# Analyst 3, idle run 3 (10 Oct 2026): names, deployers, wallets, phantom runners
Scripts: work/analyst3/m1-m13.py, w3-w7.py, d1.py. Data: e1x-runners.csv (27k pools, 21 Sep-9 Oct), names.db (new, ppdata), deployer_raw.db (new), r6weth.db (v3 tape with recipients), creators_tx.db.
"Real" = swaps>=60 and vol>=30k (kills one-print jumps). Halves = split at median birth time.

## Headline findings
1. **Phantom runners (counting fix).** Of 2,329 pools with peak cap >= $1M, only 515 are real (22%). Per TOKEN: 1M+ 769 listed vs **440 real** (~23/day); 5M+ 366 listed vs **113 real** (~6/day); 20M+ 231 listed vs **32 real** (~1.7/day). Same-symbol pools with the same frozen cap (SPY 19.5M, META 7.3M, MSTR 5.5M, CASHCAT, many USDG nohook) are copies, not runners. The runner-check list (SPY $10M, AMC, STANDARD, BONER rows) is polluted by them. 
2. **Where the real big ones are:** 65 of 113 real 5M+ and 23 of 32 real 20M+ tokens are v3 WETH (nearly all fee 100). Median 10-24 min from birth to 1M. v4 nohook native: 2 real 20M+ (both took 17+ h). USDG nohook: 13 real 5M+, 70% of them one-print or very low vol/cap. The unwatched pool types hold almost no REAL big runners; the runners are in v3 fee100, which mc5 covers only partly (53%, known).
3. **Wallet reputation, v3 (new, tape has recipients):** pools whose first-5-min buyers include >=3 wallets that bought early in earlier real runners: 13% / 9% reach real 1M (halves) vs 0.3% for none. But it is the busy-pool effect: inside the mc5 band with a buyer-count control the gain is gone (nb>5: 1.40 vs 1.32; tercile 3: 1.52 vs 1.04, n 25). At +66 s entry: no lift (nw0 2.05, nw>=4 2.02, n=1995). DISCARD for entry.
4. **Buyer concentration in first 300 s, mc5 band (new, tape):** share of first-5-min buy volume from the single largest buyer wallet. Band 330 pools, tp5 net of 7% = 1.24. Among pools with >=7 distinct buyers (220, 1.39): top1 < 0.11 -> n=146 1.58 hit5 23% (halves 1.90/1.40, CI 1.36-1.82); top1 >= 0.11 -> n=74 0.99 hit5 7%. Monotone across thresholds 0.08-0.30 (all 1.40-1.59 vs rest 0.94-1.06). Keeps 35 of 37 real winners. MATURE: gain over the dust guard is +0.07 (thr .15) to +0.19 (thr .11), in-sample, one 25-day window. To analyst2 for forward check on A1mc.
5. **Name meta (new):** hot word (name word shared with a token that touched 1M in the prior 7 d): population lift 3x (v3 8% vs 2.5%, both halves) but zero inside mc5 band (1.22 vs 1.25) and at +66 s (2.00 vs 1.98). DISCARD.
6. **Same symbol seen before:** v3 1.4x (4.3 vs 2.9%), others noise. DISCARD. (Beware: same token in several pools inflates "prior"; count tokens.)
7. **Short ticker (<=2 chars) v3:** 11%/8% vs 3.5% both halves, but 17 of 31 are the "SI" Super Inu family; in band no lift (1.34 vs 1.23, n=26). DISCARD. Meta lifecycle: SI family born 140 pools 30 Sep-4 Oct, 14 real 1M, rate falls as copies flood (4 Oct: 54 born, 5 runners). Bursts last ~5 days.
8. **Deployer repeat winners, v4 USDG nohook (deployer_raw):** deployer with >=2 pools that already touched 1M before this birth: 110 pools, 18 real 1M (16%), 12 reach 5M, vs 1.0% base; halves 24%/14%. But 11 deployers, 7 of 18 from one deployer on 7 Oct, USDG nohook is net negative after cost (a2 #82). n_eff ~4. PARKED as a watch-list rule, not a test.
9. **v3 deployer (creators_tx covers 1,512 of 6,177 v3 pools, 4-9 Oct):** first-time deployer 4.1% real 1M (46 of 1,126) vs repeat deployer 0.5% (2 of 386). Confirms the existing "no repeat deployer" arm in A1mc. Coverage 28%: dataeng to finish.
10. Data request: cross-chain meme match (below).

## Requests
- ppfeeds: cross-chain trending names. Every 30 min snapshot GeckoTerminal trending_pools for networks solana, base, bsc (duration 1h and 24h, pages 1-2; use geckogate) into /opt/sw/trending.db board "gecko_x", list "<net>_<dur>", rank, token symbol and name in extra. Keep from now. Test after 7 days: share of real-1M RH tokens whose normalized name or symbol matches an other-chain trending name <= 12 h before birth, vs matching rate of non-runners. Pass: lift >= 2 with >= 15 matched pools, both halves. Kill: lift < 1.3.
- dataeng: finish creators_tx.db for the other 4,665 v3 pools in r6weth.db meta (created_tx column, 66-char, same key as creators_tx.c.tx). Test: repeat-deployer avoid-filter and deployer-funder clusters on v3 fee100 with 6k pools instead of 1.5k.
- research: runner-check and the weekly runner map should count real runners: token-level, swaps>=60 and vol>=30k before alerting. SPY/META/MSTR/AMC rows are frozen-cap copies.