# Learn report 23 Oct 2026 (research)

Ledger complete to #758 (22 Oct miss ideas #757 killed, #758 idea). runner-capture.txt latest is 21 Oct; nightly 02:30 UTC not due yet at write time (02:00 UTC). Check 24 Oct.

## What we learned
- **22 Oct misses** (fw/miss-2026-10-22.md): 9 runners (33M, 7.3M, 6.4M, 4.4M, 4.1M, 3.6M, then 1-2M x3), none bought by any paper test. 6 of 9 are v3 pools still under $250k at +5 min (late wake-ups); already killed as #174/#75/#757. 3 non-v3 (Macca v4:native, GF pons_v2, 0x30de7a usdg fee130000) match known unwatched clusters. 1 technical: 0x30de7a never checked by mc5 (#758, feed gap).
- Real money unchanged: 65 live trades, net -0.0127 ETH; 12 near-total losses cost 0.079 ETH. Nothing new since 17 Oct.
- Runner exits (runner-exits.md, n=101): 38 trades peaked >=3x; the 0.5x stop cut many of them (e.g. 35x, 35x, 31x, 29x runners exited at ~0.45x). Open question stays the stop vs rug cost.

## Families
- Gaining: v3-new-pools (sparse60 verified).
- Losing: v3 late wake-ups, native v4 momentum, thin seeds as a whole.
- Unproven: cap-continuation floor tests, serial-symbol (#753), v4:native into mc5 with symbol-repeat gate (#642).

## Top 3 directions today
1. #758 feed gap: confirm why mc5 never checked 0x30de7a (usdg fee130000 pool type); if a pool type is unwatched, ppfeeds adds it.
2. #753 serial symbol replay: first low cap of 2nd+ pool of a symbol that ran >=1M in last 48h; ex-top3, 7-day holdout.
3. v4:native organic (Macca-type) with lookback-only gate #642: needs a replay with n>=30 before paper.