ETH Deribit Accelerating-Vol Gamma Scalp: Buy a ~10-DTE ATM Straddle Only When Short-Window Realized Vol Is RISING Through Both the 7-Day RV and DVOL (not after a spike has peaked), Band Delta-Hedge on ETH-PERPETUAL.DERIBIT (same venue), Exit at 2 DTE or When 24h RV Falls Back Below DVOL; flat ~70% of the time

Source: 'What really is Gamma scalping?' (Hans, quant.stackexchange Q37406, 2017-12-17), mechanism from the accepted answer by AlRacoon (a/37421) and the answers by OGC (a/37432) and Bram (a/37413). The question's own P&L identity is the whole mechanism: a delta-hedged long option earns P&L = integral of 1/2 * Gamma * S^2 * (sigma_realized^2 - sigma_implied^2) dt. Theta is the market's price for the expected gamma-scalping income (OGC). With discrete re-hedging the P&L is path-dependent and centred on Vega * (RV - IV) (Bram). So being long gamma pays only when realized vol over the holding window exceeds the implied vol you paid. Nothing else in the article is a signal. The research question is: when is sigma_realized over the next ~8 days predictably above the implied vol quoted today? The corpus has already tested and falsified the naive form of this. 'BTC LONG-GAMMA When Implied Vol Is CHEAP vs Realized' bought a BTC straddle when DVOL < trailing RV and hedged on BTCUSDT.BINANCE. Over 38 cycles the option legs lost $8.4k and only 5/38 cycles were profitable, while the hedge itself was fine. The analyst's diagnosis was that the gate 'does not select weeks whose subsequent realized vol exceeds the implied vol paid'. The reason is structural: trailing RV > IV is true mostly just AFTER a vol spike, and realized vol mean-reverts down faster than implied does. The gate therefore bought gamma at the top of the vol cycle. The ETH twin (EthDeribitCheapGammaBandHedgedStraddle) used the same trailing-RV gate and never produced a measured report (verifying_stuck_3_timeouts). This hypothesis changes exactly that premise and nothing cosmetic. Buy gamma only when realized vol is ACCELERATING, which means volatility clustering is in its build-up phase, not its decay phase: RV_24h > RV_7d > DVOL, and RV_7d has risen versus its value 3 days ago. Realized-vol persistence (GARCH/HAR: the next-week RV loads positively on the last day's RV) says the forward RV then stays above the trailing weekly level. Implied vol on Deribit is known to lag realized in the first days of a regime change, because dealers reprice IV off a smoothed estimate. If RV has just crossed above IV and is still rising, the IV paid is stale. Trades ONLY on DERIBIT, the venue where the signal (DVOL) and both legs live. Option legs are ETH weekly/Friday ATM call + put from the rolling-options chain expander, entered with ~8-12 DTE at the strike nearest spot. The delta hedge is ETH-PERPETUAL.DERIBIT (same venue, inverse perp, hourly data since 2019). There is NO cross-venue leg: the BTC failure hedged on Binance, and L143 bans the signal-on-A/trade-B pattern. The realized-vol inputs come from ETH-PERPETUAL.DERIBIT 1H bars; the implied input is the deribit_dvol supplementary key (ETH), read point-in-time via supp_as_of. The decision clock is 1H bars. Why ETH and Deribit (breadth and quotas): options are 5.7% of the corpus against a >=15% target, and BINANCE is 66% against a <=40% cap. ETH on Deribit is the second-deepest crypto option market. BTC was used for the falsified version, so the retest moves to a different underlying rather than re-mining BTC. Flat state: flat whenever the acceleration condition is absent, and no new straddle is opened inside 2 DTE. Expected time in market is ~30% (roughly 1 in 3-4 weekly cycles qualifies, with a ~6-day average hold). Data-age caveat (stated up front): per the EthDeribitWeeklyDeltaHedgedAtmStraddleVrp_v4 notes, the catalog's ETH weekly option candles start ~2024-07, which gives ~115 weekly cycles. At a 25-35% qualification rate that is ~30-40 straddles in total and a holdout of ~7-9. This is below the 10-trade holdout floor unless the qualification rate is at the upper end. min_trades below counts option legs + hedge fills as the engine does; the number of independent straddle cycles is the real sample and is stated honestly. The mechanism cannot be measured with more power until more chain history accrues. Do NOT widen the gate to manufacture trades: that would recreate the falsified trailing-RV version.

This implements P&L ~ 0.5*Gamma*S^2*(RV^2-IV^2) by buying gamma only in the build-up phase of a vol cluster, not after a spike (the falsified trailing-RV>IV gate). Every input is on one venue: RV from ETH-PERPETUAL.DERIBIT 1H returns (O(1) running sums), IV from deribit_dvol read point-in-time by bisect, legs from the RollingOptionsStrategy chain expander (dte_target 12, ATM C+P), and the hedge on the same perp. Option-premium bars are used only when stamped at or before the primary bar. Premium units are auto-detected (coin <1 is multiplied by spot, otherwise treated as USD). If DVOL is missing the run goes flat with a data_unavailable error log; there is no price-only fallback and no underlying smoke trade. Three implementation choices go beyond the brief and are stated here. (1) dte_target 12 with min_entry_dte 6 widens the entry window to 6-12 DTE (brief: ~8-12) so the overlapping weekly windows cover about 6 of every 7 days; the strike is fixed at expiry-12d, so a max_moneyness 0.08 guard skips entries whose strike has drifted off-ATM. (2) min_hold_hours 24 keeps the RV_24h<DVOL exit from firing on RV24 computed mostly from pre-entry returns. (3) Everything is sized off premium at risk. Sample-size caveat from the hypothesis: ETH weekly chain data starts ~2024-07, so expect ~30-40 straddle cycles. The gate was deliberately not widened.