 4H) trigger on a <=-k-sigma close-to-close return plus volume. In the pre-study only 32% of capitulation-close events were also <=-2.5-sigma return days. The NON-shock 68% (330 events on 166 distinct dates) still earned +1.43% net over 3 days, so the trigger carries expectancy those panels do not capture. (2) NR7 failed-downside-break needs a prior compression day; only 11% of these events follow an NR7 day. Excluding both shock and NR7 events leaves 289 events at +1.46% net. (3) The promoted 4H wide-range close-location CONTINUATION rule (Bybit) would SHORT these bars. On the daily horizon in spot the sign is the opposite: reversal. This is consistent with L230: absorption works on the long side, and the short mirror is not assumed. RESEARCH LEAD PRE-STUDY (OKX_SPOT daily bars, 2020-2026, 12 names, events non-overlapping per name with a 3-bar refractory, net of 0.20% spot round trip; used to shape the rule, not to accept it). Baseline rule range_mult 1.5, clv_max 0.15: n=488 events on 197 distinct dates. Net forward return after 1 day: mean +1.58%, median +1.25%. After 2 days: mean +1.54%. After 3 days: mean +1.60%, median +0.66%. After 5 days: mean +0.99%. Almost all the edge arrives on day 1, so the baseline hold is 1 day. 11 of 12 names are positive on the 3-day horizon; SOL is the exception at -0.61%. 5 of 7 calendar years are positive. 2022 was -2.1% and 2024 was -2.4%: the rule loses in grinding bear legs where a wide red close is followed by another. Control (any close below the prior day's low, 3-day hold): +0.50% net. Requiring a wide range and a low close adds about +1.1% per event over that no-filter baseline. Approximate capped book (8% of equity per leg, max 8 concurrent, 1-day hold, 0.1% fee per side): 494 trades, average trade +1.37% net, total return about +68%, max drawdown 8.7%, daily Sharpe about 0.9, positive in 6 of 7 years. With a 3-day hold the drawdown rises to 27%. The equal-weight buy-and-hold of the same 12 names drew down 83.5% over the same span (the L224/L170 buy-and-hold drawdown comparison for a long-only design). FEE ARITHMETIC: the expected capture is about 1.2-1.5% per trade, against the 0.20% spot round trip, so 6-7x cost. UNIVERSE (pre-registered rule, not chosen by performance): every OKX USDT spot pair with daily bars from on or before 2021-01-01 and continuous history to 2026-09, excluding BTC and ETH (corpus concentration), stablecoins, wrapped and staked tokens and OKB. The rule gives exactly 12 names: ADA, ATOM, AVAX, BCH, DOGE, DOT, ETC, FIL, LINK, LTC, SOL and XRP. Residual survivorship bias: OKX serves no bars for delisted spot pairs, so a long-history name that died before 2026 cannot be included. Set point_in_time_universe so each leg trades only while is_listed() is true and at least 30 days after listed_since(). EXECUTION: daily bars only. next_bar_open_fill has status use_equivalent, so entries and exits are market orders at the signal bar's close, which on this 24/7 venue is the next bar's open price. No stop orders are needed. Each name is traded independently with at most one position per instrument. This is an event panel, not a ranked or neutral book. Report results split by event type (also a <=-2.5-sigma return day vs not) so the analyst can see the incremental part.

Original Research Lead hypothesis. It tests a promoted mechanism on a universe and venue it has never seen. Lesson L227 asks for exactly this kind of replication. The rule is frozen from the promoted siblings: Bybit perps (742 trades, avg 0.985%), OKX swaps (940, 1.105%), Binance spot (10 pre-2020 alts) and the Hyperliquid 8-asset daily panel (413, 1.10%). Every one of those universes holds pre-2021 large and mid caps (ETH/SOL/XRP/DOGE/ADA/LINK/AVAX/BNB/LTC, DOT/ATOM/ETC/FIL/UNI/BCH/NEAR/AAVE/TRX/XLM). This test uses 20 names NONE of them traded, most of them listed 2022-2024 with a different holder base: unlock-driven supply, airdrop farmers and narrative rotation. That is a real out-of-universe breadth test, not a fifth re-run of the same price series. The venue is BYBIT_SPOT (4 of 4,861 experiments, 0.1%), CASH, long-only and unlevered, so there is no funding or perp liquidation engine in the price path. MECHANISM: the NR7 day is the narrowest daily high-low range of the last 7 days. A tight range packs resting stop-losses and breakout-short entries just under its low. A daily CLOSE below that NR7 day's low runs those orders into a thin liquidity pocket. The forced selling is spent by the close and price snaps back over the next few days. Buy that close, hold a fixed 3 daily bars, sell. UNIVERSE (pre-registered before any measurement; all 20 kept, none dropped after seeing results): APT, ARB, OP, SUI, SEI, TIA, INJ, LDO, IMX, STX, WLD, PEPE, GALA, SAND, MANA, CRV, GMX, JUP, ORDI and ENA, as Bybit spot USDT pairs. Point in time: a leg is eligible only while listed (is_listed) and at least 30 days after its own Bybit spot listing (listed_since); the developer sets parameters.point_in_time_universe=true so later-listed legs do not clamp the window. Residual survivorship: Bybit serves no bars for closed pairs. The names were chosen by liquidity and a disjointness rule, not by performance. RESEARCH LEAD PRE-STUDY (fixed rule, measured BEFORE writing, on Binance spot daily bars of the SAME 20 names as a price proxy: local Bybit spot bars cover only the last few days and are collected on demand, and spot prices of these liquid pairs track across venues to a few bp). NR7 downside-break long, 3-day hold, non-overlapping per name: n=1,594 events on 550 distinct entry dates (L238 asks for this count), mean gross +1.19%, median +0.54%, win rate 52.9%, per-event std 10.0%. 17 of 20 names positive (SUI -0.08%, TIA -0.27% and the rest +0.09% WLD to +3.06% CRV). By year: 2020 +4.45%, 2021 +3.30%, 2022 +1.74%, 2023 +0.60%, 2024 +1.17%, 2025 +0.17% (the weak year, as in every sibling), 2026 +1.59%. INCREMENTAL VALUE of the compression condition (L214/L217): the CONTROL, any close below the prior day's low with the same 3-day hold, earns +0.66% gross on these names (n=4,814), so the NR7 condition adds about +0.53% per event. CAPPED BOOK (6% of equity per event, at most 8 concurrent, 0.20% round trip): 1,365 trades, about +88% total, max drawdown 17.9%; yearly PnL positive in 6 of 8 calendar years (2019 is 5 events, -2.6%; 2025 -9.6% is the real losing year). FEE ARITHMETIC: gross capture about 1.19% minus the 0.20% Bybit spot round trip (0.10% taker per side) leaves about 0.99% net per trade, about 5x the cost; allowing another 0.3% for slippage on mid-cap books still leaves about 0.7%. HONEST CAVEATS: (1) on the Bybit sibling a beta-scaled BTC hedge cut the 3-day mean from +1.25% to +0.14%, so much of this edge is a market-wide rebound expressed through alt stop-runs; the design is deliberately unhedged and long-only, and drawdown is controlled by per-event size and the concurrency cap. (2) The median (+0.54%) is well below the mean: the edge leans on right-tail rebounds. (3) Lesson L170/L224 buy-and-hold comparison, pre-registered: an equal-weight buy-and-hold of these 20 names drew down far more than 70% in 2022 and again in 2024-25, against a predicted <=22% for this book, which is flat about 85% of the time per name. EXECUTION: daily bars on a 24/7 venue. Signals are evaluated at the daily close and filled with a market order at that close, the platform's faithful equivalent of a next-bar-open fill on a continuous market (capability next_bar_open_fill: use_equivalent). No stop orders (risk-control route b: fixed time exit plus capped sizing, no single position above 6% of equity).

Original Research Lead hypothesis, built on the textbook 'evaporating liquidity' finding (Nagel 2012, RFS: short-term reversal returns, i.e. the return to liquidity provision, are largest right after market-wide stress) applied to crypto alt perps. On a market-wide down-shock day, forced deleveraging and liquidation cascades hit some alts far harder than their normal BTC beta implies; that excess (idiosyncratic) drop is a liquidity-demand price concession, not news, and it is partly repaid over the following days. The strategy is flat until BTCUSDT.OKX prints a UTC daily close-to-close log return <= -shock_k x its trailing 60-day daily std (prior days only). At that close it ranks every eligible alt in a fixed 22-name OKX USDT-swap universe by its same-day RESIDUAL return r_alt - beta_alt x r_BTC (beta = OLS slope of daily log returns over the prior beta_lookback days, at least 60 valid days), buys the n_long most negative residuals equal-weight, and shorts BTCUSDT.OKX for beta-neutrality (BTC short notional = average beta of the bought names x total long notional). Everything closes on a fixed clock after hold_days daily bars. New triggers are ignored while a book is open, so events never overlap. Venue: OKX USDT perpetual swaps (3.6% of the corpus, under the BINANCE 49% concentration), long/short market-neutral (long_short bucket is 24%), event-driven flat-most-of-the-time design (lesson L224). Execution: orders are market orders at the signal bar's close; on this 24/7 venue that price is the next bar's open (use_equivalent for next_bar_open_fill, stated here). Universe is point in time: an alt is eligible at an event only if it has >= 90 days of OKX daily bars before the event (listed_since / universe_at); OKX keeps no delisted swaps, so residual survivorship bias remains and is stated as a risk. PRE-REGISTERED FIXED-RULE MEASUREMENT (OKX 1-DAY closes 2020-01 to 2026-10, shock_k 1.5, beta 90d, n_long 3, hold 5d, non-overlapping): 94 events; long weak-3 arm +2.04% mean 5-day return; BTC-beta-hedged book +1.82% per event on long notional (t ~1.97, hit rate 54%, worst event -17.9%), positive in 5 of 7 calendar years (2022 -0.3%, 2023 -1.2%). Per L230 the alternative short arm (shorting the 3 STRONGEST-residual alts) was measured too and is rejected: those names rose +1.44% over the same 5 days and the dollar-neutral alt-vs-alt spread had a -29% single event (a squeeze), so the hedge is BTC, not alt shorts. COST ARITHMETIC: per event the book pays ~0.10% RT on the alt legs plus ~0.10% x beta(~1.07) on the BTC leg, ~0.21% of long notional, vs +1.82% gross hedged capture, net ~1.6% per event (about 8x the cost). Per-trade average across the 4 positions per event (three alt legs ~+2.0%, BTC short leg ~-0.2%) is ~1.4% of notional, vs the OKX 0.15% floor. Risk route (b): signal/time exit with capped sizing, 15% of equity per alt leg (45% long gross), BTC short ~48%, gross <= ~1.0x, leverage 1x, no stop invented (the fixed clock is the exit); worst measured event ~-8% of equity.

Original Research Lead hypothesis, event-driven and BTC-beta-hedged. On a sharp BTC down day, alt perps fall by MORE than their normal beta to BTC. Alts carry higher retail leverage and thinner books, so the same BTC move sets off proportionally larger long liquidations in alts. That forced, non-informational selling should partly reverse over the next one to three days, relative to BTC. The trade buys the alt overshoot and shorts BTC in beta-matched size, so it bets on the alts' RELATIVE recovery, not on the market bouncing. TRIGGER: on each completed BTCUSDT.OKX 1-DAY bar (UTC close), z = that day's BTC simple return divided by the standard deviation of the previous vol_lookback (30) daily BTC returns, excluding the current day. An event fires when z <= -shock_k (baseline 2.0) and the strategy is flat. UNIVERSE (point in time): the 20 OKX USDT swaps listed below, all listed on OKX before 2021. At each event, a name is eligible only if it has at least 90 consecutive daily bars ending at the event bar (use universe_at()/listed_since()). Residual survivorship bias: OKX keeps no delisted swaps, so the universe cannot include coins that later died; this is stated, not fixed. POSITIONS: market orders at the event bar's close. On this 24/7 venue that price is the next bar's open (the platform's use_equivalent for a next-bar-open fill). Long legs: rank eligible alts by their shock-day residual (r_alt - beta_i*r_BTC, with beta_i the 60-day OLS beta) and buy the most-negative select_frac share (baseline 1.0, i.e. all eligible names), equal-weighted, with total alt notional alt_gross (0.60 of equity). Hedge leg: short BTCUSDT.OKX with notional beta_basket * alt_gross, where beta_basket is the 60-day OLS beta of the selected basket's equal-weight daily return on BTC, clipped to [0.5, 2.0]. Typical beta is ~1.2, so the hedge is ~72% of equity and total gross ~1.3x on 2x configured leverage. No single position exceeds ~0.9x equity even at the beta clip. EXIT: every leg closes at market at the close of the hold_days-th daily bar after entry (baseline 2). No new event is taken while positions are open (non-overlapping events). Risk-control route (b): time exit with capped sizing; no stop is invented. FIXED-RULE PRE-MEASUREMENT (Research Lead, daily closes, same 20 names, k=2.0, 2-day hold, non-overlapping events, 0.11% round-trip per leg charged on both alt and hedge notional): - OKX 2020-2026: 66 events. Hedged gross +0.78% per unit of alt notional (+0.95% with select_frac=0.5); net +0.54% (+0.72%); t ~1.4-1.6; hit rate 0.55-0.58. - Same prices on Binance: +0.99% gross, net +0.75%, t 1.86. - Bybit, where history only starts in late 2020 and several names list later: +0.39% gross, net +0.14%, t 0.75. - Most of the effect sits in 2020-2021 and 2024-2025, and 2026 to date is negative (-1.4% per event). Positive event-years: 4 of 7. This is a modest, fragile signal and the prediction below is set accordingly. PRE-REGISTERED COMPARISON (lesson L217 style): the backtest should also report the same events traded UNHEDGED (alt basket only). The hypothesis claims the HEDGED spread is positive after costs. If only the unhedged leg makes money, that is market beta after crashes (already covered by the promoted down-shock absorption panels), not this mechanism. EVIDENCE COUNT (lesson L238): the strategy takes ~21 positions per event, so position counts overstate the evidence. There are about 66 distinct entry dates over ~6.8 years (~10 a year), above the ~30-event floor of lesson L237. Why it is new: the earlier 'BTC-Shock Idiosyncratic Response Persistence' (Hyperliquid) bet that alts continuing in BTC's direction would keep going (momentum). This trade bets the opposite (reversion of the beta overshoot) and is market-neutral to BTC. The pipeline's Bybit volume-share capitulation rebound triggers per alt on that alt's own volume share; this one triggers on a market-wide BTC shock and trades a basket against a beta hedge. The venue (OKX, 3.6% of corpus) and the long/short structure both fill under-represented buckets.

Original Research Lead hypothesis. This is a FRESH, frozen-rule venue-transfer test (L228) of the promoted 'Cash-Financed vs Leverage-Financed Advances' strategy (run 282f0af9, BINANCE_SPOT, 286 trades, avg trade 4.85%, Sharpe 1.57, DD 7.8%). In that strategy both the signal and the traded leg came from Binance. Here every input comes from a DIFFERENT exchange, OKX: the volume-share signal is built from OKX spot and OKX USDT-swap daily volume, and only the OKX SPOT leg is traded. If the claim that cash-financed advances persist is about who is buying, and not about one exchange's data, it has to reappear on OKX's own flow mix. If it does not, that weakens the paper strategy's premise, which is useful too. Results must be reported separately from the Binance strategy; nothing here is pooled with it. Mechanism: spot_share = OKX spot base volume / (OKX spot base volume + OKX swap base volume), same coin, same UTC day. A price advance during which the spot share of volume rises unusually is being paid for in cash. Coins leave exchange float into fully paid wallets, so there is no funding clock and no liquidation overhang, and that flow persists for days. An advance carried by swap volume is leverage-financed and unwinds mechanically. Universe (rule fixed before any return was examined): every non-BTC, non-ETH OKX USDT market that has BOTH OKX_SPOT and OKX swap 1-DAY history in the local catalog starting on or before 2021-01-31. That gives 12 names: ATOM, FIL, ETC, BCH, LINK, ADA, DOT, AVAX, SOL, XRP, DOGE, LTC. Eight overlap the Binance universe and four are new (ATOM, FIL, ETC, BCH). Weak names are deliberately kept (ETC was negative in the pre-study). The 12 OKX swap instruments are VOLUME FEEDS ONLY and must never receive an order. Rules are frozen at the Binance strategy's optimized values: z-window 90 days, entry_z 1.5, trend lookback 77 days with trend_min 0, hold 5 days, inverse-volatility weights, max 11.5% per name, gross cap 92%. PRE-STUDY (local catalog, OKX 1-DAY, 2020-2026, gross, non-overlapping per name, fill at the signal close): 607 events on 415 distinct entry dates. Mean 5-day forward return +2.37%, median +0.10%, hit rate 50%. By year: 2020 +3.5%, 2021 +6.0%, 2022 -1.7%, 2023 +2.9%, 2024 +2.8%, 2025 -1.9%, 2026 +1.5%. Mean is positive on 11 of 12 names (ETC -1.0%). MATCHED CONTROL, same uptrend filter with no share condition: 2,637 events, mean +1.57%, median -0.22%. The share filter adds about +0.8% mean and +0.3% median per event. That is the incremental expectancy pre-registered here (L217). The edge is right-skewed: the mean far exceeds the median, and 2022 and 2025 (alt bear years) are negative. It is a risk-on amplifier, flat most of the time. Fee arithmetic: OKX spot taker round trip about 0.20% (0.10% per side) plus about 0.02% impact. Gross mean +2.37% leaves about +2.1% net per trade, roughly 10x cost. Even the incremental +0.8% over the control is 4x cost. Firing rate: about 100 events a year across the panel in both bull and recovery regimes, so this is not a crash-only trigger. Execution substitution: next_bar_open_fill is use_equivalent. Decisions are made on the closed UTC daily bar once all of that coin's spot and swap bars for the same close timestamp have arrived. Orders are market orders at that bar's close, which on a 24/7 venue equals the next bar's open. No stops are used and no exchange-resting order is needed. Diversity: OKX_SPOT is 0.2% of experiments and OKX 3.6%. Two venue suffixes make this cross-venue (10% vs a 15% target). No BTC or ETH, so it adds no corpus concentration. Pure OHLCV, no supplementary feed (L189). Data overlap (L237): every leg has daily bars from 2021-01-24 at the latest, giving more than 5.7 years of joint history with 607 events.

Original Research Lead hypothesis: an out-of-universe, out-of-venue breadth test of a promoted mechanism, with the rules frozen. The source is the paper-stage 'Bybit Spot 4H Volume-Confirmed Capitulation Absorption Panel' (405 trades, average trade +1.78%, PF 1.80, Sharpe 1.26, max DD 15.0%). It traded 11 coins on BYBIT_SPOT: SOL, XRP, ADA, DOGE, LINK, LTC, DOT, AVAX, ATOM, FIL and NEAR. Following lessons L228 and L224, this run tests the same rule on coins it has NEVER traded, on a different cash venue. It shares no coin, and so no price series, with the source panel. It is a fresh test, not a replication on the same data. MECHANISM: a forced or panic seller (stop cascades, a margin call met by selling spot, a fund exiting) leaves an unusually large 4H drop on unusually high volume. A patient cash buyer who takes the other side is paid through a partial rebound over the next 1-2 days. Spot cannot be liquidated and pays no funding, so this is a plain liquidity-provision bet. UNIVERSE, fixed by a data rule BEFORE any return was examined (survivorship-honest, point in time). Take the BINANCE_SPOT USDT pairs that had daily bars by 2020-01-01. Exclude BTC, ETH and BNB, stablecoins and fiat or gold tokens, leveraged UP/DOWN tokens, and the 11 coins of the source study. Rank the rest by median daily quote volume over calendar 2020 and take the top 15: BCH, EOS, TRX, XTZ, VET, ETC, ZEC, XLM, NEO, DASH, XMR, ONT, MATIC, QTUM, ALGO. Three of them were later delisted from Binance spot: XMR (2024-02), MATIC (2024-09) and EOS (2025-05). They stay in the universe. The base template closes a position at its last bar, and the strategy checks is_listed() before every entry. The selection uses 2020 data only, so trading starts 2021-01-01, with late-2020 bars used only as indicator warm-up. Set parameters['point_in_time_universe']=True so the later delistings do not clamp the window. SIGNAL, the source rule unchanged. On every closed 4H bar of each coin: shock_z = ln(close/prev_close) / std of 4H log returns over the previous vol_lookback bars (the current bar is excluded). volume_z = (ln(volume) - mean of ln(volume) over the previous vol_lookback bars) / their std. An event is shock_z <= -shock_z_min AND volume_z >= volume_z_min. After an event, the coin cannot fire again for 24h (6 bars), and it never fires while a position in it is open. EXECUTION: a market BUY at the signal bar's close. On a 24/7 venue that price is the next bar's open; the registry lists this as the use_equivalent for next_bar_open_fill. The exit is a market SELL after hold_bars closed 4H bars. Exits are on a time clock only: no stop, no target, no signal exit, as in the source. SIZING / RISK CONTROL route (b), a time exit plus capped sizing. Each position is leg_weight = 10% of current equity, with at most max_concurrent = 5 open at once (50% gross, 1x cash, no leverage). When more coins fire on the same bar than there are free slots, the most negative shock_z is filled first. Under the conservative risk profile, each leg contributes about 0.10 x ~100% annualised alt volatility = ~10%, far under the 120% per-position limit. Even a -50% gap on a leg costs only 5% of equity, under the 15% per-trade limit. PRE-STUDY: catalog BINANCE_SPOT 4H bars, 2020-06 to 2026-10. z-scores use prior bars only and vol_lookback = 180 (30 days); entry is at the signal close, with one event per coin per 24h; returns are gross. There were 1,185 events on 349 distinct entry dates (lesson L238: events cluster, so the date count is the honest sample size). The mean forward return was +0.87% at 12h, +1.71% at 24h (median +1.12%) and +1.53% at 48h (median +0.89%, 56% positive). By year, the 24h mean was positive in all seven years: 2020 +0.5%, 2021 +4.6%, 2022 +0.5%, 2023 +1.1%, 2024 +2.0%, 2025 +2.0%, 2026 YTD +0.6%. The 48h mean was positive in 6 of 7 years, all except 2020 (-0.2%). The 48h mean was positive for 14 of 15 coins; the exception is ONT at -0.5%. CAVEAT: this pre-study covers the whole history, holdout included. The holdout is therefore not unseen by the Research Lead, and the rules were frozen from the SOURCE study, not from this pre-study. COST ARITHMETIC: the round trip on Binance spot is 0.20% in taker fees plus about 0.05% modelled impact, so 0.25%. The gross 48h mean is ~1.5%. Haircut by 50% for capacity skipping and fill reality, it is ~0.75% gross, ~0.5% net per trade: 2x+ the cost floor, and about 6x the cost before the haircut. PRE-REGISTERED COMPARISONS: (1) Lesson L170: report the strategy's max drawdown next to an equal-weight buy-and-hold of the same 15 coins over the same window. That basket drew down far more than 70% in 2021-2022; the strategy should stay near 20%. (2) The traded horizon is 48h, frozen from the source. The 24h horizon is reported beside it through secondary_horizons_days, because the pre-study shows the 24h capture is as large. This is reported only; the rule is not changed. CORPUS / QUOTA: BINANCE_SPOT is 10.4% of experiments against a target of at least 15%, and lesson L192 says spot-tagged long-only ideas have the highest survival rate. All 15 coins are old-generation names that have barely been used in the corpus, so this spends unused data rather than re-mining BTC. The design is event-driven and flat most of the time, which is the lesson-L224 pattern. It is not a momentum variant.

Original Research Lead hypothesis, a Wyckoff 'effort vs result' absorption event on daily spot bars. On a day when an alt trades at least twice its normal volume but the candle body stays small (open-to-close move no more than half of the 20-day ATR), and the close sits in the bottom third of the 20-day high-low range, heavy supply met equal demand near the lows: someone absorbed the selling without letting price fall. The panel buys that name on BINANCE_SPOT and holds it for a fixed 3 days. It is NOT the down-shock/capitulation absorption family (those need a <=-2.5 to -3 sigma return; this one requires a SMALL return), NOT NR7 (range compression with no volume condition) and NOT a breakout. It is long-only on the no-leverage spot venue, which is the highest-survival venue and is under quota (10.4% vs >=15%). Venue/instrument choice: 22 alts, none of them BTC or ETH, so it adds no BTC concentration. Fixed-rule pre-measurement on these exact names (Binance spot daily bars 2018-2026, close-to-close 3-day forward return, no costs): the condition fired 343 times on 176 distinct dates. Mean forward return was +2.04% (median +1.40%, 59.5% positive) against +0.36% for all name-days, so the incremental gross expectancy attributed to the signal is about +1.7% per trade. Calendar-year means were positive in 8 of 9 years (2020 was -3.7%). Drop the range-position filter and the mean falls to +1.73% (1,251 events); tighten the body to 0.3 ATR and it is +1.11% (209 events). The edge depends on all three conditions together, so all three are tunable with ranges. Execution: the platform fills a bar-close decision on a 24/7 venue at that bar's close, which is the faithful equivalent of the next day's open; this substitution is stated here. Universe (point in time, as best the data allows): a fixed, pre-registered list of 22 Binance spot USDT pairs, chosen because they were listed before 2021-01-01, not because they survived. The base template's listing calendar closes and excludes any leg once it is delisted, and a leg trades only while is_listed() is true. Residual survivorship caveat: the list was drawn from names still in the catalog today.

Original Research Lead hypothesis, with the mechanism taken from George & Hwang (2004, Journal of Finance, 'The 52-Week High and Momentum Investing'): traders anchor on the 52-week high, so they sell into it and good news is under-absorbed until price clears the anchor. Once it does, the continuation is under-priced. Nothing in the factory corpus tests 52-week-high or all-time-high anchoring; a title/FTS search for 52-week, all-time-high and anchoring returned nothing. The design is the event-plus-fixed-clock, flat-most-of-the-time form that L224 favours, on BINANCE_SPOT, the venue L192 ranks highest. It is a different mechanism from the factory's crowded families: it is not a shock absorption, not NR7, not dual-timeframe confluence, and not a cross-sectional rebalance. Single venue, BINANCE_SPOT (CASH, 1x, 0.20% taker round trip), 1-DAY bars, pure OHLCV, no supplementary feed. Universe (fixed candidate list, survivorship handled): 36 Binance spot USDT pairs that were listed before 2021 and were large caps then, including names delisted or collapsed since: ETH BNB XRP ADA LTC LINK DOGE TRX XLM ETC BCH DOT ATOM VET EOS NEO IOTA ZEC DASH XTZ ALGO FIL UNI AAVE SOL AVAX, plus QTUM ONT ZRX BAT ICX FTT, and the delisted XMR WAVES MATIC OMG (all have bars until their Binance delisting). At each bar a name is eligible only while is_listed() is true and it has at least lookback_days+1 daily bars (universe_at() / listed_since()); the base template closes a position at a delisting. LUNAUSDT is excluded on purpose: the Binance spot symbol spans the Terra collapse and the LUNA 2.0 relaunch, so its price series is not one asset. Event definition, per name, on daily closes: close[t] > max(close[t-lookback_days .. t-1]) (a new trailing-365-day closing high), AND the previous such new high on that name was at least min_gap_days bars ago (or never in the available history). That second condition is what makes it an anchor break: the name has spent at least a quarter below its yearly high, and today it clears it. Repeated new highs inside a running trend do not fire. Execution: a market order at the signal bar's close. The capability registry lists next_bar_open_fill as use_equivalent, and on a 24/7 venue the signal-bar close is the next bar's open, which is the substitution named here. Pre-study (fixed rules lookback 365, gap 90, measured on Binance spot daily closes for the 26 surviving names only, before costs, used ONLY to pre-register the firing rate per L238 and L224, not to accept the idea): 97 events from 2019 to 2026, on 83 distinct entry dates in 37 distinct calendar months. By year: 2019 2, 2020 19, 2021 22, 2022 0, 2023 16, 2024 28, 2025 7, 2026 3. The 2022 bear market produces no events at all, so the strategy is flat through it rather than losing in it. The forward 10-day close-to-close return had mean +9.4% and median +4.8%, interquartile -9.4% to +16.8%. The full panel with the 10 extra names should give roughly 110-140 events. The effective sample is the ~40 distinct entry months, not the trade count, and the analyst should read significance that way. Fee arithmetic: a 10-day hold of an alt that just cleared its yearly high moves several percent. Even capturing a third of the pre-study mean (about 3%) clears the 0.20% spot round trip 15 times over. Turnover is tiny: about 15-20 round trips per year at 10% of equity each. Risk control route (b) plus a protective stop. Exposure ends on a signal (the fixed time limit), each position is 10% of equity (never above 100%), and at most 10 positions are open at once (100% gross, unlevered spot). An intrabar protective stop sits stop_pct (15%) below the entry fill, so the worst stop loss is about 1.5% of equity per position. On the BINANCE_SPOT backtest and paper venue it is a platform-matched stop (reduce-only stop-market, triggered inside the bar). It is not an exchange-resting order, and the idea does not need one. Benchmark the analyst should apply (lesson 170 for long-only): equal-weight buy-and-hold of the same eligible panel over the same window. The claim is a much smaller drawdown than buy-and-hold, because the strategy holds nothing in bear markets, and a higher return per unit of exposure. The claim is NOT a higher total return than buy-and-hold in bull years. Risk profile (conservative): spot 1x. With 10 names each at about 90-130% annualized vol and high correlation at full book, realized strategy vol should be about 25-45%, inside the 60% band. Each 10% position contributes about 9-13% annualized vol, far under the 120% per-position cap.

Original Research Lead hypothesis, built on George & Hwang (2004, 'The 52-Week High and Momentum Investing', Journal of Finance 59(5)). Mechanism: investors anchor on a stock's 52-week high. Good news arriving when price is near that high is under-reacted to, because traders treat the high as a ceiling and sell into it. Bad news on stocks far below their high is also under-reacted to, because holders hesitate to sell at a large loss. Prices therefore drift toward the information: stocks near their 52-week high keep outperforming, and stocks far below it keep underperforming. George & Hwang show this nearness measure subsumes most of Jegadeesh-Titman return momentum and, unlike momentum, does not reverse at long horizons. It is a different signal from the abandoned S&P residual-momentum run (fee_edge, past-return ranking). Here the rank is price relative to its own anchor, not past return. Venue and data: USEQ US stocks, 1-DAY bars, pure OHLCV with no supplementary feed (L189). Zero commission and about 1 bp half-spread make a monthly long-short book cheap to run. USEQ is 9.7% of the corpus. Stock-level S&P baskets have only 2 prior runs in 1,869 hypotheses. Long-short and monthly-horizon buckets are both under target. UNIVERSE, point in time (survivorship rule): every stock that was an S&P 500 member at any time from 2005-01-01 to the present and has daily bars in the catalog. That is 675 names from `python -m src.data.listings sp500 --start 2005-01-01 --end 2026-10-01 --with-data`, listed in `instruments` after SPY. SPY.USEQ is the primary instrument and is used ONLY as the session clock; it is never traded. On each rebalance, candidates are `self.universe_at(index='sp500')` with at least `lookback_sessions` completed daily bars of their own. Set `parameters['point_in_time_universe'] = True`. A held name that leaves the index is sold at the next rebalance (index_member False). Residual bias: companies acquired or bankrupt are mostly missing from the free source; state this in the report. SIGNAL per candidate at the rebalance close: ratio = close / max(daily HIGH over the last lookback_sessions bars, including today). Values near 1 are at the anchor; low values are far below it. BOOK: dollar-neutral, gross 1.0x of equity. The long book (50% of equity) is equal-weighted across the names in the top `entry_frac` of ratio; the short book (50% of equity) is equal-weighted across the bottom `entry_frac`. HYSTERESIS to cut turnover: a held long stays while its ratio rank is inside the top `exit_frac`, a held short stays while inside the bottom `exit_frac`. Vacated slots are refilled from the entry band, and both books are re-equal-weighted to 50% each at every rebalance. Re-equalizing is skipped for a leg whose weight drift is under 20% of its target, to avoid dust trades. EXECUTION: decide at the close of the last session of each calendar month (helpers: src.strategies.helpers.sessions / FactoryStrategy session helpers; never hard-code dates). Market orders fill at the next session's open (platform next_open_fill on USEQ, supported). Shorts use the USEQ Reg-T margin account. COST ARITHMETIC: the round trip is about 0.02-0.05% of notional (zero commission, 1 bp half-spread plus impact on S&P names). George & Hwang report about 0.45-0.65% per month for the 52-week-high long-short over 1963-2001 on all CRSP stocks. For large caps after publication I expect roughly a third of that, about 0.15-0.25% per month on the 1.0x gross book. With hysteresis, expected one-way monthly turnover is about 25-35% of each book. Monthly cost is then about 0.3 x 2 books x 0.5 x 0.04% = 0.012% of equity, more than 10x below the expected gross. Per trade: a leg is held about 3-4 months on average and captures about +/-0.4-0.8% relative to the opposite book, so the expected avg_trade_return_pct is about 0.3%, against the 0.05% USEQ floor. Pre-registered evidence count (L238): about 250 monthly rebalance periods over 2005-10 to 2026-09. Expected after-cost return is about +0.15% per period on equity. The per-name trade count (several thousand) overstates the independent evidence; the analyst should judge the period count. RISK CONTROL route (b): signal exits at the monthly rebalance with capped sizing. A single leg holds about 1-2% of equity, far below the 100% cap. Gross leverage is 1.0x, inside the 2x Reg-T limit and the 5x conservative-profile cap. Expected annualized volatility is 8-12%, inside the 60% band. No stop-loss: the source strategy has none, and per-name stops would turn a monthly cross-sectional book into a noise-driven trader. Not modelled: borrow cost on shorts (S&P names are general collateral at about 0.25-0.5%/yr, a known small drag) and dividends (short legs owe dividends in reality, long legs receive them; roughly a wash, with a small bias in favour of the long book because higher-yield value names tend to sit far from their highs). No performance pre-study was run; the numbers above come from the literature and the cost model.

Original Research Lead hypothesis. It pre-registers the NEW hypothesis that the analyst named when closing the OKX transfer of the daily down-shock absorption panel (run 7276bf13, closed underpowered: 397 trades, avg trade 0.97%, PF 1.41, but 2022 lost 12.4% and drawdown hit 23.2%). The analyst wrote: 'A regime filter (for example, skip entries when BTC is below its 200-day average) would be a NEW hypothesis to pre-register, not an amendment.' This run is that test. It moves to BINANCE_SPOT (10.4% of the corpus against a 15% target; spot-tagged runs survive best, L192), the correct venue for a long-only, unlevered idea. Binance spot also has deeper history than Hyperliquid, so the test window holds a full bear year (2022) and the 2026 drawdown. MECHANISM: a UTC day that closes with an unusually large drop on unusually high dollar volume is dominated by forced and impatient sellers: perp liquidations spilling into spot, stop cascades, de-risking. Price-insensitive selling pushes the close below where patient liquidity clears, and the buyer is paid by a partial rebound over the next 1-3 days. This is the long-side absorption design that L224/L230 record as working: Hyperliquid daily panel 7710f417 promoted (191 trades, 2.49%), Bybit spot 4H panel promoted (405 trades, 1.78%). The new element is the regime gate. In a sustained bear market, a down-shock is information (the next leg down), not a liquidity overshoot. That is exactly where the OKX transfer lost its money. The gate is the textbook 200-day SMA of BTC, fixed in advance. It is NOT searched. UNIVERSE (rule-based, point in time, includes later-delisted names): the 40 largest Binance spot USDT pairs by Q4-2020 (Oct-Dec) quote volume, among pairs listed on 2020-12-31. Excluded: BTC itself, stablecoins and fiat pairs, PAXG/WBTC, and leveraged UP/DOWN tokens. Result: ETH, XRP, LTC, LINK, BNB, YFI, BCH, ADA, GRT, XLM, EOS, DOT, UNI, TRX, YFII, SUSHI, VET, SXP, FIL, ZEC, ZIL, AAVE, ETC, XMR, DASH, NEO, THETA, CRV, ATOM, XTZ, WAVES, OMG, BAND, EGLD, ONT, NEAR, REN, RSR, SNX, ALGO. Several were later delisted from Binance spot (XMR, WAVES, OMG, REN, YFII and others); they stay in the universe. The base template's listing calendar closes a position at a delisting at the last price, and is_listed() blocks entries after it. Because membership is fixed with information as of end-2020, NO entries are allowed before 2021-01-01 (parameter trade_start). Every traded event therefore comes after the selection date, which removes the survivorship and look-ahead bias that a 'today's names' basket would carry. BTCUSDT.BINANCE_SPOT 1-DAY is a signal-only leg for the regime gate and is never traded. PRE-STUDY (Research Lead, local catalog BINANCE_SPOT 1-DAY, rolling stats from past bars only, entry at the signal close, 0.25% round trip charged = 0.20% fee + 0.05% slippage, non-overlapping per name, events from 2021-01-01). Frozen HL rules: shock_z <= -2.5, volume_z >= 1.0, 60-day lookback, 2-day clock. - Ungated: 740 events on 161 distinct entry dates, mean net +1.98%, median +1.44%, win 60%. By year: 2021 +2.3%, 2022 -3.83% (109 events), 2023 +2.05%, 2024 +2.83%, 2025 +7.65%, 2026 +2.21%. Capped portfolio (8 concurrent x 8%): 464 trades, +53%, Sharpe 0.65, max DD 26.1%. This reproduces the OKX failure mode. - Gated (BTC close > SMA200 at the signal close): 477 events on 99 distinct entry dates (L238: entry dates are the real sample, not trades), mean net +3.11%, median +2.01%, win 66%. By year: 2021 +2.19% (120), 2023 +2.04% (142), 2024 +2.77% (128), 2025 +8.26% (68), 2026 +0.72% (19). 2022 has zero gated events. Capped portfolio: 302 trades, +79%, Sharpe 1.00, max DD 7.0%. - The complementary BTC<SMA200 arm: 263 events, mean -0.06%, 3-day mean -1.24%. HONEST CAVEATS, disclosed because I saw them before writing this: (1) I looked at the gate split in the pre-study, so the gate is not blind. It was proposed by the analyst before this study, and its 200-day length is the textbook value, frozen and not tuned. (2) The gate costs money in 2026: most 2026 events fire with BTC below its SMA200, and those were positive (+2.61%). (3) 2025 carries about 38% of the gated pre-study profit. (4) Distinct entry dates are about 99 over 5.8 years, about 17 a year, so the evidence is moderate. LONG-ONLY DRAWDOWN COMPARISON (L170/L224): an equal-weight buy-and-hold of the same 40 names drew down more than 80% in 2021-22. The pre-registered limit for this panel is 15%. EXECUTION SUBSTITUTION (capability next_bar_open_fill = use_equivalent): entries and exits are market orders at the signal bar's close. On this 24/7 venue that price is the next bar's open. Spot CASH account: long-only, no leverage.

Derived by the Research Lead from 发明者量化-小小梦, '数字货币手动期现对冲策略' (FMZ digest-topic/7932, 2021-09-08, strategy source https://www.fmz.com/strategy/314352). What the source gives: a semi-automatic, button-driven futures-spot hedge on OKX. It monitors pairs of an OKX USDT-margined dated quarterly future and the same coin's spot (default pairs LTC-USDT-211231|LTC_USDT and BTC-USDT-211231|BTC_USDT). It shows the 正对冲差价 (future bid1 - spot ask1), and on a human click opens 正套 (short the future, buy the same coin amount of spot) or closes it with 平正套. It also rebalances leg quantities every 5 minutes (keepBalance) and does not support 反套 (long future / short spot). Settings: futures leverage 5-10x, 50% futures margin reserve, a fixed 0.1-coin clip. The images show the parameter panel, a status table with the LTC basis at about 4.6-4.7 USDT on about 180 (≈2.6%) and BTC at about 1,345 on about 45,500 (≈3.0%), a 0.1 BTC 正套 opened at a 1,393.8 spread, and the author's note after closing it: '亏了!!! 利润差价小的时候平仓覆盖不了手续费' (closing on a small spread did not cover fees; plan the close spread net of fees and slippage). What the source does NOT give: when to open or close. A human decides both. Every timing rule below is mine (operationalized). WHY THIS IS A RESTATEMENT (labelled change 1, the reason for this retry). The platform holds no OKX dated quarterlies, only the OKX USDT perpetual swaps (LTCUSDT.OKX, BTCUSDT.OKX) and OKX spot (LTCUSDT.OKX_SPOT, BTCUSDT.OKX_SPOT). The previous contract kept a quarterly-style 100 bp premium trigger on the perp, and it never fired: in about 59,700 LTC OKX hourly bars (2019-12 to 2026-10) the perp/spot close premium has median 0.0 bp, 99th percentile 19 bp and std 7.5 bp. A perpetual's basis carries no time-to-expiry, so it never reaches quarterly levels. A perp does pay its premium out continuously as funding, which a 正套 short collects. So on a perp the same trade (long spot, short the derivative, flat-to-positive basis) earns from (i) premium convergence plus (ii) funding received while the premium stays positive. The entry is therefore a rolling-mean premium measured in a few bp, which fires on this instrument. The direction, the legs, the venue (OKX) and the 正套-only restriction stay as in the source; 反套 stays unsupported, as in the source. LABELLED CHANGE 2: both pairs from the source's default 对冲控制表 (LTC and BTC) are kept, as independent sleeves. BTC is over-represented in the corpus, but the source names it explicitly and doubling the events matters (see the firing rate below). LABELLED CHANGE 3: leverage 1x on the perp leg, instead of the source's 5-10x margin setting. The hedge is delta-neutral, so leverage only changes margin usage, and 1x keeps the liquidation buffer far beyond 8 daily moves. LABELLED CHANGE 4: sizing as a % of equity instead of a fixed 0.1-coin clip. LABELLED CHANGE 5: the 5-minute keepBalance becomes a per-bar leg-balance check (self.legs_balanced()). FIRING RATE AND COST ARITHMETIC (feasibility measurement made only so the entry can fire; L223 pre-registration). Rule measured: 24h-mean premium > 3 bp → enter; exit when the 24h mean ≤ 0 or after 7 days. Funding comes from the platform's OKX daily funding aggregates (from 2021-01). LTC: 53 episodes (2021: 29, 2022: 5, 2023: 6, 2024: 10, 2025: 2), median hold 7.0 days, median premium convergence 6.7 bp, median funding received 50 bp, median gross 59 bp, mean 70 bp. BTC: 47 episodes, median gross 52 bp, mean 68 bp. That makes about 100 episodes in total. Cost per episode on the hedged notional: spot round trip 0.20% plus perp round trip 0.10% = 0.30%, so the expected net is about +0.25-0.40% per episode on the hedged notional, roughly 2x cost on the mean. The measurement is gross of slippage, and the events cluster in 2021 and 2024 bull phases. The backtest decides; these numbers are predictions only. RISK CONTROL ROUTE (b): signal and time exits with capped, unlevered sizing. Each pair's spot leg is at most 40% of equity, and its perp short is the same coin quantity (≤40% notional at 1x). Combined gross across both pairs is at most 160% of equity, while net delta is about 0. Intended realized volatility is about 3-8% annualized. EVALUATION NOTE FOR THE ANALYST: each episode closes TWO positions (spot long, perp short) whose price PnLs largely cancel. Read per-position avg_trade_return_pct as roughly half of the episode's net capture on hedged notional. The pre-registered numbers below are stated at the per-position level, and the pair-level figure is about twice that. Funding must be in PnL (the funding ledger for OKX swaps); without it, the carry half of the edge is missing.

Original Research Lead hypothesis. It tests the promoted Hyperliquid daily down-shock absorption panel (run 7710f417: 191 trades, avg trade 2.49%, PF 2.55, Sharpe 0.88, DD 8.0%) as a FRESH test on a different venue and a longer history (lesson L228: freeze the source rules and report the destination separately). The rules are frozen exactly as they were on HL: 60-day lookback, shock_z <= -2.5, volume_z >= 1.0, 2-day clock, 8% per leg, at most 8 concurrent. Only the venue and universe change. Why this test is informative and not a cosmetic copy: (1) OKX is 3.5% of the corpus against a >=5% quota. (2) OKX USDT swap daily history starts in 2020, while HL's traded history starts in 2023-03. The HL result has never been through a sustained bear market. This run includes 2021-05 and the whole of 2022 (LUNA, 3AC, FTX), which is exactly the regime the HL write-up named as the weak spot. (3) It follows L224/L230: a long-side, flat-most-of-the-time absorption event with a fixed clock and no short arm. Mechanism: a UTC day that closes with an unusually large drop on unusually high volume is dominated by price-insensitive sellers (cross-margin liquidations, auto-deleveraging, stop cascades, retail de-risking). That pushes the close below where discretionary liquidity clears, and a patient buyer is paid by a partial rebound over the next 1-2 days. The HL controls showed the premium is mostly the market-wide crash rebound. A BTC-hedged version cut it by about 2/3. A quick check of the idiosyncratic, BTC-hedged trigger on OKX spot alts also came out negative (232 events, mean hedged 3-day forward -0.40%). So the design stays UNHEDGED and long-only, and its risk is the sustained bear leg. That risk is stated, not hidden. UNIVERSE (point in time, fixed before any return was examined): BTC, ETH, SOL, DOGE, AVAX, LINK, LTC, BCH, ATOM, DOT, ADA, XRP, NEAR, UNI, AAVE, FIL, ETC, TRX, XLM, COMP, SUSHI as <BASE>USDT.OKX swaps (EOS dropped because it has no local daily history). A name is eligible on a day only when is_listed() is true AND it has at least `lookback` prior daily bars with volume > 0. Survivorship residual: OKX keeps no delisted swaps, so the universe is limited to names still listed today. All 21 are large, long-lived names, so the bias is small but non-zero, and it is named here. PRE-REGISTERED EVENT COUNT (local catalog OKX 1-DAY, rolling stats from past bars only, non-overlapping per name, 0.12% round trip charged, no concurrency cap, so this is an event count and NOT the strategy result): 590 events on 140 DISTINCT event dates, 2020-2026 (lesson L238: the distinct dates are the real sample, because crash days fire on many names at once). Mean net 2-day forward +0.90% per event, median +0.97%. Per-year mean per event: 2020 +4.1%, 2021 +1.1%, 2022 -4.0%, 2023 +1.2%, 2024 +2.0%, 2025 +2.7%, 2026 +2.1%. 2022 is negative, as the mechanism predicts for a sustained bear leg, and the 8-name cap is what bounds it. Lesson-170 drawdown comparison: equal-weight buy-and-hold of this universe drew down more than 80% in 2021-22. This strategy is in the market roughly 5-10% of days, at most 64% gross. Execution substitution (capability next_bar_open_fill = use_equivalent): entries and exits are market orders at the signal bar's close, which on this 24/7 venue is the next bar's open. Risk-control route (b): time exit with capped sizing (8% of equity per leg, at most 8 legs, 64% gross maximum, 1x, no leverage). No stop is added, matching the frozen source rule. Worst case per trade: an 8% leg that loses 40% in 2 days costs 3.2% of equity, inside the 15% max_trade_loss limit.