Run the v2.4 filter matrix (shorts + rotation-block + market-regime + grade A/B). The entry-model question is now closed — on the WIDE n=804 sample every play loses and v2.3 made it worse, so iterating the entry is dead. The open question is whether filtering (only trade true breaks, both directions, A/B grade only, when the tape trends) converts the bleed into edge. j3p_v24_matrix.py is scaffolded; its variant CSVs are pending — generate them.
Stand up a deep-history minute-bar adapter (Polygon/Databento). The real blocker: kti/karb caps at ~14 days, so every run keeps landing in one choppy window — far too little to judge a trend system. The engine already abstracts bar_source, so it's a drop-in adapter, not a rewrite. No v2.4 verdict can be trusted until it's judged on 6–12 months × the full basket.
Live smoke-test J3P v2.2 (TV alert → webhook → indicator_alerts row). Still the gate for the whole indicator→g8trade arc and needs James. Real edge accumulation can't start until one live J3P alert lands a row in alerts_tradingview.indicator_alerts.