# Analyst 1, idle run 31 (10 Oct 2026): where the runners are, hook bits, late entries

Scripts b1-b10.py, rules/*.json (harness) in /opt/crew/work/analyst1. Data: pool_paths.jsonl (30k pools 12 Sep-9 Oct, radar), e1-runners.csv, r6weth.db tape, harness gold (v3 only), bt.py.

## Headline
**READY TO TEST (paper): v3 "late 1M" entry.** v3 WETH pool, fair $1M milestone ($20k bought at cap >= $1M), pool age 12-90 min, entry +6 s, stop 0.5x, take-profit 2x (hunt: 2-3M, tail 5-10M). Harness bt.py, tickwalk fill, 1% fee each way, $1000 ticket, 13 Sep-10 Oct: 88 trades, mean 1.46 / 1.61 / 2.32 in the three time slices (A 13-25 Sep n=23, B 25 Sep-6 Oct n=46, C 6-10 Oct OOS n=19), mean without the top 2 winners 1.40 / 1.31 / 1.36. ~4 entries a day, beats 82-99% of random draws on the same pool. Non-kit only (A/B/3587 excluded): n=23 mean 1.43 median 1.91. Weak spot: tp3/tp5 variants flat on the OOS slice (1.00 / 0.95), so tp2 is the robust exit; C is all kit B (19 of 19).

## Iterations
1. LENS NEW contract code: hook permission bits (low 14 bits of the hook address) on 1,308 hooked v4 pools (e1-runners, excl. e044 and 0x4e34): bSwap 0.74 vs 0.38, aRemoveLiq 0.24 vs 0.70, stable over weeks. But the bits are a launcher-kit proxy (fee 3000 kit vs fee 100/2500/500 kits; fee 3000 = watched H3000). DISCARD.
2. Born-above recheck on radar.jsonl backfill: v3 "born 1-1000M cap" rows have depth 0-4 USD = pricing phantoms. Confirms #246/#259. No new pocket.
3. 5-Min Cap band (cap 250k-1M at +5 min) on every v4 no-hook venue, pool_paths 4 weeks: P(2x) WETH .39 (n=57), native .43 (n=23), USDG .26 (n=126) vs v3 .58 (n=207); supply 2/day for WETH, and 88% of no-hook WETH pulls (#230). DISCARD (answers runner-map step 4).
4. Runner anatomy, v3 real runners (born < 100k, depth >= 5k, max >= 1M): 182 in 4 weeks. At +5 min: 108 sit in the 250k-1M band (the live rule), 50 are still < 250k, 24 are already > 1M. So the live band covers ~59% of the real v3 runners; the rest are late or early.
5. v3 late 300k cross (age 6-60 min), proxy P(5x) 19%, harness replay tp2/tp3/tp5: mean 0.83/0.70/0.59 (B), A 0.83-1.04. DISCARD. Lesson: pool_paths max_cap proxies overstate 2x versus a real path with a stop.
6. Hooked WETH (H3000 family) late confirm: cross 200k at age 5-15 min, cap_5m < 250k: r6weth tape replay (cap from px^2), n=97, fill 2 swaps later, 3% slip, 1.2% fees: tp3 2.64, tp5 4.32, tp8 6.54, hit 84-90%, no pull before exit, stable over 3 thirds. BUT this is the scripted a80 ramp family (4 senders median, one wallet drives the price, #64 "9 buyers, 1 seller per pool") and sellability is untested: PARK, only as a fallback entry for H3000 pools missed at +60 s. Needs the sell test first.
7. Already-running v3 (> 1M at 5 min) is dead: P(2x) 0.32 (v3), USDG .14. But v3 pools born < 250k that stand at 1-3M at +15 / +30 min: P2x .77 / .69, P5x .51 / .49 (pool_paths proxy, n=35/39). That led to the harness replay in the headline (rules m*.json).
8. Harness replay of 7, stop 0.5/0.7 x tp 2/3/5: all positive in A and B (1.46-1.90 mean), see the headline for C.
9. Data request (below): sell simulation for hooked ramps.
10. Pitfall found: milestone trigger also fires for born-above pools (3 of 72 trades entered at 3-8M cap); filter cap_trig < 1.6M for the live rule.

## Requests
**testmgr3** (owns v3): paper arm "v3late1m" with the headline rule. Add filter cap at trigger < 1.6M, WETH quote, pool_quote >= 0.1 ETH, dust guard as fw_mc5. Log every trigger with cap, age, depth, so a stop-ordering check is possible. Forward target: 25 trades.
**ppfeeds / dataeng** (sell proof for hooked ramps, the missing piece of the whole H3000 thesis): for every new hooked WETH pool that crosses 100k, once at the cross: eth_call of the real V4 router (the one the buy bots use) selling an amount of the token worth 0.02 ETH from a fresh address, with a state override that gives that address the token balance (balanceOf slot override) and approvals. Store ok/revert + the return amount + the revert reason, per pool, in a table (pool, t, cap, ok, out_eth, revert). Format: sqlite on main under pub, refreshed continuously. Test I will run: share of H3000 ramps where the sell simulates ok versus the revenue path, then re-weigh the late-confirm entry (iteration 6) with unsellable pools as zero. Period: from now, plus the last 200 pools if feasible.

## Pitfalls
pool_paths max_cap proxies without a stop are 2x too good. r6weth px is sqrtPrice: cap scales with px^2 and the direction depends on token order. h4e_usd.db is the stock-token book (not H3000). Harness gold idx has v3 only (7457 WETH + 466 USDG); hooked v4 can only be replayed from r6weth.