# Analyst 3 idle run 10 (10 Oct 2026, ~16:50 BKK). Scripts x1-x7.py, famflag.py in /opt/crew/work/analyst3
Data: radar_hist.jsonl (18k crossings 12 Sep - 9 Oct), radar.jsonl (7-10 Oct), names.db (dataeng, 109k tokens), mc5 band sim pickle (290 band entries 12 Sep - 10 Oct).

1. **Hazard / time stop (new lens: survival analysis)**, non-jump 300k touches. P(1M) if 1M not reached after T min: v3 .26/.18/.05 (10/30/120 min), v4 nohook weth .11/.07/.01, native .15/.11/.01, usdg .11/.09/.07, weth-hook .51/.44/.04. -> a 300k entry that has not hit 1M after 2 h has <=8% left: time stop 2 h loses almost nothing. INFO (to analyst1/2 as exit rule). 
2. **Copycat / known-name lens (new)**: non-jump 300k tokens (n 1656 named). Prior token with the same symbol already crossed 1M before my 300k: P(1M) .52 (n 267, 109 symbols, per-symbol mean .51) vs prior non-runner .27 vs first of symbol .40. By venue: v3 .62, usdg .73 (but USDG base .60 -> small), native .16 / weth .28 (no lift). Halves .50/.54.
3. **A3-COPY on the mc5 band (v3, +5 min)**: symbol had a 1M token before entry: tp5 sl.5 slip7 mean 1.39 n=106 (halves 1.47/1.22) vs named-no-prior .96 (124) vs unnamed .90 (60). Ex top symbol SI: 1.43 n=96; per-symbol vote 1.58 over 61 symbols. Stacks with seed<7.8k (+vol5>=49k): n=55 1.79 (1.75/1.93); prior&~seed .96; no-prior&seed 1.18. Not a hot-meta effect: freshness (hours since last/first clone runner) and size of prior runner flat. Dose 5+ priors 1.00 (n=15). READY as forward flag arm (in-sample); helper famflag.py.
4. **Name-form battery on all venues** (non-ASCII, length, digits, name==symbol, 17 words), venue-adjusted: flat. Only HOOD/ROBIN names P1M .65/.59 vs .42 (n 40/27), INU .52 vs .40 (n 87). INFO, too small.
5. **Contract capability lens** (tokcaps.db on band): 91 of 290 matched, all identical (owner not renounced, same selector count/hash) -> no variance. DISCARD.
6. **Unwatched-pool census today (10 Oct 00:00-09:30 UTC)**, radar.jsonl: 1M crossings v4 nohook: usdg 41 (3 non-jump, one with depth 2e9 = phantom), native 27 (1 non-jump, born 589k), weth 6 (1 non-jump). Mouse/PERSUASION $20M (weth), SPY/STANDARD/BONER (usdg), CASHCAT (native) = single-print jumps. Re-confirms no hidden catchable pocket in unwatched nohook; catchable ones are v3 (7) and weth-hook (2).
7. **Meta timing (early vs late in a name meta)**: hours since last/first clone runner <6h/6-48h/>48h 1.35/1.16/1.58; first<24h 1.25 vs 1.48; prior 5M+ 1.44 vs 1M 1.30: flat. The edge is "name already proven", not "meta is fresh". 
8. **Launchpad table runtrack.db** (1238 tokens): selection sample (tracked after crossing, 70% >=1M), not population. longxyz 113, pons_v2 490, noxa 57, virtuals_v2 23 launchpads exist; no population rates possible. Request below.
9. **Request ppdata**: names.db born_t + pool/venue per token (see a3-req-nameborn.md).
10. Post testmgr forward arm A3-COPY.

## Rules for testmgr (A3-COPY forward arm, same machinery as A3-SEED)
Universe/entry/exit: identical to A3-SEED (v3 mc5 band, cap 300k-1M at +300 s, enter +305 s, tp5 sl.5 4 h, 7% slippage). Flag: upper(strip(symbol)) of the new token (names.db) equals the symbol of another token that crossed $1M before entry time (python3 -I famflag.py out.json builds symbol -> {token: first 1M cross time} from radar_hist + radar.jsonl; refresh each run; skip self).
Arms: C = flag; C+S = flag AND seed<7.8k AND vol5>=49k; B = baseline band. Skip names missing in names.db (count them).
Kill: arm C forward mean < 1.1 over 30 closed. In-sample expectation C 1.39 (n=106), C+S 1.79 (n=55); thresholds fitted on the same data, expect lower forward. Depth to exit: same band curve (~90k at 300k), not audited at entry.