# A1 answers 10 Oct: #357 daily miss review, #361 S rule OOS, #366 5-Min Cap on v4 no-hook

## #357 Misses of 9 Oct (20 runners >= $1M, from review/misses.csv)
Did any paper test catch one? No: all 20 are MISSED / NOT SEEN for every bot listed (mc5, aipaper, aipaper30, aipaper5, fml2).

| Why missed | Runners | Detail |
|---|---|---|
| Venue not watched (mc5 = v3 WETH only) | 15 of 17 mc5 rows | v4 no-hook WETH 8 (WOMBAG MBLOOM STLZA CRNKP RROVR FRACFOX Clanker + BMBFI), v4 no-hook USDG 4 (AAPL $309M, BONER, STANDARD, V4), v4 no-hook ETH 2 (OpenAI, LOL), v4 hook-USDG 1 (REALM), Pons v2 1 (NASDANQ) |
| Filter (cap outside band) | 1 | BMBFI cap $2.6M at 5 min (over $1M); v4 no-hook WETH |
| In band, execution/skip | 1 | LOL (letscash) "in band but skip": check why the skip fired |
| Never scored by the AI pickers | aipaper5 20/20 | aipaper5 never scored anything on 9 Oct: technical (feed or scorer down?), ask testmgr |
| Scored below the line (aipaper/30) | 13 / 10 | scores 0.05-0.83; OpenAI 0.83 and PAPER 0.83 (aipaper) and QI 0.77 (aipaper30) were close; the 5 USDG runners score 0.06-0.31 |
| fml2 skip_quote / skip_sellquote | 6 | all v4 no-hook WETH (WOMBAG BMBFI MBLOOM CRNKP RROVR FRACFOX): quote check fails on v4 no-hook, so fml2 cannot trade this venue at all |
| Never seen by anyone | DAEMON (v3 no hook WETH) | all three AI bots "never scored" |

Proposals:
1. Biggest block = venue. 14 of 20 runners are v4 no-hook (WETH/USDG/ETH). Fix fml2 skip_quote on v4 no-hook (ppfeeds/testmgr: is the quote function v3-only?).
2. aipaper5 scoring nothing on 9 Oct needs checking (testmgr).
3. Venue extension of mc5 itself does NOT pay (see #366 below): the v3 band rule is not what catches these.
4. LOL "in band but skip": log the skip reason.

## #361 H100-S out of sample (ppdata tape to 83.962M, hooks...144 fee 100 WETH)
Frozen rule: slot1 at creation in {1e12, 1e13}, entry +60 s, reserve >= 0.5 ETH, pos 0.02 ETH, 20 min, tp5.
OOS = hooked fee-100 WETH pools created after 7 Oct 19:37 (31 pools with tape, all slot1 joined).
- S pools: n = 2 only (tp5 3.49). Rest (slot1 1.1e14): n = 23, tp5 1.80, 39% zero (was 2.14 / 21%).
- The S population collapsed: 17-27/day on 24-29 Sep, 2-9/day since 2 Oct, 2/day on 7-9 Oct. Default pools also fell (1.80).
- Verdict: NOT confirmed, NOT rejected (n=2). No new data can help; forward test (testmgr #342) stays the only way. Given 2 S pools/day, a 40-trade verdict needs ~3 weeks: suggest testmgr keeps it as a cheap arm, no more analyst time. Scripts oos1.py oos2.py.

## #366 5-Min Cap on v4 no-hook (replay, WETH 18 days + native-ETH 18 days; USDG not done, tape decimals differ)
Rule as live mc5: first swap +300 s check, cap (price x supply x ETHUSD) $250k-1M, pool net quote >= 0.1, price moved; enter +306 s, stop 0.5x, pull exit (active liquidity -50%), 4 h, fees from pool (1%+), impact model, $50 position. Calibration on v3 WETH with the same code: band n=199 mean 2.74x, median 0.50 (tail edge as live).

| v4 no-hook | band $250k-1M | n | mean x | median | halves |
|---|---|---|---|---|---|
| WETH | 250k-1M | 40 | 0.85 | 0.50 | 0.71/1.00 |
| native ETH | 250k-1M | 16 | 0.55 | 0.35 | 0.66/0.44 |
| both | 250k-1M | 56 | 0.77 | 0.49 | 0.78/0.85/0.67 (thirds) |
Band hits only ~3 trades/day and has no tail (best 3.5x, 0 trades >= 5x). **5-Min Cap does not hold on v4 no-hook after fees. Not sent to testmgr as 5-Min Cap.**

Side finding (not the 5-Min Cap): same entry (+306 s, stop 0.5, 4 h) on v4 no-hook coins with cap UNDER $250k:
n=1260, mean 1.70x, median 1.20, 68% >= 1x, thirds 1.67/1.65/1.78, ex top 5 1.65; <$50k: n=220 mean 2.15. Direction check OK (price rises with buys). Caveat: this is a sim with price impact only, no real fill; every pool shows one recipient (router), so I cannot prove outside money: a few self-pumped coins would look the same. Also v3 shows the same pattern for <$50k (2.48), so it may be the model (stop 0.5 clipping losers, drift up on survivors) rather than a venue edge. Sent to testmgr as a paper arm with a sellability probe, not as ready.

Scripts: mc5v4.py (build), mc5v4b.py (replay), Poos.pkl, mc5v4_*.pkl in /opt/crew/work/analyst1.