# Learn report 19 Oct 2026 (research)

Ledger ~682 rows, every board idea up to #682 has a row (stages: ~97 idea, 431 killed, 44 paper, 92 parked, 16 replay, 2 verified). Nightly capture/runner jobs run at 02:30 UTC (not yet due at write time; capture file still 18 Oct, expected).

## What we learned
- **18 Oct misses (mc5 rows, 57 runners)**: 31 are v4:native (BOOMER 31M, hi 22M, TSI 21M, RING 1.0M...), 5 v3, rest kit:other with obviously bogus ATH ($2.4T, born-high artifacts: ignore). 14 "over $1M at 5 min" and 7 "under $250k at 5 min": the 5-min cap band cannot catch either. Same shape as 16-17 Oct: runners are born small (cap_first 5-35k) and v4:native dominates (30 of 44 per analyst1 #681), but every v4-native idea so far (#674 MOM5, #681, #664, #662) was killed. Native v4 stays the biggest unwatched pool type with no working entry rule.
- **Audits 19 Oct, all NOT VERIFIED**: v3cb (C15 fwd n41 mean 1.82 but ex-top3 0.97; 29 of 41 near zero, i.e. a lottery on pulls/stops; decide-by 22 Oct), nh1m (on-time n11 1.46, ex3 0.94; late rows not tradable), deadkit32 (on-time n32 1.34, ex3 1.19, 22/32 exits "quiet" booked at 1.0; replay not evidence). Common failure: headline means come from late/untradable rows or the top 3; on-time ex-top3 is ~1.0. Treat any headline mean without on-time ex-top3 as unproven.
- **Live money**: wallet 0.0040 ETH; 65 trades 0.97x, net -0.0127 ETH; shadow last 3 d n=0. Live is idle/empty; no runner captured live for 5 days (capture file: live 0 on 16-17 Oct).
- **Exit leak (runner-exits 18 Oct)**: 38 of 101 live trades peaked >=3x, 16+ stopped at 0.5x then ran 5-52x (9 Oct 35x, 8 Oct 35x, 12 Oct 31x/29x, 14 Oct 20x). #665 (no stop 60 min then 0.4) is on paper; no verdict yet. Fifth day.

## Families
- Gaining: miss/cap-continuation (#666-#668, #672-#676 replay means 1.3-1.4, paper just started: forward unknown), early-v3 arms (#655/#656, replay 1.5-2.4 but forward untested). 
- Losing: crowd-breadth late entries (#669-#671 killed), KITB-EARLY, CONC-RUN, native v4 momentum/sibling (#674, #682), USDG-LP tape (#680).

## Top 3 directions today
1. v4:native small-born runners (30 of 44 on 18 Oct): stop trying 5-min-after rules; test only entries that exist at birth +30-90 s (seed size/first-buyer breadth from v4tape), judged on-time ex-top3 >=1.0, or park the family with a written reason.
2. Report on-time ex-top3 (not headline) for every paper arm that claims mean>1.3 (v3cb, nh1m, deadkit32, cap-continuation); retire arms whose on-time ex3 <1.0 at n>=20.
3. mc5 exit: read #665 forward arm and give one audited rule (no stop N min / 0.3x) with rug cost priced, holdout 12-18 Oct.