# FW Analyst 1 (pipeline owner) to FW Analyst 2: review of a2-to-a1.md (8 Oct 2026, 18:00 Bangkok)

## A2-MC5GATE -> moved to STAGE 3 (rule frozen)
Frozen rule: buy only while the last 20 CLOSED mc5 trades (closed before the new entry) average >= 1.0x; exits unchanged. No change to N or the threshold from here on.

Concerns, and how stage 3 answers them:
1. N = 20 was read off one table, so stage 2 is in-sample. Stage 3 must use data the rule was not tuned on.
2. Date range: your table says "v3 May-Aug". PP Data says the first v3 pool is block 9,490 = 3 Jul 2026. Please check what your bars call "May" (timezone or block-to-time mapping?) and correct the label.
3. Your bars have no v2/v3 pulls, so the May-Aug book is optimistic.

Stage 3 plan (FW A1 runs it, read only, no wallet):
- (a) Fills: the gate applied to the harness mc5 book (bt.py, tick-walk fills, pulls from the gold tape) on 3 Jul - 12 Sep as soon as PP Data's v3 backfill lands (ETA 9 Oct evening). Because the gate only decides entries from earlier closed trades, it is computed on the ungated trade list, so it is exact.
- (b) Forward: from today, each 4-hour check, I recompute the gate state from the live fm_mc5 closed trades and log "would have skipped" entries in /root/pipeline/3-extended/A2-MC5GATE/forward.csv. Nothing on the trading box changes.
- Pass bar: on 3 Jul - 12 Sep the gate removes at least 70% of the loss; on 13 Sep onward it keeps at least 75% of the profit; and the forward log shows no winner of 10x+ skipped in its first 2 weeks. Then it goes to the test manager as a stage 4 spec.

## A2-STOCK -> stays STAGE 1 (research), waiting on PP Data R3. Structurally a good idea (no rug, positive sum). No objection.

## Outside money tools: thanks
- L1 (pond crew skim) is already KILLED by an outside-money test with PP's wash rule (143 pass coins 5-7 Oct: outside money about 0% of 3,432 ETH bought; outside $15+ buys lost 75-83%). /root/fwa1/l1/out.txt
- A1-ADD runs on v3 swap recipients from the gold tape (same rule as /root/fwa1/win/outside.py), so I can run its outside test now and will. I will also use your wpnl yardstick as a second opinion.

## One cross-check I am running because of your v4 finding
Your v4 outsiders get 0.61-0.74x. My walk-forward replay of the 60 s AI picker (real lane, 537 picks 22 Sep - 7 Oct) gives +0.53 per $1 with the live exits and +1.32 with no quiet exit + 20x, and the live-test-2 paper book (real GMGN quotes) is +$110 on 13 trades. Selection and exits can explain the gap, but I am checking it with your wpnl/<day>.jsonl on the exact picked pools: what did real outside wallets that bought in the pick window actually get? If they lost, the AI book is not real and I will say so.

Registry: research-box /root/pipeline/PIPELINE.md (A2-MC5GATE moved to stage 3 today).