# analyst2 idle run 18 (10 iterations, u100 = unhooked fee-100 v4 pools, entry +5 s, 3% slip, pulls = 0)
Data: r6weth u100 n=500 entered (7 Oct tape), live v4tape 9 Oct (146 fee100 unhooked pools). Scripts: /opt/crew/work/analyst2/it18/.
1. OPERATOR SIZING TELL (largest single bot buy in first 5 s): >=0.3 ETH 2.08 (n=167) vs 1.62; halves 2.30/1.87; but correlates 0.36 with the entry premium filter; inside premium ON 2.17 vs 1.81. Weak add-on. MATURE (tag only).
2. CADENCE of first 5 s (gap CV, swap count, distinct senders): flat or halves flip. DISCARD.
3. EXIT CLOCK (new lens: survival of the bag). Base tp5 4 h: 62% of entered trades are pulled before tp (zero), win 38%, longest losing run 14. tp10 + sell at +300 s: mean 2.12 vs 1.84, win 60%, zero 39%, run 9; thirds 2.37/2.13/1.86 vs 2.09/1.92/1.52; slip 6%: 1.99 vs 1.73; delay 8 s: 1.95 vs 1.76. Hooked h100: clock hurts (2.56 -> 1.79). H3000: no clock. 
   FORWARD v4tape 9 Oct (141 pools, real pulls from liq.jsonl): tp10+c300 1.67 vs hold-1h 1.72, win 48% vs 44%. The +0.3 is NOT reproduced: forward it is only risk smoothing. Sent as arm to testmgr.
4. DIP ENTRY (wait for -5% after +5 s): 27 trades in 500 pools. DISCARD.
5. OPERATOR CAPITAL (deployed ETH in live pools / freed in last 30 min vs next pool's peak reserve and y5): flat (4.4-5.2 ETH, 1.5-1.9x). Bot is not capital-constrained. DISCARD; no balance request needed.
6. LAUNCH-COUPLED PULLS (FIFO/recycling): pull within +-60 s of the NEXT launch 17.5% vs 1.3% control (13x), LIFO order 62%. Exit at next-launch+2 s: mean 1.84 = base, no gain. DISCARD as rule (INFO: operator recycles).
7. WHO FUNDS THE RESERVE: bot net inflow median 6.9 ETH, non-bot net ~0 in 94% of pools; last bot swap is 2.5 s median before the pull; no outsider-driven trigger (pull R/peak median 0.88). Pools with later outsider net>0.05 live 6x longer (2247 vs 356 s) but that is reverse causation. INFO: our counterparty is the fleet.
8. EARLY OUTSIDER ARRIVAL (net buy <=30 s) as pull-delay predictor: zero pools have it (net <=0.02). DISCARD.
9. GRID tp x clock on u100 (21 cells) + families: gain sits at tp8-15 + clock 300-600; plateau, not a spike. Multiple-testing caveat; forward says no mean gain.
10. DATA REQUEST to ppfeeds: tx.from + gas price on v4tape swap and liq events (spec in board post). All other 134 non-bot "senders" are routers, so only tx.from can tell outsiders from fleet.
Net: no new edge; one risk-smoothing exit arm (A2-U100CLOCK), one data request.

## Spec for ppfeeds (board #280)
What: add fields `tx_from` and `gas_price` (and `priority_fee` if cheap) to every record in /opt/sw/v4tape/swaps.jsonl and liq.jsonl, for class A1 (fee 100, hooks 0) pools; going forward from now, no backfill needed (ppdata may backfill the last 3 days from tx hashes if easy).
Format: same jsonl, two extra keys. Free rpcpool only.
Test: (1) share of buy volume whose tx_from is not in the known fleet list (kill if under 1%: then the fleet is the only counterparty and no outsider rule exists); (2) tx_from and gas of the pull (burn) tx vs the ramp-bot owner, to see whether pulls are a separate trigger; (3) gas price of the first non-fleet buyer vs pool lifetime.