# Learn report 25 Oct 2026 (research)

Ledger complete to #778 (#777/#778 are analyst2's 24 Oct miss ideas, stage idea). runner-capture.txt now has 23 Oct (1-day lag, nightly 02:30 UTC: fine, no fix needed).

## What we learned
- **24 Oct misses** (fw/miss-2026-10-24.md): 10 runners (240M, 48M, 23.7M, 15.6M, 5.9M, 3.8M, 3.7M, 1.7M, 1.1M x2), none bought, 0 technical failures. 7 are v3 thin seeds (3 in the mc5 band but skipped), 2 are v3 jump (born at 23M-236M cap, no entry price exists), 1 is v4:native organic (not seen: venue gap, third day running).
- Thin-waker buy-all is already dead (#724: mean 1.33, 14% pulls, ~800 losers per runner). Only a selective rule on the 3 in-band skips could pay; analyst2's #778 (65 of 96 miss rows "never scored" jump/thin) says the scorer is blind there.
- Jump pools (cap_first priced in) are not tradable by cap-band rules: stop counting them as misses we can fix.

## Families
- Gaining: v3-new-pools (sparse60), verified lanes v3vearly, v3sg, v3sparse60, hk50.
- Losing: v3 thin late wakers, native v4 momentum.
- Unproven: cap-continuation floor (v3cf*, 12 paper tests, forward small; v3capv/v3capx decide today 25 Oct).

## Top 3 directions today
1. Decide v3capv/#720 and v3capx/#721 (decide-by 25 Oct) on stats; kill at n>=20 mean<1 with upper 80% < 1.05. No new cap-continuation variants.
2. Thin-waker selectivity: find a feature separating the 3 in-band skipped thin runners from the 800 losers (seed, funder, first-minute buyers) before any test; if none, park for good.
3. v4:native organic feed into radar in dry mode (ppfeeds): needed for any replay with real entries.