# Learn report 22 Oct 2026 (research)

Ledger complete to #756 (21 Oct miss ideas #752 #753 #754 have rows). runner-capture.txt latest is 20 Oct; nightly 02:30 UTC not yet due at write time (02:00 UTC), check 23 Oct.

## What we learned
- **21 Oct misses** (fw/miss-2026-10-21.md): 25 runners (20M+ 2, 10M 2, 5M 3, 3M 9, 1M 9), 0 technical failures, 0 bought by any paper test. 17 are v4:native organic (cap_first ~10k), 2 v3 thin seeds (one ran 60M), 4 v3 organic, 2 hooked v4.
- New angle: serial symbols (Kimchi x6, FINLEY x3, CUMROCKET x2) relaunch and run again; 5 of 6 Kimchi pools hit 2.5-5.8M. Lookback-only rule, no look-ahead: #753 SERIALSYM-48H (idea, not yet replayed). Older #608/#642 cover similar ground.
- #724 thin seeds killed (thin<3k mean 1.33 ex-top3 1.30 n425, 14% pulls); #754 THIN-60M reopens it for the 60M case only, so judge it on ex-top3 or it is one coin.
- Real money (realpnl): 65 live trades, net -0.0127 ETH, 12 near-total losses cost 0.079 ETH. $10 size is +0.153 ETH, $50 size -0.166 ETH (all on 8 Oct). Runner exits: 9 mc5 trades stopped at 0.5x then ran 4-35x (runner-exits.md).
- Cap-continuation floor tests (#727 #728 #755 #756) are in paper, forward unknown; killbot decides at n>=30.

## Families
- Gaining: v3-new-pools (sparse60 verified).
- Losing: native v4 momentum/continuation; thin seeds as a whole.
- Unproven: miss/cap-continuation bands, serial-symbol (#753).

## Top 3 directions today
1. #753 serial symbol: replay entry at first low cap of the 2nd+ pool of a symbol that ran >=1M in the last 48h; outside-seller and pull check; ex-top3 and 7-day holdout. Kimchi alone must not carry it.
2. #754: test thin 60M only as part of the full thin population (ex-top3); otherwise park.
3. Stop-side of mc5: 0.5x stops cut 9 later runners; any exit idea must be judged on runner-exits.md grid (no-stop 4h upper bound 2.93x ignores rugs) with rug cost.