# Analyst 3 idle run 8 (10 Oct 2026). Scripts t1-t9.py in work/analyst3. 9 iterations (3 on unwatched pool types, long scans).

1. Aged nohook "awakening" (v4 nohook, age>=1h at 300k, cap_first<=100k; radar_hist): USDG 110 pools P1M 34%, but 13 of 110 hits have a gap >=60 s between 300k and 1M (rest same-second = jump). Native 30 (P1M 13%), WETH 14. DISCARD (jump trap).
2. Hook x quote map of real 300k touches (e1x, swaps>=60, vol>=30k): nohook WETH 290 P1M 13% P3M 0%; nohook native 155 16%/2%; nohook USDG 217 78%/43% but median gap 0 s (jump). Confirms ceiling ~1M on unwatched WETH/native nohook; EV about 0.15 x 3.3 < 1. INFO.
3. NEW LENS name form: all-caps symbol. v3 band (c5 300k-1M, n=291 minus Robinhood Token names): upper n=207 hit5 26% r 2.59 vs mixed n=81 10% / 1.84; halves 2.79/2.38 vs 1.67/2.01. Real-runner population v3: P1M upper .25/.20 vs mixed .16/.06 per half. v4 hooked WETH lift (24x) is the H3000 kit confound (mixed has no fee 3000). With path sim (tp5 sl.5 slip7 fees, 4 h, +5 min entry) upper & 5-min vol>=$49k n=151 mean 1.35 (halves 1.56/1.18) vs mixed 0.95 (1.15/0.80). Absolute levels lower than analyst2's replay (my sim is simple); direction is the claim. READY as forward filter arm -> testmgr.
4. NEW LENS IPO-pop / first-minute oversubscription on band: 5-min USD volume <$49k (97 of 290, born near the band with 3 swaps median) = 0.78 mean, hit5 0%, vs 1.26 for the rest; n5<10-80 swaps same effect. These are stale band pools that never trigger the stop and hold a slot for 4 h. Same finding as #548 (swaps<10); threshold refined: swaps>=40 or vol>=$40-49k. c0 (first-print cap) and pop60 are the same pools. MATURE (duplicate), folded into arm of 3.
5. Young climbers (age 60-1800 s at 300k, non-jump start) by venue, radar_hist: hooked WETH 133 P1M 93% P3M 48% (watched, H3000); hooked native 124 P1M 50% but enterable 27%, P3M 12%; USDG nohook 75 P1M 31% enterable 7%; native nohook 111 25%/23% P3M 2%; WETH nohook 332 14%/12%, P3M 2%; v3 220 25%, P3M 5%. Unwatched types cannot reach 3M; hooked native is the one pocket with some enterable 3M (n=15). INFO.
6. Regime: band tp5 sim by day vs ETH/BTC daily return (21 days): same-day spearman -0.31 (down days 1.36 vs up days 1.01), prev-day +0.22. Weak and inverted vs the anecdote. DISCARD as pause rule; n too small.
7. All-caps across venues (real runners e1x): v3 and hooked WETH only; nohook WETH/native P1M 2-6% regardless. Nothing in unwatched pools. INFO.
8. Blockscout (token info, verified flag) is behind Cloudflare from research-box; cannot fetch. No test.
9. DATA REQUEST ppdata: pre-launch token distribution (Transfer events before first swap, recipients, share outside pool, top-10 at +5 min) for band pools + controls: a3-req-prelaunch.md.

Verdict: one forward arm (A3-CAPS) to testmgr; unwatched nohook WETH/native/USDG confirmed ceiling/jump again.

## A3-CAPS arm rules (for testmgr)
Entry: the live mc5 band rule unchanged (v3, cap 250k-1M at +5 min, entry at +5 min) AND symbol is all upper-case (python str.isupper() on names.db symbol) AND name does not contain "Robinhood Token" AND USD volume of swaps in the first 300 s >= 49,000 (or >= 40 swaps). Exit: unchanged mc5 (stop 0.5x, tp5 arm, 4 h time stop). Slip 7%, fees both sides.
Evidence: n=151 mean 1.35 (halves 1.56/1.18) vs mixed-case 0.95 (1.15/0.80); thresholds in-sample, so judge forward only. Kill: forward mean < 1.0 over 30 trades. Paper only, no live change.