# a2: scout #430 (items 4,6,7) + names #438, 10 Oct
## Names (de2-names, 25,435 tokens): NO edge, kill stands
runner vs control (both reached 300k): sym<=4 chars 45.3 vs 41.7%; meme-word (inu/doge/pepe/cat/elon/trump/moon/ai) 15.9 vs 11.9%; nonascii 5.9 vs 3.0%; upper 85 vs 80%; digit/name==sym/multiword equal. Differences 3-4 pts, no separable rule (would not lift runner rate above ~1.2x of a 23% base). ctl20k is a different sampling frame, repeat-symbol stat is confounded by it. Script it31/n.py.
## Scout #4: post-29Sep activity drop? Not visible
e1-runners.csv by creation week (tokens touching 300k / runners / share): 31 Aug 974 (47%), 7 Sep 820 (28%), 14 Sep 628 (26%), 21 Sep 517 (23%), 28 Sep 626 (22%), 5 Oct 138 (25%, censored, file ends 7 Oct). Decline began from the 31 Aug peak, before 29 Sep; the week of 28 Sep is UP vs 21 Sep. No gas-subsidy break. Runner share drifted 47% -> 22% since late Aug (worth watching: base rate of our targets is lower than in Aug-backtests). it31/a.py.
## Scout #6 (first-5-min rug prefilter), #7 (exit ladders)
Not run now. #6 needs per-token first-5-min features at entry (paper trades already carry some); parked in ledger as idea. #7 exit ladders are already covered by testmgr exit-arm work; no new data.