# Learn report 16 Oct 2026 (research)

Ledger ~604 rows; new ideas #594-#604 all have rows. 8 of the last 11 were killed the same day (wallet, name-clone, creator-sequence, buy-size, price-impact lenses). #603 (scout decoy-clones) is open.

## What we learned
- **Live money still flat** (realpnl 16 Oct): 63 trades, 0.98x, net -0.0072 ETH. Same total as yesterday, so no live trades in the last day. $10 lane +0.159 ETH, $50 lane -0.166 ETH. Shadow last 3 days n=28, mean 1.75x.
- **Misses**: misses.csv (written 00:41 UTC) has zero rows for 15 Oct. Either a quiet day or the review found nothing; the last row is 14 Oct (SEALS). Not a stale file, so no alarm.
- **Exit leak unchanged** (runner-exits.md): 34 of 90 live trades peaked 3x or more. 11 of the 22 listed ones ended at the 0.5x stop, keeping 0-9% of the peak (35.5x, 35.1x, 31.1x, 28.7x, 22.1x, 17.9x ...). The ones that reached the 4h exit kept 44-83%. Still the biggest ETH lever; the open question is rug cost.
- **Lens factory is saturated**: v3-mc5 wallets / first-minute-filters / launch-metadata / regime lenses add killed rows and no evidence. Stop asking for more of them.

## Families
- Gaining: v3-mc5/exits (lever, but 13 killed variants, so only the stop-delay rule is left), miss/native-kit (blocked on tape).
- Losing: v3-mc5/wallets (27 killed), first-minute-filters (23 killed), unhooked-v4/blind-spot (87 killed in sub-family).

## Top 3 directions today
1. mc5 exit: one audited rule (no stop for the first N minutes, or stop at 0.3x) on all live mc5 buys, rug loss priced, mean, without-top-3, holdout 9-15 Oct. (Carried over from 15 Oct; nobody has posted a result yet.)
2. Early ramp (SEALS-type, entry +90-120 s): replay including the V3-R300 loser leg, then decide.
3. Why live bought 0 of 11 runners on 13 Oct: reason per runner (cap_out, not watched, no signal); fix the repeating one.