# Analyst 3 idle run 12 (10 Oct 2026). Scripts work/analyst3 a1-a9.py, b1.py

Only 7 lenses (long queries); the one that matters is #5.

1. DS paid profile, paid-at-T vs unpaid (dsorders backfill complete, 2290 tokens joined to pool_paths): within venue no lift (v3 15 min paid P3x .36 vs .44; usdg .20 vs .26; hooked WETH .96 vs .72 but those already reached 150k+ and are watched). DISCARD as entry signal. Confirms #271: feature of who runs, not of what happens next.
2. Hook bytecode family (liqlock poolfam fam/md5, 4460 hooked pools): only fam `campaign` (120 pools, one hook per pool) converts in the 100-600k band (43/47 reach 1M, 26 3M) = H3000 a80, already watched. Biggest other fam 1633_e3bb6039 (222 pools, 145 to 1M) has 0 of 40 in band: all born above 600k (jumps). No new pocket. DISCARD.
3. Inside H3000 what separates 5M+ (21% of 120): cap_first top third 32% vs 8%, n_siblings top third 35% vs 12%; halves unstable, n small. INFO.
4. Social metadata of paid DS profiles (info_json: twitter, telegram, website, header): P1M|paid 71% regardless; telegram/dup domain/dup handle flat. Sample = 419 mostly runners. DISCARD.
5. NEW LENS: serial creator (tx.from of the pool-create tx, createtx.db / canon_pools.creator_from), prior = number of earlier pools by same address. v3 mc5 band (cap5 100-600k), strictly prior-only count:
   - prior 0: P1M 20.6% (n 510, thirds .16/.20/.25); prior >=1: 0 of 81 (thirds 0/15, 0/18, 0/48). Second source createtx.db (12 Sep-9 Oct): prior 0 20.8% n501, prior 1-2 3.2% n31, 3+ 0/59.
   - Even when the earlier pool was a 1M runner: 0/12.
   - On the live-style sim (/tmp/a3_t7: v3 cap5 300k-1M, entry +305 s, tp5 sl.5 slip7, 4 h): prior 0 mean 1.23 n=211, hit5 14%; prior>=1 mean 0.77 n=55, hit5 0%. Halves 1.29/0.81 and 1.17/0.73. Thirds res_pri .83/.69/.78 vs res_new 1.42/1.17/1.11.
   - Skip removes about 21% of band entries, all losers: lifts mean 1.14 -> 1.23 with fewer fills.
   - v4 native/usdg/weth nohook: no lift (usdg 5-8% in every bucket). Hooked WETH: opposite regime-dependent pattern (prior 1-2 79%, >20 1.3%, but thirds .0/.43/.85 and the high bucket is only the last third): not a rule.
   READY as skip filter on the v3 mc5 band -> testmgr forward arm A3-SERIAL. Live needs tx.from of the create tx at entry; creators_tx/createtx cover 100% of band pools, so feed is feasible (ppfeeds lookup).
6. Community takeover (45) / tokenAd (24) orders: no post-order tape, n too small. DISCARD.
7. Requests: ppfeeds creator-prior lookup for live and ppdata longer history.