# analyst1 iteration run 32 (10 Oct 2026, idle) - scripts c1-c9.py in /opt/crew/work/analyst1

Focus: unwatched pools (v4 no-hook native/WETH/USDG, hook farms), plus P2 WETH path tape (8329 pools, 21 Sep-8 Oct, with pull + reserve model from sim2.py).

| # | Lens (new = *) | Test | Verdict |
|---|---|---|---|
| 1 | Held-cap band on unwatched no-hook pools | cap_60m held 400k-1M, depth>=3000: P1M native .22 (n69), WETH .19 (100), USDG .25 (119); P3M 0-.03 with depth. Fee3000 WETH nohook "91% P1M" = depth-0 decoy twins (#279) | discard: no 3M tier, depth gate kills it |
| 2 | * Third-party LP adds as pull-survival marker | 439 nohook WETH/native touchers: every pool has exactly 1 LP sender; 2+ add events n=9 survive 40-100% vs 11% | discard (n=9) |
| 3 | * Hook-address reputation across all 1473 hooks | Top hooks by 1M count (15 pools, 15 reached 1M within 1-2 h) all have real-depth share 0.00 = wash farms. Hooks with real depth (e5e7..e044 168 pools, 0310.. 27) produce 3 / 0 | discard: "runner hooks" are decoy farms; confirms the 88% pull finding |
| 4 | Reserve-backed real-runner map by family (P2, WETH): mult>=30, sustained, reserve>=2 ETH at peak, not pulled 1h, >=8 senders | 254 real: v3 fee100 168 (9.5% of pools), H3000 a80 x4 hooks 59 (of ~110 pools, 55%!), v4 fee100 nohook 16 (1%), rest ~10. a80: 6 pools/day, 3.5 real/day, no decay 21 Sep-7 Oct | info: only H3000 a80 has per-pool rate >50%; its open question is still sellability (scripted ramp, Ben $5 sell test #221) |
| 5 | * Dip-buy inside a confirmed runner (peak >=20x held >=5 prints/300 s, price falls to 0.5x peak; exit back to peak x0.75 / stop .7, 4h, sim2 impact+pull) | n=225-506, mean 1.06-1.08, halves 1.05/1.07, random-time baseline same pools 0.98; WETH v3 fee100 1.05, v4 fee3000 1.24 (n25) | park: +8% over baseline = 1.2 sigma, not a runner hunt |
| 6 | First-12-print structure on v3 fee100 (1771 pools) | prints in first 3 s: >=7 of 12 -> real runner 16% vs 3%; reserve after 20 prints <0.5 ETH 2.5% .. 5-20 ETH 26% .. >20 ETH 48% | duplicate of known first-minute speed/volume filters (#221, #277) |
| 7 | * Early big seller (inversion) | one sell >10% of gross inflow inside first 20 prints, reserve>=2 ETH: 7.7% (n26) vs 20.4% (n515); halves 10%/0% | park: skip-flag, n too small |
| 8 | Regime/supply of the best family | a80 pools/day 4-11, real-runner/day 0-6, no decay; v3 fee100 60-138 pools/day, 4-17 real | info |
| 9 | Request: sell-probe stream (below) | - | request to ppfeeds |
| 10 | Not re-run: tier ladder / pull / name / wallet lenses already in ledger (#225-248) | - | - |

Net: no new rule ready for testmgr. Honest finding: every nominal-runner pocket outside v3/H3000 (native/USDG nohook, 15-pool hook farms) fails the depth/pull gate; the only unwatched-by-count pocket with a high real rate is the H3000 a80 family (4 hook addresses: ...4a80, ...ca80, ...8a80, ...0a80), where the single open question is that nobody has sold into it.
Pitfalls: P2 is WETH-quote only (no native/USDG); max_cap in pool_paths is last-print/one-print contaminated; cap_60m in pull pools is a stale last print.

## Request to ppfeeds: periodic sell-probe (extends #266)
What: for every pool that has touched 100k cap (any venue/quote), every 60 s for its first 4 h (and at the moment of each 100k/300k/1M cross) run an eth_call quote/sell simulation of selling $20 and $100 of the token via the pool's router with a balance override. Log: pool, ts, block, ok/revert flag, amount_out (quote units), implied price impact vs last swap price, pool reserve (quote side).
Period: forward from now, plus a one-off backfill at the 100k cross for the last 14 days if archive state is available on a free RPC.
Format: sqlite /opt/sw/ppdata/sellprobe.db table probe(pool,t,block,sz_usd,ok,out_quote,impact,reserve_quote); radar_hist gets a column sellable_100k.
Test I will run: (a) H3000 a80 and v4 nohook: share of pools that stay sellable at +10/+30/+60 min after 300k; (b) whether a falling probe impact or reserve (pre-pull drain) leads the pull by >30 s; (c) skip rule: do not enter pools with ok=false or impact>15% at $20; compare mc5 band mean 1.0 vs filtered on 200+ pools.