# A2 iterate run 19 (10 Oct 2026): 10 iterations, scripts /opt/crew/work/analyst2/it19/

Base: u100 (v4 fee100 unhooked, WETH-quoted tape to 7 Oct), entry +5 s, tp5, stop 0.5, pull-aware, n=500 pools, mean 1.84.

| # | Lens | Result | Verdict |
|---|------|--------|---------|
| 1 | Other fee classes (2500/10000/20000/...) | pickle only has the known classes; each other class has <70 swapped pools in r6weth. Hooked fee3000 +5 s: tp5 4.72 n=93 (known) | DISCARD (n) |
| 2 | LP staged pull = early warning (NEW) | 1369 pulls: 1368 are ONE removal event (0 partial steps, lead 0 s) | DISCARD |
| 3 | Overlap of my own arms (NEW, method) | premium>=1.2 vs reserve>=1 ETH at entry: Jaccard 0.84 (same signal). BUSY60 and RES60 use 60 s of data: look-ahead at +5 s entry, only valid with +60 s entry. 0 flags 1.12 vs >=2 flags 2.14 | INFO: one signal, do not stack arms |
| 4 | Inversion: autopsy of losers | px(+5s)/px0 quartiles win rate .28/.34/.44/.46 mean 1.35/1.64/2.13/2.25; creation lag short better (2.02 vs 1.55); R0 low better; nothing beyond the known premium | INFO |
| 5 | Exit latency (NEW, sell side) | tp3 hits n=305: pull comes >=62 s after the hit in 90%, <5 s in 2 of 305. Selling 0/2/5/10/30 s late: 1.74/1.84/1.86/1.88/1.89 (price keeps rising). tp5 n=190: 0 of 190 pulled within 5 s | INFO positive: exit latency is not the risk |
| 6 | Archive sell replay (NEW tooling) | eth_call + stateOverride works at past blocks: native fee100 pool 40k blocks back OK (sell 0.01% supply returns ETH). Old WETH-ERC20 u100 pool 16M blocks back: empty revert x9 (cause open: depth or quote type). | DATA REQUEST to ppfeeds (below) |
| 7 | CORRECTION | v4tape swaps: sender 0x8876789976decbfcbbbe364623c63652db8c0904 = Universal Router (60% of all 81k swaps; sellsim UR constant). "0x8876 is the first swapper in 500/500 pools" is NOT evidence about an operator bot. Operator identity only via tx_from. | CORRECTION in ledger |
| 8 | Buyer diversity in first 60 s (txfrom, 232 pools) | distinct tx_from rises with swap count; within swap-count terciles low/high diversity 1.14/0.81, 0.89/1.49, 1.11/1.09 (+60 s entry). Noise | DISCARD |
| 9 | Names / theme (v4tape token_name just appeared) | 22 of 272 pools named. Same name relaunched 6x in 7 min ("opcode", A1 class, nonce 2 each, dead in 6-258 swaps); H3000 EOA 0x26e813 launches fintech/AI-named pools back to back (nonce 62385 -> 62645) | MATURE: rerun at >=150 named u100 pools (dataeng #266 controls) |
| 10 | Synthesis | no new paper arm; FAST5 stays the single u100 idea, its two sub-signals are one signal | - |

## Data request (ppfeeds, owner of /opt/sw/sellsim)
Historical sell replay, `sellsim_hist`:
- What: for v4tape A1-class pools (fee100 unhooked, native quote) since 9 Oct 00:00 UTC: (a) every pool that hit tp3 (price >= 3x the +5 s tick price): run the existing sellsim call at the hit block; (b) the same at the +5 s entry block as control; (c) 100 random pools without a hit at +120 s. Sell amount = 0.25 ETH worth of tokens at that block's pool price (not 1% of supply).
- Needs: block tag in eth_call (sellprobe.call uses "latest"); pool key from Initialize log of the creation block (works up to ~40k blocks back).
- Format: /opt/sw/sellsim/sellsim_hist.jsonl, one row per call: pool_id, kind (entry|tp3|ctrl), block, amount_in, eth_out, sell_ok, revert. Mirror to fw/a1feed/.
- Test I run: realized eth_out / (tick price x amount x (1-fee)) at tp3 blocks. Pass: median >= 0.85 and sell_ok >= 90% in hits. Fail (sell_ok < 70% or median < 0.6): FAST5 tp numbers are a mirage, tell testmgr and Ben. Purpose: gives n~100+ sell proofs now instead of waiting for 60 live rows, without money.