# PP vector replay vs mc5-vectors.csv (FW Analyst 1, sup-to-a1 v35 task 2)

## READ THIS FIRST
- 7 Oct 2026 16:00 Bangkok. **PASS 30 of 30, 0 mismatches.** I reran PP's replay myself on research-box against my own file, with the installed bot build: fm_mc5 md5 e3ad18065c262e097b69813145411a58, mc5-vectors.csv md5 b9b25cc764c59c96655cd3ed40395239 (same file as published). Every stop row has the same block, and the stop value is within 0.002.
- **Not covered yet. Here is what the replay does not prove:**
  - **Buy or skip on all 831 pools:** 38 BUY and 793 SKIP (band, dust guard, fee tier). The replay starts after our buy, so the bot's entry decision is untested against the file.
  - **6 time stop exits:** 0x7608c38ea8cc2082b043e886185979942cdeb58e, 0x5560b76440e9b6660ab6ca8c7ae1941c92abffcf, 0x5078092c196727b4798c5c4d583d69146bec3929, 0xb4b0dfeaa2e2a838f0f7a304f005ebe32390d4ef, 0x3939ce55ec4e0ff8df9aa6469a795c793da06d55, 0x8da021f34c2f2a95696c4fd842b0cd0cc4e18468.
  - **2 pull exits:** 0x351722e71c93290847c1e3b5a585369b11c42236, 0xad4cfc3433988aa9495d94b4163f84e2422bb1d2.
- **Ask to PP:** extend rp46 to (1) the entry decision on all 831 rows (cap at +300 s, band, dust guard, fee tier) and (2) the time and pull exit block on the 8 rows above. Then the vectors prove the whole mc5 module, not only the stop.

## Proof
- Command (research-box): `python3 /opt/sw/rp46.py /opt/sw/fm_mc5_v46.py /root/fwa1/vec/vec.db /root/fwa1/vec/out/mc5ref.json /root/fwa1/vec/mc5-vectors.csv`
- Full output, one line per pool: http://159.223.88.168/hx-c9ois/fw/lab/rp46-out.txt (research-box /root/fwa1/vec/rp46-out.txt).
- Pool 0x86303aff9563213a7c5f877bbe15eacde98d7c50: bot block 82262763, value 0.4952, the same as lab.py.

Ledger: FW-018. Version: v1, 7 Oct 2026 16:00 Bangkok, FW Analyst 1. Latest: http://159.223.88.168/hx-c9ois/fw/lab/vector-replay-diff.md