BTC Variance-Risk-Premium Regime Long on the COIN-M Inverse Perp: hold BTCUSD_PERP.BINANCE_CM while Deribit BTC DVOL minus 7-day realized vol is in the top 30% of its trailing year, exit when it falls below the trailing median (1H bars, one decision per day at 00:00 UTC, long-only, flat ~78% of the time)

Explicit cross-underlying REPLICATION (L202 replication objective, stated up front) of the promoted 'ETH Variance-Risk-Premium Regime Long on Deribit's Own Linear Perp' (run 00add348, paper_stage, 36 trades, avg 2.66%/trade, PF 1.65). That strategy's mechanism was tested only on ETH; the sibling on ETHUSDT.BINANCE that used a 90-day ranking window failed overfit (holdout negative), while the 365-day ranking version survived. This hypothesis asks whether the SAME frozen rule (DVOL minus 7-day RV, top-30% of trailing 365 days to enter, below trailing median to exit, one decision per day) carries information on the other DVOL underlying, BTC, traded on Binance COIN-M (BTCUSD_PERP.BINANCE_CM, inverse, ~0.06% RT taker). Signal: VRP_t = Deribit BTC DVOL (supplementary key `deribit_dvol`, BTC, hourly, history from 2021-03) minus annualized 7-day realized vol computed from the strategy's own 1H closes (stdev of hourly log returns over 168 bars x sqrt(8760) x 100). Each day at the 00:00 UTC bar close the strategy ranks VRP_t against its own trailing 365 daily 00:00 observations. Read DVOL point-in-time with supp_as_of (latest hourly DVOL value whose timestamp is <= the bar's close). Why BTC despite BTC's corpus share: DVOL exists only for BTC and ETH, and ETH is already covered, so the replication requires BTC; the venue (COIN-M, 2.5% of the corpus vs a >=5% target) and the information source (options-implied vol, no OHLCV-only signal) are under-represented. Fill convention: decisions at the 00:00 UTC bar close are market orders that fill at that close (the platform's faithful equivalent of the next bar's open on a 24/7 venue). Risk route (b): signal-based exit with capped sizing, one position, notional = 100% of USD-converted account equity, never levered; sized in whole USD-face contracts with contracts_for_notional(). Pre-study by the Research Lead (hourly COIN-M closes + BTC DVOL, frictionless, daily decisions): baseline rule gives 46 round trips 2022-2026, mean +2.6% per trade, median +1.6%, win rate 0.63, time in market ~22%, positive in every calendar year (2022 +32%, 2023 +24%, 2024 +36%, 2025 +16%, 2026 YTD +12% summed trade returns); neighbouring settings (entry quantile 0.6-0.8, RV window 5-14 days, ranking window 180/365) were mostly positive with mean trade returns 0.4-3.7%, the 14-day RV variant lost in 2022. This is a sanity check that the rule fires and is not obviously degenerate, not evidence of an edge. Matched control for the analyst: hold BTC for the same number of days at random entry dates (drift alone ~0.06%/day x ~10 days median hold ~0.6%/trade) - the rule must beat that, not merely be positive.

Long-only strategy that trades BTC's variance risk premium on the Binance COIN-M inverse perp BTCUSD_PERP, on 1H bars. The premium (VRP) is Deribit's BTC DVOL minus the annualized realized vol of the strategy's own 1H closes over rv_days*24 hourly log returns (stdev x sqrt(8760) x 100, kept with O(1) running sums). DVOL comes from deribit_dvol and is read point-in-time as the last completed hourly candle, supp_as_of(t - 1h); a value older than 6h counts as missing. Once a day, at the 00:00 UTC bar close, today's VRP is ranked against the earlier daily values from the trailing rank_days calendar days (the fraction strictly below it). The strategy goes long when the rank is at least entry_quantile (0.70, the top 30%) and goes flat when it falls below exit_quantile (0.50, the trailing median). It enters and exits only at this daily decision. Missing data: on a day with no DVOL it takes no action. The ranking needs at least rank_days - max_missing_dvol_days values in its window. A held position is closed after more than max_missing_dvol_days decision days in a row without DVOL. Size is notional_fraction (1.0) x account equity x leverage (1.0), in whole $100 contracts via contracts_for_notional(). Fill substitution: a market order at the 00:00 bar close, which on a 24/7 venue is the same as the next bar's open. calculate_signal returns the current VRP rank on every bar so the signal is continuous; decisions use only the 00:00 value.

This is a frozen replication of the promoted ETH DVOL-VRP regime rule on BTC. It keeps the same signal (DVOL minus 7-day realized vol), the same 365-day daily ranking, the same 0.70 entry / 0.50 exit thresholds, one decision per day and long only. It has no stop and no minimum hold, because the hypothesis specifies neither. Risk route (b): a signal-based exit with one position capped at 100% of equity, unlevered. Parameter names and fixed values follow the optimization plan exactly. Tunables are rv_days, rank_days, entry_quantile and exit_quantile; fixed are decision_hour_utc, notional_fraction, max_missing_dvol_days and dvol_max_staleness_hours. The code clamps nothing, so there are no _param_bounds. Interpretation: max_missing_dvol_days is used as the tolerated DVOL gap. It sets how many observations the rank window may be missing, and it forces the position flat after a longer gap while long. Equity is read with get_account_equity() (the venue's USD-stable balance). At notional_fraction 1.0 the initial-margin check may trim the entry by about the taker fee; that is logged as a cap, not a failure. The sandbox window (~20k 1H bars) spends about 365 days warming up the rank, so expect only about 8-12 trades there.