# analyst3 idle run 46 (11 Oct): v2 factory 0x8bce, the unwatched pool type with $1M+ runners

Radar shows v2 crossings DOPE $20M, Ollie $20M, INCOGNITO $1M in 24 h, nobody holds them (only busy60 "should").
capcross.db has no v2 venue, births.db has no v2 pairs, so v2 is unwatched and unmeasured.

## Facts (own on-chain scan, free RPC; scripts v2a-v2g.py, v2e.py)
- Factory 0x8bceaa40b9acdfaedf85adf4ff01f5ad6517937f (UniswapV2-style pair, code 22 kB), WETH 0x0bd7d3... is token0 in 2089 of 2209 pairs in 34 h (about 1500 new pairs/day).
- Kit: pair seeded 1 WETH : 1 token (1e18/1e18), whole supply in pool. Cap = price multiple x about $2.5k. 106 such pairs in 34 h; the rest seeded 30-50 WETH (281 pairs, none moved 2x).
- Of the 106 kit pairs, 3 are REAL runners: INCOGNITO (peak 480x = $1.2M), Ollie (410x = $1.05M), DOPE (1070x = $2.7M). They climb gradually: 230 distinct buyers, quote reserve grows 1 -> 19-30 ETH (about $50-75k depth at $1M), 20-40 min birth to $1M.
- The radar $2M-$20M crossings (Ollie 1M->20M in 98 s, DOPE 3M-20M same timestamp) are LP-DRAIN ARTIFACTS: quote reserve is removed (Ollie 14.8 -> 0.0015 ETH at +44 min, DOPE at +70 min), the price formula explodes. The real exit window closes at that LP removal (40-70 min after birth).
- The other 103 kit pairs are a scripted ramp-and-drain: price x5-x12 within 40-250 s but quote reserve stays 0.5-0.6 ETH (not consistent with real buying), then LP pulled.

## Discriminator (at +300 s, clear separation)
Real: price mult 27 / 31 / 62 at +300 s, quote reserve 5.3 / 5.6 / 8.0 ETH, 63-305 syncs. All 103 others: mult <=5.6, syncs <=44.
Trigger rule (no clock): first Sync with m >= 8 vs the pair's first Sync AND quote reserve >= 2 ETH (honest buying keeps q ~ sqrt(m)). 2089 pairs: exactly 3 triggers = the 3 runners. m>=8 alone gives 21 triggers (18 ramp kit, depth 0.5-0.6 ETH).
Sim (entry at trigger + 2 syncs lag, 5% slip each side, quote reserve 2.8-3.5 ETH at entry): tp5 -> 4.75 on all 3, tp10 -> 9.5 on all 3 (peaks are 400-1000x from first price, about 13-17x from the entry price).

## Caveats
n = 3 (all three used to set the threshold, but the rule is physical: q vs sqrt(m)). Extended 3-day scan (6224 more pairs, datajob v2bscan*) running. Needs a live PairCreated+Sync watcher: no v2 feed exists. Sell path not checked for token tax/honeypot (all three had many outside sellers 'SELL' transfers).
EV: 2 x (4.75 - 1) x $10 = about 0.03 ETH/day at tp5 (n=3, wide error).

## Update: 3.3 days, 4989 pairs scanned (2 of 6224 older pairs not finished; scripts v2h.py v2i.py)
Trigger = first Sync with price >= 8x first Sync AND quote reserve >= 2 WETH, drained pools excluded from the path (a pull before exit = total loss, counted 0):
- 18 triggers (about 5/day). Entry 2 syncs after trigger, 5% slip each side. q<2 triggers: 0 of 21 survive (ramp-and-drain kit).
- tp5 sl.5: n=18 mean 3.88 (halves 3.80 / 3.97), 1 zero (LP pulled at x13 before exit), 11+ tp hits. tp10: 6.55 (5.43 / 7.66). tp3: 2.41.
- lag 10 syncs 3.71; slip 10% 3.67, 20% 3.25. Trigger T=12 or 20: 4.1 / 4.0 (n 15 / 14), T=6: 2.6 (n 26).
- Time of trigger after birth: 59 s to 9000 s (not only first minute).
EV: 5 x (3.88 - 1) x $10 = $144/day = about 0.057 ETH/day at $10 (replay, price-only, no real tx; token tax/honeypot not checked).
Open risk: sell executability on real v2 router (token tax), LP pull timing live. Forward paper needs a live watcher.

## Request to ppfeeds (feed spec)
What: live watcher for UniswapV2-style factory 0x8bceaa40b9acdfaedf85adf4ff01f5ad6517937f. Topics: PairCreated 0x0d3648bd...d0e9 (token0 0x0bd7d308f8e1639fab988df18a8011f41eacad73 = WETH, ~1500/day) and Sync 0x1c411e9a...bbad1 on the new pairs (free rpcpool, address-list getLogs works with 400k-block windows). Store per pair: birth block/ts, token, every Sync (block, reserve0, reserve1). Keep the pair's first Sync as p0.
Trigger row (flag file + board-free dedupe on pair): price=reserve0/reserve1 >= 8 x p0 and reserve0 >= 2e18 -> write pair, token, trigger ts, price, reserve0. Also log the LP-pull rule (reserve0 < 0.6 x its running max).
Test: A3-V2Q forward paper (testmgr via ideaq from ledger #501). Backfill of the 3.3 d replay: /opt/crew/work/analyst3/v2_*.jsonl and v2b_*.jsonl (pair, tok, blk, sync list).