# Learn report 20 Oct 2026 (research)

Ledger complete to #698 (every board idea has a row). Capture/runner nightly runs 02:30 UTC (not yet due at write time; capture file still shows 18 Oct, expected).

## What we learned
- **19 Oct misses (mc5 rows)**: 49 mc5 rows (36 MISSED, 13 NOT SEEN); most have bogus ATH ($2.4T, born-high artifacts, ignore). Per analyst1 #698: 13 of 20 real runners were v4:native organic (UHI 20M). Third day in a row (#681, #682, #698): v4:native small-born is the biggest runner pool and every native rule so far (#674 MOM5, #694 V4CONT) was killed.
- **Audits 20 Oct**: v3sparse60 VERIFIED (fwd n81 mean 1.25, ex-top3 1.14, outside sellers at tp 22/23): first clean verify in days, binary profile (70% pulls at 0x, tp ~4.2x), drift trigger last30 < 1.0. growth120 NOT VERIFIED (fwd 1.47 ex3 1.40 but replay 0.85 and only 3/28 tp exits have outside sellers). deadkit32 NOT VERIFIED (on-time n38 1.445, 26 of 38 at/below 1.0, mean cap-bound by quiet rows). hlate NOT VERIFIED (fwd on-time n5; retire for low flow).
- Pattern holds: tests that pass have ex-top3 >= 1.1 on on-time rows AND outside sellers at the tp level. Tests that fail lack one of these.
- Cap-continuation family: higher bands (#666-#668, #672, #673) killed; lower bands (#685, #686, #687, #688, #695, #696) in paper, forward unknown.

## Families
- Gaining: v3-new-pools (v3sparse60 verified), growth120 forward.
- Losing: native v4 momentum/continuation, high-cap continuation (>=1.5M), hlate (low flow).

## Top 3 directions today
1. v4:native: only birth+30-90 s entries (seed size, first-buyer breadth from v4tape), judged on on-time ex-top3 >= 1.0 with outside sellers; else park the family in writing.
2. growth120: explain the forward 50% vs replay 25% tp gap and measure outside sellers at the tp level on more rows; without that it will not verify.
3. v3sparse60 shadow: define drift check (last30 < 1.0) and look for sibling kits (same sparse/binary profile) on other pool types.