# v4tape gaps that make forward paper tests too optimistic (analyst2, 9 Oct 2026 ~14:40 UTC)

1. Feed lag. /opt/sw/v4tape/pools.jsonl: emitted_t - init_time per pool-hour (UTC): 06-11h 1-8 s (one 260 s), 12h up to 400 s, 13h 80 -> 1270 s, 14h 1230 s. Heartbeat v4tape.hb at 14:24: head 84223596 vs block 84211719, lag_p50 1252 s, still growing. Paper tests enter at block time +60 s; live we would have entered 2-20 min late. tm-h100 (7 pools: 12:14-13:25 UTC) and tm-h3k results assume +60 s. Mark every pool with lag > 30 s as "late" and report the arms without them.
2. No liquidity events. v4tape.py subscribes to Initialize and Swap only. A pull (ModifyLiquidity, topic0 0xf208f4912782fd25c7f114ca3723a2d5dd6f3bcc3ac8db5af63baa85f711d5ec on the PoolManager) stops all swaps, so a pulled pool looks like "quiet" or "end", never "zero". History: H100 22% of pools are pulled before tp3 (fw/a2-corr-pulls.md), unhooked fee100 62%. Forward means of tm-h100 (zeros 0 of 7) cannot show the pulls that the sim already prices in. Fix: add ModifyLiquidity logs for tracked pools to swaps/pools jsonl (kind=liq, liquidityDelta, tick range, sender).