Idiosyncratic Attention-Shock Continuation, BTC-Beta-Hedged, HALF-SIZE with a hard netted-hedge cap: long a Bybit alt perp / short beta x BTCUSDT.BYBIT for 5 days after a >=2.5-sigma BTC-residual UP day on >=2x 30-day median dollar volume (21 pre-2022 Bybit USDT perps, 1-DAY, all-fixed plan)

Revision of the Research Lead's Bybit idiosyncratic attention-shock continuation panel. The analyst found the implementation sound and the signal intact; the premise failed on its RISK SPECIFICATION. The contract fixed leg_notional_pct = 0.10 with max_concurrent = 6 and beta up to 2.0, so the netted BTC short could grow large, and the drawdown broke the pre-registered 20% limit in the latest walk-forward windows. With no risk-admissible candidate there, the optimization was rejected. The report-only sweep showed Sharpe flat in leg_notional_pct, so this revision changes ONLY the sizing and risk construction. Signal, universe, timeframe, direction and mechanism stay as registered. MECHANISM (unchanged): limited attention and slow capital rotation. An alt that jumps far more than its BTC beta predicts, on unusually heavy dollar volume, has had a coin-specific information or flow event (listing, partnership, upgrade, narrative rotation, buyback). That repricing diffuses over several days. Long the alt, short beta x BTC: the bet is only on the residual continuing. The full-history P&L split showed the edge sits in the alt leg; the BTC hedge contributed noise of up to about +/-$57k per year at the old size. CHANGES: (1) leg_notional_pct 0.10 -> 0.05 of equity per alt leg. (2) New fixed rule: the netted BTC hedge, sum over open events of beta_i x alt_notional_i, is capped at btc_hedge_cap_frac = 0.30 of equity. If a new entry would push it above the cap, every open alt leg and the new one are scaled down pro rata (alt notional and their hedge together) so the cap holds; scaled-down legs below min_notional are skipped. With 6 x 5% x beta<=2.0 the uncapped worst case is 60%, so the cap binds only in crowded, high-beta clusters. Every position stays far below 100% of equity by construction, satisfying risk-control route (b) (signal/time exit with capped sizing). (3) Leverage 2.0 -> 1.0 (gross worst case is 30% alts + 30% BTC hedge = 60% of equity). Everything else is frozen as registered: hold_days 5, z_threshold 2.5, vol_ratio_min 2.0, beta_lookback 60, resid_sigma_lookback 60, volume_median_lookback 30, beta clip 0.5-2.0, max_concurrent 6, min_listing_days 90, point-in-time universe. The plan stays ALL-FIXED: no sweep point justifies re-tuning, and every extra look raises the deflation bar. HONEST STATEMENT: the most recent period (2026 YTD) is negative in both the pre-study (-2.09% per event) and the failed backtest. The signal is deliberately NOT changed to recover it, because that would be fitting to the held-back period. This is a re-registration of the same signal at a sizing the evidence says it can carry. PRE-REGISTERED EVIDENCE COUNT: about 480 alt events on about 325 distinct entry dates (2021 to 2026-10), unchanged by sizing except for rare cap-skips. COST ARITHMETIC: pre-study mean net residual +1.51% of alt notional per event after 0.22% for both legs' round trips (trimmed mean +0.88%), i.e. gross capture about 1.1-1.7% vs 0.22% two-leg round trip, more than 4x cost. Expected per-trade avg on the backtest's per-position accounting about 0.5-0.8%. PAPER KILL CRITERIA (pre-registered): after at least 40 paper events, stop if the average alt-leg trade (hedge included) is below +0.25% of alt notional. Execution: market orders at the signal bar's daily close; on a 24/7 venue that is the registered equivalent of a next-bar-open fill (capability next_bar_open_fill: use_equivalent). The 5-day exit is a time stop tracked from the event's entry bar; the BTC hedge is netted on one BTCUSDT.BYBIT position (netting account; hedge_mode not needed because only the net BTC short matters) and resized when an event opens or closes, subject to hedge_rebalance_min_usd.