Hypotheses

Market-level re-reading of the factory's promoted NR7 failed-downside-break panels (Bybit, OKX, Binance spot, Hyperliquid). These sources are the factory's own results, not an external seed. A Research Lead pre-study on BINANCE_CM daily bars (2020-08 to 2026-10, 16 alt USD-margined inverse perps) showed three things. (1) The per-name long arm replicates on COIN-M: 1,564 non-overlapping events, mean +1.07% net of 0.06% round trip, positive in 6 of 7 years. (2) The mirror short arm (short a close ABOVE the NR7 high) loses: 1,408 events, mean -0.58%. This is consistent with L230, so there is no short arm. (3) KEY: hedging each per-name event with 60-day-beta BTC drives the mean to -0.01%. The 'reversal' is a market-wide rebound, not idiosyncratic alt mean reversion. The honest test of that mechanism is therefore a MARKET-LEVEL event: the BREADTH of same-day NR7 downside breaks across a panel, traded through one liquid low-fee instrument. ETH is used instead of BTC for corpus concentration. ETH's own NR7 signal is not used, because it adds little: ETH own-NR7 events made +0.84% (n=101), while breadth days without an ETH-own event made +1.37%. Signal panel (fixed, pre-registered; chosen by history length, not performance): ADA, LINK, DOT, XRP, LTC, BCH, ETC, BNB, DOGE, FIL, SOL, ATOM, AVAX, TRX, XLM, EOS, all <BASE>USD_PERP.BINANCE_CM 1-DAY. Each name counts only when it is listed (is_listed / listed_since >= 30 days; set point_in_time_universe). The breadth fraction uses the names eligible that day as its denominator. PRE-STUDY for the traded leg (ETHUSD_PERP.BINANCE_CM, non-overlapping events, 3-day hold, net 0.06% RT): breadth >= 15% gave n=211, mean +1.52%, win 59%, positive in 2020, 2021, 2022, 2023, 2024 and 2026, negative in 2025 (-1.38%). The unconditional ETH 3-day drift is +0.48% gross, so the signal's incremental expectancy is about +1.0% per event. The surface is smooth: count thresholds k=2..5 give +0.98 to +1.30%, and fractions 15/25/35% give +1.52/+1.03/+1.20%. The same signal on BTCUSD_PERP.BINANCE_CM gives +0.96 to +1.28%, which corroborates the mechanism. At 50% notional the event book compounds to about 4.3x with about 25% max drawdown; the worst single event was -27% on ETH, i.e. -13.7% of equity. Execution: market order at the signal day's close. On this 24/7 venue that is the registered equivalent of a next-bar-open fill (next_bar_open_fill -> use_equivalent). Inverse sizing uses contracts_for_notional() (whole USD-face contracts; supported only on BINANCE_CM). Risk route (b): fixed time exit plus capped sizing at 50% of equity, unlevered, no stop. A protective stop is optional and off by default, because the per-name siblings lost Sharpe with a tight stop. COIN-M margin is coin-denominated, so equity moves with ETH while a position is open; this is inherent to the venue and is noted as a risk.

A breadth test of the factory's most-replicated promoted mechanism on a new venue type and new names. The mechanism has been promoted on Bybit perps, OKX swaps, Binance spot and Hyperliquid perps (742-1,090 trades, PF 1.31-1.47, average trade 0.97-1.10%). Here it moves to Binance COIN-M inverse perpetuals (BINANCE_CM, 2.7% of the corpus against a >=5% target) and to 12 names that no sibling panel traded. Following lesson L228, this is a fresh test: the rules are frozen from the siblings and the result is reported on its own, not inferred from them. MECHANISM: the NR7 day is the day with the narrowest high-low range of the last 7 daily bars. It packs resting stop-losses and breakout-short entries just below a tight range. A daily CLOSE below that NR7 day's low runs those stops into a thin liquidity pocket. The forced selling is spent by the close, and price tends to revert over the next few days. The sibling rule is kept exactly: there is no reclaim condition. UNIVERSE (pre-registered and fixed): EOSUSD_PERP, XTZUSD_PERP, VETUSD_PERP, THETAUSD_PERP, ICXUSD_PERP, ZILUSD_PERP, KNCUSD_PERP, CHZUSD_PERP, MANAUSD_PERP, SANDUSD_PERP, ALGOUSD_PERP and EGLDUSD_PERP on BINANCE_CM. Selection rule: older COIN-M alt perps (daily history from 2022-08 or earlier), excluding BTC/ETH/SOL for corpus concentration and excluding every name used by the Bybit, OKX and Binance-spot NR7 siblings (so ADA, XRP, DOGE, LINK, LTC, BNB, AVAX, DOT, ATOM, FIL, ETC, BCH, UNI, NEAR, TRX, XLM and AAVE are out). The 12 names were fixed before testing, and none was dropped for its results (VET and CHZ were negative in the pre-study and stay in). Several contracts stopped trading on COIN-M (for example EOS in 2025-05, THETA/MANA/EGLD in 2025-11, CHZ in 2025-10). The developer must set parameters['point_in_time_universe']=True, so each leg trades only while is_listed() and at least 30 days after listed_since(). The base template closes a position at a delisting. RESEARCH LEAD PRE-STUDY: Binance COIN-M daily bars 2020-09 to 2026-10, events non-overlapping per name, gross of fees, used only to check the firing rate and the fee arithmetic, never to accept the rule. The 12-name panel produced about 673 events (about 150 per year), mean 3-day forward return about +1.0%, median about +0.85%, win rate about 54%. Every year from 2021 to 2026 had a positive mean; 2022 was the weakest at +0.29% across 244 events. EXECUTION: daily signals on completed bars. Entry is a market order at the signal bar's close; on a 24/7 market this is the platform's equivalent of a next-bar-open fill. Contracts are inverse USD-face contracts sized with contracts_for_notional(). Fees are the BINANCE_CM taker rate on both sides.

Faithful port of the Quantiacs open-source strategy 'Stateful-machine-learning-neural-network-strategy' (author: Quantiacs, MIT licence, commit ce91142). Credit to the author; the source's reported Sharpe 0.85 (2006-01-01..2024-10-17) is unverified and our own backtest with real costs decides. MARKET: US equities, venue USEQ, 1-DAY bars, long-only, no leverage (gross <= 1.0). UNIVERSE: the S&P 500 AS OF EACH DECISION DAY (point-in-time membership via index_member()/universe_at(), former members included) — this replaces Quantiacs' 'is_liquid' field, which in the Quantiacs S&P 500 dataset flags current index membership. SPY.USEQ is the primary bar_type only as the session clock (never traded); every S&P 500 member that was ever in the index during the backtest span is subscribed as an extra instrument. MECHANISM: each day, from the index members as of the last bar, select the 15 assets with the LOWEST rolling standard deviation of daily close-to-close returns over min(450, available_bars-1) days (qnt.filter.filter_volatility metric='std', ascending=True; assets lacking a full window are not eligible). For those 15 assets (kept in alphabetical symbol order — the order Quantiacs' xarray selection yields, and the order that maps them onto the network's 15 outputs) compute 6 features per asset, flatten to a 90-vector, feed a fully connected network 90->128 ReLU->64 ReLU->15 Sigmoid, and read the 15 outputs as P(next close > today's close) for the 15 assets by position. MODEL: implemented in pure numpy (torch is outside the sandbox import allowlist; this is the only substitution in the model): PyTorch-default Linear init (weights and biases U(-1/sqrt(fan_in), +1/sqrt(fan_in))) from a local numpy Generator seeded 42 (never the global RNG), full-batch training for 100 epochs, binary cross-entropy (mean over all time x asset cells), Adam lr=0.002 (betas 0.9/0.999, eps 1e-8). Training is retrained every 90 calendar days (qnt backtest_ml retrain_interval=90) on the most recent 250 trading days (train_period=250); between retrains the frozen model predicts each day (predict_each_day=True). The training sample re-selects the 15 lowest-volatility members as of the last training day using the same selector with rolling window min(450, 249) = 249 returns. TARGET: 1 if next day's close > today's close else 0; the final training row has no next close and is labelled 0 (the source's xr.where on NaN). FEATURES (exact source formulas, in this order): f0 = (RSI7-50)*sqrt(7)/65; f1 = (RSI20-50)*sqrt(20)/65; f2 = candle = (close-open)/(close*2*max(ATR14,0.01)); f3 = (RSI60-50)*sqrt(60)/65; f4 = (RSI150-50)*sqrt(150)/65; f5 = ATR14/close. RSI and ATR are qnt.ta.rsi / qnt.ta.atr with their library-default smoothing (developer replicates the qnt implementation; ATR uses true range with previous close). Inputs are forward-filled then back-filled then NaN->1 (OHLC) / NaN->0 (indicators) within the window, as in the source. Flattening follows transpose('time','feature','asset').reshape(-1, 90): index = feature*15 + asset. PORTFOLIO STATE (the 'stateful' part): each day start from yesterday's target weights (positions carry over); for each of the 15 selected assets whose prediction > 0.7, set its target weight to prediction*0.1 (max 0.1); assets not re-predicted keep their previous weight; multiply by index membership (non-members go to 0). Exits zero a weight: max-hold 217 trading days since the position opened (qnte.max_hold_long), take-profit when close >= 1.23 x entry price, stop-loss when close <= 0.95 x entry price (qnte take_profit/stop_loss_long_percentage, close-based). If the sum of weights exceeds 1, scale all weights by 1/sum. A zeroed asset only re-enters when the model again predicts > 0.7. EXECUTION: Quantiacs applies weights decided on day t's close at day t+1's open; on USEQ that is the platform's supported next_open_fill (market order decided on the close fills at the next session's open). Entry price for TP/SL is the actual entry fill price (the next-session open), matching the source's 'open price' semantics. Target weights are fractions of current equity; held positions are rebalanced to target each day (Quantiacs fractions semantics), with an engineering band: only send a rebalance order when the share delta exceeds 10% of the target position or the target changes, to avoid micro-orders (whole shares, zero commission, spread/impact modelled). RISK CONTROL: route (b) — signal/time exits (217-day clock, close-based TP/SL) plus capped sizing (<= 10% of equity per name, gross <= 100%, unlevered); a 5% close-based stop on a <= 10% position risks ~0.5% of equity, more only on gaps. Expected per-trade arithmetic: low-vol large caps held up to 217 days with a 23%/5% asymmetric bracket; a typical round trip should capture well over 1% versus ~0.04% USEQ round-trip spread+impact cost.

Derived by the Research Lead from Sebastien Valeyre, Breaking the Trend: How to Avoid Cherry-Picked Signals: port the explicit naive 1/N continuous-position variant in Sections 4.1 and 4.3, rather than introduce an approximate ARP matrix cleaner. Use daily bars on six predeclared Hyperliquid perpetuals, with EMA relaxation time 120 days, 40-day volatility estimation and 20-day position smoothing as source baselines. This tests transfer of the source mechanism to crypto, not reproduction of its 70 traditional-futures dataset or claimed Sharpe. Universe, 10% annual portfolio volatility target and conservative exposure caps are Research Lead adaptations because the paper does not specify an executable risk budget. Market orders at signal-bar closes are the capability-registry equivalent of next-bar-open fills on continuous venues. No exchange-resting orders, opposing positions on one instrument or selectable same-bar policy are required. All performance numbers below are unverified predictions.

Derived by the Research Lead from the MQL5 i-Regr H&L indicator: test whether an escape from a previously projected high/low regression channel identifies persistent repricing rather than ordinary movement along a trend. Source URL: https://www.mql5.com/en/code/8436 . The locally archived article describes separate regressions of highs and lows, but supplies no trading rules or performance evidence. All entry thresholds, instrument choices, exits and risk settings below are Research Lead choices. Independently trade XMRUSD.HYPERLIQUID, BCHUSD.HYPERLIQUID and LTCUSD.HYPERLIQUID on 4-HOUR bars, long and short, holding at most one net position per instrument. Use each instrument own bars, with XMR primary and BCH/LTC as extra instruments. Market entries and timed exits execute at the signal bar close under the continuous-market convention. Protective stops trigger intrabar in the platform. No exchange-hosted protection, independent opposing executors, tick latency edge or selectable same-bar policy is required. Use all available overlapping history without assuming Hyperliquid supplies multi-year sub-hourly data.

Independent venue replication of the factory's promoted Bybit, OKX and Binance-spot compression-trap mechanism. On Hyperliquid daily bars, buy a close below the preceding NR7 day's low and exit three days later. Preserve the source's actual rule: no reclaim above the range is required despite the historical failed-breakout label. Use eight independently traded, predeclared perpetuals; no cross-sectional ranking or current-liquidity selection. Daily bars avoid Hyperliquid's short intraday-history cap. Market orders at completed signal-bar closes implement the registered continuous-market equivalent of next-bar-open fills. This is a new venue test, not an assertion that sibling results transfer. No performance pre-study was run. Commission assumption is 2 x 0.045% = 0.09% per round trip, matching [Hyperliquid base-tier perpetual fees](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees).

Authorize a fixed risk-budget revision of the existing SOLUSDT.BINANCE 1-DAY contract: position_pct=0.10 with config.leverage=2.0, hence entry notional=equity × 0.10 × 2.0=20% of equity, down from 30%. Preserve the implemented daily-close EMA and Chandelier rules and their indicator definitions. Risk control route (b): signal exits with capped sizing, one long position, no pyramiding, no shorts and no continuous rebalancing. The account margin setting is 2.0; intended economic exposure is 0.20x at entry. Market orders execute at the signal daily close under the supplied continuous-market convention. There is no intrabar protective-stop requirement. This is a risk-admissibility experiment within the existing lineage; reduced sizing does not create or validate a trading edge. Preregister an unmeasured working capture hypothesis of 0.50% gross per trade less 0.10% modeled round-trip commissions = 0.40% before additional spread/impact; gross capture is 5 times commissions. Predict at least 0.30% net average trade return after all modeled costs. These are falsifiable predictions, not observed measurements.

Faithful port of Quantiacs' MIT-licensed strategy-ml-nn at commit ebd9ea79fe175390a7e8a38141b3d755d8608a71. Follow the executable stock rules despite the notebook's cryptocurrency wording: independently train and trade AAPL, AMZN and MSFT using daily bars. Preserve the requested source defaults and raw signed model outputs. Adapt venue, prices, whole-share sizing and costs to USEQ. Daily close decisions execute at the next session open, as supported by the execution registry. The source supplies no explicit stop, take-profit or probability threshold. Historical testing uses annual model retraining; its separate daily post-submission retraining default is recorded but must not silently replace the historical experiment. The twelve supplied figures show documentation logos, notebook demonstrations, data coordinates and generic performance charts; none establishes this strategy's trading performance.

Derived by the Research Lead from Lim, Zohren and Roberts, Enhancing Time Series Momentum Strategies Using Deep Neural Networks: test direct LSTM position outputs trained on turnover-adjusted Sharpe within volatility-scaled time-series momentum. Use daily LTC, XRP and ADA Hyperliquid perpetuals. Preserve the actual recurrent architecture, the source's 15% annualised volatility target and its 10-bps one-way turnover penalty. Adapt the original continuous-futures universe, initial training period and exposure caps explicitly. Implement the LSTM and Adam using permitted NumPy operations; do not substitute an MLP, classifier or heuristic. The supplied full text resolves details absent from the abstract. All four supplied figures were inspected: they show asset-level performance dispersion, turnover, cost sensitivity and regularisation comparisons, rather than marked trading entries. Their reported performance is unverified on these markets. Continuous-market orders execute at signal-bar closes, the registry's equivalent of next-bar-open execution.

Test whether directional pressure distributed across many 4-hour bars predicts multi-day continuation better than an equally large move concentrated in one jump. Trade HBAR and ICP independently, long and short. Use market orders at the signal bar close, the registry’s faithful equivalent of next-bar-open execution on continuous markets. Protective stops trigger intrabar through platform matching. Use only actual available history, approximately 5000 four-hour candles per instrument before gaps and warm-up. Compare against a displacement-only diagnostic that removes the sign-count and jump-concentration filters; predict an incremental 0.15 percentage points of average net trade return. Diagnostic holdout results cannot rewrite this experiment. Base-tier taker commissions are 0.045% per fill according to the [official Hyperliquid fee schedule](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees).

Derived by the Research Lead from ChaoZhang’s [Heiken Ashi and Ichimoku strategy](https://www.fmz.com/strategy/427137), whose local article and Pine rules were read. Preserve its signal predicates on independently traded ALGO and ICP perpetuals, adapting to completed 4-HOUR bars and reduced allocation. Market orders at the signal close implement the registry’s continuous-market equivalent of next-bar-open fills. Synthetic candles generate signals only; fills use real prices. Baselines: tenkan_bars=9, kijun_bars=24, span_b_bars=51, displacement_bars=24. Share parameters across instruments and use all available same-venue history.

Trade XMRUSD.HYPERLIQUID and ZENUSD.HYPERLIQUID independently, long and short, on completed 4-HOUR bars. Test whether shallow, low-volume retracements after high-volume directional impulses predict renewed movement in the impulse direction. Market orders execute at the signal close, the supported equivalent of next-bar-open execution on continuous markets. Protective stops trigger intrabar through platform matching. Both exact bar directories exist. Use each instrument's available history independently. This is an original, untested hypothesis.

Test whether volatile but reversing ETH price paths support profitable discrete gamma harvesting when weekly Deribit straddles underprice variation observable at the hedge cadence. Buy actual same-expiry ATM calls and puts, and hedge their combined delta with ETHUSDT.BYBIT. Use synchronized 4-HOUR bars and RollingOptionsStrategy. ETH-PERPETUAL.DERIBIT supplies an untraded clock and reference price. Market orders execute at completed signal-bar closes, the capability registry's equivalent of next-bar-open execution on continuous venues. Resolve historical option contracts through the supported weekly chain and append every full option instrument id to the persisted instruments and extra_instruments before verification. Use one bar type per instrument and full available common history. Official [Deribit Standard fees](https://support.deribit.com/hc/en-us/articles/25944746248989-Fees) charge options 0.03% of underlying notional per fill, subject to the premium cap; assume no combo discount.

Derived by the Research Lead from Alexander Fedosov's Strategy builder based on Merrill patterns: test causal five-swing price continuation on independently traded ZEC and DASH perpetuals using 4-HOUR bars. The [source article](https://www.mql5.com/en/articles/7218) supplies a pattern-testing framework, not these trading rules or profitability evidence. Archived and online article text were read. Use all available history per instrument, acknowledging Hyperliquid's approximately 5000-candle limit. Market orders at signal-bar closes implement the registry's continuous-market equivalent of next-bar-open fills. Protective stops trigger intrabar in the platform. Baselines are pivot_span=2, pattern_window_bars=60 and min_impulse_atr=2.0. Economic estimates are predictions, not measured results.

Derived by the Research Lead from ChaoZhang’s Dual EMA Momentum Trend Trading Strategy with Full Body Candle Signal System, https://www.fmz.com/strategy/472974, read locally including its Pine rules. Preserve its candle-body test, close-above/below-both-EMAs entry and opposite-signal reversal. Adapt the BTC 2H example to independently traded ATOM and FIL Hyperliquid perpetuals on 4-HOUR bars with capped sizing. The source requires neither EMA crossovers nor EMA ordering. Market orders execute at completed signal-bar closes, the registry’s equivalent of next-bar-open execution on continuous venues. Use all available history; Hyperliquid’s 5000-candle limit permits approximately 2.3 years at 4H.

Test whether return-volume covariance predicts a multi-day directional move when net price drift is small. High-volume advances offset by low-volume declines suggest accumulation; the mirror suggests distribution. Independently trade ALGO and ICP perpetuals on 4-HOUR bars using OHLCV only. This is an original hypothesis, not a demonstrated edge or an aggressor-flow measurement. Market orders execute at the completed signal bar close, the registry equivalent of next-bar-open execution on continuous venues. Risk control route (b): signal/time exits with 20% equity per sleeve and 40% maximum gross exposure. Official commission reference: [Hyperliquid fees](https://hyperliquid.gitbook.io/hyperliquid-docs/trading/fees).

Derived by the Research Lead from ColorBearsGap: test whether directional displacement leaving a three-bar range gap continues after the gap survives another completed bar. The [source description](https://www.mql5.com/en/code/14319) was read; attached code and smoothing formulas were not inspected. The raw-OHLC gap definition and trading rules below are Research Lead choices, not a reproduction. Trade ZEC and ETC independently on 4-HOUR bars. Market orders at the decision bar close substitute for next-bar-open execution under the capability registry. Protective stops trigger intrabar through platform matching. Evaluate each instrument over its actual available history.

Event-driven, market-neutral test of the index-deletion price-pressure reversal (Harris & Gurel 1986 price-pressure hypothesis; Chen, Noronha & Singal 2004 'The Price Response to S&P 500 Index Additions and Deletions', Journal of Finance). When a stock is removed from the S&P 500, index funds and benchmarked managers must sell it around the effective date, whatever its value. The forced, information-free selling pushes the price below its equilibrium value. Once the selling is finished, the price should partly recover over the following weeks as liquidity providers unwind. Chen-Noronha-Singal report large post-effective recoveries for deletions. Greenwood & Sammon (2022, 'The Disappearing Index Effect') show the effect has shrunk for additions. This test asks whether the deletion reversal survives in 2010-2026, net of costs and hedged against the market. The universe is point in time. Events come only from the platform's historical S&P 500 membership (index_member(), built from fja05680/sp500 + Wikipedia intervals). An event fires on the first daily bar on which a leg that was a member on the previous bar is no longer one. The candidate legs are the 108 former members whose daily USEQ bars continue at least 60 days after a 2010-2026 removal (data_check=ok in sp500_membership.parquet). Acquired or bankrupt names that Yahoo dropped have no post-removal bars, so they never enter. We cannot buy them in reality either, but their absence can bias results upward. Execution: daily bars on a session venue, so a decision at bar N's close fills at bar N+1's open (next_open_fill, supported). No intrabar requirement. Mechanism diversity: this is the factory's first index-reconstitution or forced-flow event strategy. It needs no supplementary feed beyond the membership calendar. It is long/short and runs on USEQ, which holds only 9% of the corpus. It uses about 108 never-used single-stock instruments instead of BTC/majors.

An out-of-universe, out-of-venue breadth test of the one factory survivor family that keeps passing (lesson L219): the promoted Binance seven-major 1D/1W macro trend basket (997 trades, avg trade 3.22%, Sharpe 1.16). That strategy traded BTC/ETH/BNB/SOL/XRP/ADA/LINK on Binance USD-M. This hypothesis applies the SAME construction, with no ranking and no relative strength, to a DISJOINT universe of eight older mid-cap alts on a different venue, BYBIT linear USDT perps (3.0% of the corpus, under its 5% quota): DOTUSDT, LTCUSDT, BCHUSDT, ETCUSDT, FILUSDT, AAVEUSDT, XLMUSDT and TRXUSDT on BYBIT. None of the seven promoted majors is included, so a pass is evidence the mechanism is not a property of those particular large caps. Each name is traded independently on its own absolute trend. Long when its fast (about 1 month) and slow (about 12 weeks) daily-close momentum signs are both positive, short when both are negative, flat when they disagree. The slow screen uses daily closes over a ~12-week window, standing in for the weekly bars, so the strategy subscribes to one bar type per leg (8 x 1-DAY). That keeps the verification bar load small. Several names (DOT, ATOM-class, ETC, LTC) have already died as SINGLE-NAME 4H/1D confluence tests. Those were faster (4H primary), single-instrument and needed per-name significance. This design holds trades for weeks, so each trade targets a multi-percent leg, and it pools eight weakly related trend legs into one book, the way the promoted basket did. Universe point-in-time: a leg is eligible only from its first Bybit daily bar plus warm-up (listed_since/universe_at). Bars start 2020-10 (LINK/LTC), 2020-12 (BCH), 2021-03 (DOT), 2021-05/07 (AAVE, FIL, ETC) and 2021-09 (XLM, TRX). A leg that lists later simply joins later, and a delisting closes the leg at the last price. Set parameters['point_in_time_universe']=True so later-listed legs do not clamp the window. Fills: market orders at the signal bar's close. On a 24/7 venue that equals the next bar's open, the platform's documented next_bar_open_fill equivalent. Protective exit: a reduce-only intrabar ATR trailing stop (_submit_stop_market) that triggers inside the bar in the platform's simulated venue. It is not an exchange-resting order, and the idea does not need offline protection. Risk control route (b): signal exit plus capped sizing, with no position above 25% of equity and gross exposure up to 100% at 1x leverage. The trailing stop is an additional protection, not route (a). Pure OHLCV for signals. Bybit funding is used only for cost accounting.

Cross-venue, market-neutral funding carry between two perps of the SAME coin on two venues. Hyperliquid is retail-heavy, structurally long-crowded and settles funding hourly. OKX USDT swaps price funding off a different participant mix, often near the interest floor. When a coin's HL funding stays above its OKX funding for days, a pair that is short the HL perp and long the OKX perp at equal notional collects the gap with almost no price exposure. The reverse also holds: short OKX and long HL when OKX funding is richer. This extends the promoted single-coin 'DOGE Hyperliquid-Bybit Persistent Funding-Spread Carry' (44 trades, +1.01% avg trade) to a pre-declared 10-coin panel and a different counter-venue (OKX: 3.3% corpus share; HL: 12.7%). DOGE is excluded so the panel does not overlap that run. Pre-proposal check on our own funding_daily aggregates (HL minus OKX daily funding sum, 28 liquid alts, 2023-05 to 2026-10, 31.8k coin-days): mean gap +0.013%/day. When the trailing 7-day mean gap was above +0.05%/day (3,724 coin-days), the gap summed over the NEXT 14 days averaged +0.98% (median +0.76%) and was positive 95.7% of the time. When it was below -0.05%/day (1,772 coin-days), the next 14 days averaged -0.88% (90% same sign). So the gap is persistent in both directions. It decays with the regime, though: next-14-day capture after a positive trigger averaged 1.42% in 2023, 0.92% in 2024, 0.31% in 2025 and 0.26% in 2026 YTD. That decay is the main risk and is pre-registered below. Panel, a hand choice declared as such: TIA, LDO, AAVE, NEAR, AVAX, LINK, SOL, INJ, FIL, SUI. These are the non-DOGE coins with the most trigger days that are also listed with daily bars on both venues. A coin is eligible only on days when both legs have a bar and both funding series have an observation; this is the point-in-time rule, since a later-listed coin simply joins when both legs exist. Execution: daily decision bars on a 24/7 venue. The next_bar_open fill is a use_equivalent capability, so orders are market orders at the signal bar's close, which equals the next bar's open on a continuous market. No stops rest at the exchange; the protective exit is a bar-close basis check (see exit_logic). Funding is read only via supp_as_of / supp_window, point in time.

Cross-venue, market-neutral relative-value trade between a crypto-native gold token and the spot-gold market it can be redeemed against. PAX Gold (PAXG) is a fully allocated, redeemable claim on one troy ounce of London Good Delivery gold. Its fair value is spot XAU/USD plus a slowly varying structural premium or discount (mint and redemption fees, the USDT/USD basis, crypto-side demand). PAXG trades 24/7 in thin crypto liquidity. Crypto treasuries and traders use it as a risk-off parking asset, so crypto-side liquidation waves, stablecoin flows and weekend trading push PAXG away from its usual basis to gold, while spot gold (a deep OTC market) is unaffected. Creation and redemption arbitrage is slow (KYC, minimum sizes, days to settle), so these gaps are not closed at once but revert over hours to days. Strategy: on each 1-HOUR bar where BOTH legs have a fresh close (spot-gold session open), compute basis_t = PAXG_close / XAU_close - 1 and its deviation dev_t = basis_t minus the rolling median of basis over the last `median_lookback` aligned observations. When PAXG is CHEAP against its own recent basis (dev_t <= -entry_dev_pct), buy PAXG spot and short an equal USD notional of spot gold on SPOTMETAL. Close both legs when the deviation has reverted (dev_t >= -exit_dev_pct) or after `max_hold_hours`. Only the discount side is traded: PAXG on BINANCE_SPOT cannot be shorted, and the spot listing (from 2020-08) gives about 6 years of history against about 19 months for the PAXG perp. The hedge removes exposure to the gold price. XAUUSD.SPOTMETAL is the spot-gold anchor, as the factory rules require for gold as a spot market; the PAXG token on Binance is the crypto-venue instrument. Prior art: the knowledge base holds no PAXG, XAUT or PAXG-vs-gold hypothesis, and SPOTMETAL carries 0.3% of experiments, so this is a new price series and a new mechanism (dislocation of a redeemable token from its underlying, not momentum or breakout). A scoping look at hourly PAXG spot from Dec-2025 to Oct-2026 used the Binance gold perp as a stand-in, because the SPOTMETAL 1h catalog is still being backfilled; the strategy itself must use XAUUSD.SPOTMETAL. Against a 168-bar median the deviation had a std of about 0.20%. The 13 episodes beyond -0.5% in about 10 months reverted on average +0.35% within 24h and +0.54% within 72h; episodes beyond -0.8% reverted about +0.8% in 72h. More and larger gaps are expected in the 2020-2022 crypto crash windows (May 2021, May/Jun/Nov 2022). Execution: a bar-close market decision on the 24/7 PAXG leg fills at the signal bar's close, the faithful equivalent of a next-bar-open fill. The gold leg is on a session venue and fills at the next bar's open (next_open_fill, supported). Entries are only evaluated while the gold session is open, so both legs go on together. An exit that triggers while gold is closed (weekend or the daily 17:00-18:00 NY break) is executed at the first bar after gold reopens, with the PAXG leg closed at the same bar, so the pair is never left unhedged by design. Overnight financing on the short gold leg (receives fed funds - 2.5%/yr, so it pays when fed funds is below 2.5%) is charged by the platform's funding ledger. No supplementary feed is needed: OHLCV of both legs only.

Post-earnings-announcement drift (Ball & Brown 1968; Bernard & Thomas 1989; Chan, Jegadeesh & Lakonishok 1996): after a large information shock, prices under-react, so the move CONTINUES for weeks. We have no earnings calendar, but an earnings release shows up in the tape as an abnormally large overnight gap on abnormally heavy volume. This hypothesis trades that proxy on US large caps, with SPY as a beta hedge so the bet is on the stock's own drift, not the market. Prior art: 'Earnings-Gap Continuation (PEAD Proxy)' (run 441e153d) was abandoned with infrastructure=1. It hit the Layer-3 sandbox 300 s timeout because of O(n) per-bar work, so it says nothing about the idea, and it never produced a backtest. This version changes the design to avoid the timeout and to fix three validity gaps. (1) Every rolling statistic is an O(1) deque or running-sum update. Nothing rescans history per bar. (2) The universe is pre-registered point-in-time to avoid survivorship bias. It is the 15 largest S&P 500 members by market cap at 2015-12-31: AAPL, GOOGL, MSFT, BRK-B, XOM, AMZN, META (then FB), GE, JNJ, WFC, JPM, T, PG, VZ, WMT. The list was chosen from information available on that date, not from today's survivors. Trading starts 2016-01-04 (earlier bars are warm-up only), and a name trades only while index_member() says it is in the S&P 500. (3) The gap is market-adjusted (stock gap minus SPY gap), so macro gaps such as CPI or FOMC days do not fire. A 'gap held' filter keeps only shocks the regular session did not reverse. Execution: USEQ is a session venue, so a decision taken on the event day's close fills at the NEXT session's open (platform rule next_open_fill, supported). Exits fill the same way. Diversity: USEQ is 9.2% of the corpus. Long/short (23%) and market-neutral are under-represented. None of the 15 names is in the over-concentrated BTC/Binance set. The mechanism differs from the factory's dead families (momentum confluence, short-term reversal, calendar OPEX).

Derived by the Research Lead from the MQL5 Code Base indicator 'Timeframe Flow' (https://www.mql5.com/en/code/36346). The source draws a user-chosen higher-timeframe candle (for example a D1 candle on H1 charts) at EVERY bar, together with current-timeframe ATR bands, and suggests crosses of the midline (open + close)/2 as exits. It states no entry threshold, sizing or market. In this operationalization, on XAUUSD.SPOTMETAL 1-HOUR MID bars (Dukascopy, history from 2003) I compute at each closed H1 bar a ROLLING higher-timeframe candle over the last N H1 bars (baseline N = 24, about one trading day). Its open is the open of bar t-N+1 and its close is the current close. N counts bars, not clock time, so weekend and 17:00-18:00 NY break closures are skipped and are not treated as gaps. Signal strength is the rolling body measured in current-timeframe ATR units: S = (HTF_close - HTF_open) / ATR_H1(14). We go LONG when S >= +k and SHORT when S <= -k (baseline k = 3.0), meaning a day-scale directional move that is large relative to hourly noise. The position exits when price crosses the rolling candle's midline. Because the midline includes the current close, close < (HTF_open + HTF_close)/2 is the same as the rolling candle turning bearish (close < open of bar t-N+1); the short case mirrors this. That is the source's midline-cross exit, implemented causally. An intrabar protective stop at entry -/+ stop_atr x ATR_H1 (baseline 4.0) is placed as a reduce-only stop-market order. The platform matches it inside the bar in backtest and paper, and live mirrors those fills; no exchange-resting order is required or claimed. Next-bar-open execution on this session venue is native (next_open_fill: supported): a decision on bar N's close fills at bar N+1's open. Positions may be held overnight and pay or receive the modeled SPOTMETAL financing at every 17:00 NY roll (long pays fed funds + 2.5%/yr, short receives fed funds - 2.5%/yr, Wednesday x3); this cost is part of the test. Why gold and why now: corpus concentration is BTC-heavy, and SPOTMETAL holds 0.3% of all experiments. Prior gold tests covered session breakouts (Asian range / London open), GVZ variance-premium regimes, macro-driver (real-yield) trend, PAXG basis and gold/silver ratio. None tested the plain price-based intraday-to-daily trend persistence documented for gold and other futures in time-series momentum work (Moskowitz-Ooi-Pedersen 2012; Hurst-Ooi-Pedersen 2017). This also follows lesson L219: a separately justified directional-momentum test with an instrument-specific, pre-registered expectancy, on a market with much lower cost than crypto (about 0.015% RT spread, zero commission). This is NOT another crypto dual-timeframe confluence clone: the instrument, venue, cost structure and session calendar all differ.

A pre-registered breadth test of the factory's best-performing pure-OHLCV event mechanism on a venue (OKX, 3.3% of the corpus) and on eight symbols that have never been used with it. The mechanism was promoted on ETHUSD.HYPERLIQUID 4H (Sharpe 2.11) and then survived optimization as an 8-alt Bybit panel (SOL/XRP/ADA/LINK/DOGE/LTC/DOT/BNB). This test uses a DISJOINT universe (ATOM, NEAR, ETC, FIL, UNI, AAVE, TRX, BCH) on a different exchange, so it shows whether the effect is a general property of alt-perp price discovery or something specific to those 9 series. Event: a 4H bar whose true range is at least range_mult x the mean true range of the previous atr_len bars (excluding the event bar) and whose close lies in the outer close_zone fraction of its range. A close in the top zone is a buy-side imbalance that was not absorbed by bar end. A close in the bottom zone is the mirror. Enter in the close's direction, hold for a fixed 48 hours, then exit. Each leg runs independently and holds at most one netted position per instrument. Execution substitution (capability next_bar_open_fill = use_equivalent): the market order is sent at the event bar's close. On a 24/7 venue that price is the next bar's open. Universe is point-in-time: a leg is eligible only once it has been listed on OKX for at least atr_len+10 4H bars, using universe_at()/is_listed(), and delisted legs are closed at their last price by the base template. No funding feed is needed for the signal; funding is charged to PnL by the engine. Fee arithmetic: the Bybit panel's per-event 48h move after a qualifying bar averaged about 0.18% net. These OKX names are thinner and more volatile (4H ATR typically 2.5-4% vs 1.5-2.5% for the Bybit names), so an event bar is >= 5-8% of range, and a 5-10% continuation of that range is about 0.35-0.6% gross. With 0.10% OKX round-trip taker plus about 0.05% modeled impact, the expected net is about 0.25-0.4%, which is above 2x cost. Prior art: ETH HL (promoted), Bybit 8-alt panel (passed optimization, avg 0.18%), ETH 4H thrust long-only (promoted), BTC wide-range long with a regime filter (overfit: single-symbol BTC, long-only, filtered). This design deliberately keeps the unfiltered long-short form that survived.