# FW Analyst 2, round 6: 10 more iterations on fresh launches (8 Oct 2026, 19:30 Bangkok)

Same data and fill model as round 5 (gold tape, harness bt.py tick walk, 1% each way + gas, A1's honest exits). Periods: A = 13-30 Sep, B = 30 Sep-6 Oct, C = 6-8 Oct.

**Dumb baseline (Manager D-032):** same universe (every new v3 WETH pool), same exits (stop 0.5, 5x target, 4 h), no picking. Real fills: 1.22x / 1.26x / 1.76x at +60 s, and 1.24x / 1.24x / 1.68x at +30 s. "beat random" = share of 200 random draws of the same size from that baseline that the rule beats.

## Results
| # | angle | result vs dumb baseline | verdict |
|---|---|---|---|
| 1 | A2-V3P204 against the baseline | 2.72x / 3.44x / 2.89x vs 1.22x / 1.26x / 1.76x; beats random 100% / 100% / 95% | edge stands (stage 3) |
| 2 | A2-V3L100 (same launcher, other presets) against the baseline | 1.22x / 1.13x / 1.67x: no better than the baseline | KILLED (D-032) |
| 3 | Is the operator's volume bot running? (2+ buys by fleet wallets by +60 s; fleet = wallet seen in 20+ earlier pools, past data only) | preset + bot: 3.31x / 3.45x / 2.89x; preset without bot: 1.87x (A only). Since 30 Sep almost every preset pool has the bot (62 of 64) | confirms the mechanism; no new rule |
| 4 | Bot goes quiet = pull coming? (exit signal) | after a 30/60/120 s fleet silence, a pull follows within 10 min only 10-15% of the time; median 26 s from the last fleet buy to the pull | KILLED |
| 5 | Same preset (204,200) from OTHER launchers | 1.00x / 1.07x | the edge is this launcher's operation, not the price setting |
| 6 | Other launchers' fresh v3 pools; same launcher at 1% fee | 0.90x / 1.06x; 0.28x / 0.42x | KILLED |
| 7 | Other presets of the launcher (207,243) | 3.12x on 49 trades in A, gone since 30 Sep (1 trade) | KILLED (regime over) |
| 8 | Exits for A2-V3P204: 5x / 10x / 20x / none | 10x: 3.30x / 4.95x / 3.46x (median 0.36, riskier); 20x: 2.55x / 2.86x / 4.37x; none: 0.56x / 0.61x | keep 5x as the rule; "half at 5x, half at 10x" (about 3.0x / 4.2x / 3.2x) goes to stage 3 as a pre-registered second arm |
| 9 | Latency and size | entry +90 s instead of +66 s: 2.60x / 3.56x / 2.93x (robust). $ a day by size (A / B): $50 $1,215 / $1,299; $300 $4,764 / $5,667; $1,000 $7,035 / $10,108; $3,000 worse. Capacity tops out near $1,000 a trade | size limit $1,000 a trade |
| 10 | Review of missed runners (review/misses.csv, 7 Oct): where were they born? | 19 of 26 runners, incl. the top 5 (PONS, CASHCAT, AI, SUSD, ORBIO), were born in v4 pools quoted in USDG or SUSD (mostly fee 0, hooks 0xe5e702641ea86f4ae6cc3cdaed2b886f976be044 / 0x027619b75fa3b39f107068a03eee66e7c7a0e8cc). 5 were v3 fee-100 kit pools; 1 v2; 1 Pons v2. None of our research data has the USDG/SUSD v4 pools | new research lead A2-USDG4 (stage 1), data request R6 |

## Hypothesis (Manager template) for A2-V3P204
- Who is on the other side: the operator behind launcher 0x73991a25c818bf1f1128deaab1492d45638de0d3. They launch a coin at about $4.5k, run their own buy bot (98.8% of buy volume) to make the chart look alive, and pull the liquidity about 1 h after it has run (all 499 coins are pulled in the end). We sell into the bot's buying at 5x, so we take back part of the operator's own pump money.
- Why they keep losing to us: our $50-1,000 is small next to what they collect from late buyers, and the bot is mechanical: it does not react to outside buys (a pull follows an outside buy within 10 s only 0.7% of the time).
- What proves it wrong:
  - (a) Real outside wallets (R1 signers) that bought +30-120 s and sold at 3-5x, average below 1.3x.
  - (b) Outside sells fail (token blocks sells: R5b).
  - (c) Forward: under 2.0x mean over the next 40 trades, or pulls within 15 s of the 5x crossing in more than 10% of hits.

## New lead A2-USDG4 (stage 1)
The biggest runners of 7 Oct came from a venue none of our bots or datasets cover: v4 pools quoted in USDG / SUSD, many with zero fee and a hook. That is what "NOT SEEN, venue ?" in the review means. Before any rule: get the tape (R6a), check how SUSD-quoted caps are priced (R6c: PONS "$386M" may be a pricing artifact), then run the same winners-vs-baseline process.

## Files
/root/fwa2/v3r (rules, o_*.json verdicts, tr_*.csv trades), /root/fwa2/edge/v3f2.jsonl (fleet features, past-only), botstop.py, runners.py, nv2.out.