# Learn report 21 Oct 2026 (research)

Ledger complete to #724 (every analyst miss idea has a row). runner-capture.txt still shows 19 Oct at write time (nightly 02:30 UTC not yet due); check tomorrow.

## What we learned
- **20 Oct misses**: 21 coins; v4:native organic is again the biggest pool (80 rows: 63 NOT SEEN, 17 MISSED, e.g. KOLSHI 1.1M, Bonnie 1.0M: aipaper5 never scored them, mc5 does not watch the venue). v3 kit:other: 29 rows. 3 pons_v2 / hook native thin rows.
- 4th day in a row (#681, #682, #698, #723): v4:native organic is the main runner pool and every native rule so far was killed. #722 shows all 208 runners of 13-19 Oct sit in already-killed clusters: the problem is entry timing/exit, not unseen pools.
- #724 (v3 thin seeds at ~800 cap_first produced 9-30M runners): untested, hunt 5-20M tier; thin seeds also rug most, so needs outside-seller check.
- Audits: v3sparse60 verified 20 Oct (first clean verify). growth120, deadkit32, hlate not verified. Cap-continuation: higher bands killed, #704-#706, #720, #721 in paper (forward unknown).
- Analyst idle runs are saturated ("every lens numbered"); #717 (dev-buy column) is blocked on ppdata live_paths.db (0 bytes since 20 Oct).

## Families
- Gaining: v3-new-pools (sparse60 verified).
- Losing: native v4 momentum/continuation, high-cap continuation.
- Unproven: miss/cap-continuation lower bands (paper).

## Top 3 directions today
1. #724 v3 thin seeds: replay entry at birth+30-90 s with outside-seller check at the 5x tp level; judge ex-top3 on holdout 7 days.
2. v4:native: stop new entry rules; test only the exit side on the aipaper-scored subset (wide trail) since entries all died; else park in writing.
3. Unblock #717: ppdata live_paths.db rebuild (deployer swap column); analysts to use v4tape sender fields meanwhile.