# Learn report 12 Oct 2026 (research)

Ledger: 511 rows. Miss review file (review/misses.csv) still ends at born_day 10 Oct (74 rows, written 11 Oct 17:05 UTC); no 11 Oct births yet, so this report leans on runner-exits and the audits.

## What we learned
- **Exit is the leak, not the pick** (fw/runner-exits.md): of 51 live mc5 trades, 20 peaked >=3x, but 14 of those were stopped at 0.5x (entry caps $262k-$978k) and then ran 3.5-35x. Peak kept on runners is 29%. No-stop 4h replays at 2.93x (1.43x without top 3), but this ignores rugs, so it is an upper bound.
- **Capture** (fw/runner-capture.txt): live bought 5 of 7 $1M+ runners on 10 Oct, 2 of 10 on 9 Oct; peak kept 8-11%.
- **Audits today are all FIX, none verified:** seed (27 of 31 "forward" rows are replay; real forward n=4, mean 0.55x) and fast5-NA (n=35 mean 1.25x, 46% zeros, no audited replay, no sellability). Pattern again: replay edge, thin or wrong forward, no outside-seller proof.

## Families
- Gaining evidence: v3-mc5/exits (a wider stop or no stop on $400k+ entries; only family with a plausible +ETH/day lever).
- Losing/flat: seed first-print (forward 0.55x), fast5/native (at the kill line), unhooked-v4 blind-spot (still 0 alive).

## Top 3 directions today
1. mc5 exit: audited replay (holdout last 7 days, rugs/pull-aware, outside-wallet sell sim) of stop 0.5x vs wider/delayed stop (e.g. no stop for first N min, or stop 0.3x only after a minimum-liquidity check) on ALL mc5 buys, not just the runners. Report mean, without-top-3, and rug cost.
2. Quantify the rug side of "no stop": how many of the 31 non-runner mc5 trades went to ~0 and what a 4h hold would have cost vs the 14 runners saved.
3. Stop adding replay-only arms: any new arm must state its forward n by day 5 (killbot low flow) and include sellability from the start.