Evening GMT+7. I stopped guessing who to mirror and treated it like underwriting. Inside @easydotfunX I ran a pre‑copy ROI sim wired to my constraints: leader handle, 45‑min window, size curve, maker bias, slippage cap, and a max funding drag. It backfilled last fills, projected tracking error and variance bands, and flagged seat crowding before I committed. Rule set: only mirror if tracking error < 8 bps and crowd score < 0.55, auto‑pause if either breaches mid‑session Setup • Two leaders: high‑tempo clicker vs patient rotator • Same window, reduce‑only exits, identical constraints Measured • Drift vs projected bands • Mirror fill latency and overlap % • Funding paid vs expected, cancel ratio on crowd spikes Results • Patient rotator stayed inside band; net landed within 3 bps of sim, overlap clean • Clicker blew variance when crowding rose; sim warned, auto‑paused, I avoided two sloppy chases • In calm tape, funding drag > slippage; during bursts, slippage dominated and the model caught it Why it matters: copy isn’t rank, it’s fit. Sim first, mirror second. Small asks: per‑leader risk labels and a live tracking‑error light beside the send button. Who else is underwriting mirrors before they click on @easydotfunX?