# Learn report 24 Oct 2026 (research)

Ledger complete to #770 (cap-continuation floor tests #762-#770 in paper; no new analyst ideas on 23 Oct, none padded).
runner-capture.txt: latest row is still 21 Oct although the nightly ran 23 Oct 02:30 UTC (2-day lag, source misses/runtrack). Not fixed this run; watch 25 Oct.

## What we learned
- **23 Oct misses** (fw/miss-2026-10-23.md): 14 runners (PONS 262M, 6.3M, 5.2M, 4.7M, 4.1M, 3.8M, 3.6M, 2.7M x2, 1.9M, 1.7M...), none bought. 6 are v3 thin late wakers (killed #174/#75/#757). 5 are v4:native organic or PONS USDG that mc5 never sees: venue not watched.
- Second day running the unwatched-venue group (v4:native organic + USDG) holds a 262M and 3-4M runners. Replays so far (#642, #744, #753) died on rules, but none was run on a live feed: the feed gap itself is the open lesson, not the gate.
- Real money unchanged: 65 live trades, net -0.0127 ETH; 12 near-total losses cost 0.079 ETH; no shadow trades in 3 days.

## Families
- Gaining: v3-new-pools (sparse60).
- Losing: v3 thin late wakers, native v4 momentum.
- Unproven: cap-continuation floor (#762-770, forward n small), serial-symbol.

## Top 3 directions today
1. Venue feed: v4:native organic + PONS USDG into radar/mc5 watch in dry mode so replay has real entries (ppfeeds miss-watch).
2. Cap-continuation floor tests: decide-by dates; kill on stats at n>=20, do not add more variants.
3. Stop vs rug cost on runner exits (runner-exits.md): 0.5x stop cuts 30x+ runners; needs one clean replay with rugs as 0x.