# FW Analyst 2, round 5: 10 iterations on freshly launched coins (8 Oct 2026, 18:30 Bangkok)

Data: harness gold tape (v3 WETH/USDG pools, every swap, mint and burn), 6,853 new pools, 12 Sep to 8 Oct 13:00 Bangkok. This is about 96% of all v3 pools created in that time, so it is not a winners-only sample. Real fills: harness bt.py tick walk, 1% each way + $0.05 gas, A1's honest exits (stop on the live value, sell at the end of the next block, own buy kept, exact time stop). Periods: A = 13-30 Sep, B = 30 Sep-6 Oct, C = 6-8 Oct.

Why v3: on v4 every outside-money test failed (the launchpad fleet rugs outside buyers). v3 is where the real-money mc5 makes money. The fresh-launch side of v3 had not been tested before.

## Headline
1. Blind "buy every new v3 pool at +30 s or +60 s, stop 0.5x, sell at 5x, 4 h time stop" makes money on real fills in every period: mean 1.22-1.26x (A, B), 1.68-1.76x (C), about 130-150 trades a day. Almost all of it comes from ONE launcher: v3 pools with a 0.01% fee (fee 100), made by the launcher contract 0x73991a25c818bf1f1128deaab1492d45638de0d3 (4,485 of 4,930 fee-100 pools).
2. Inside that launcher, one launch preset carries the edge: the creator's single liquidity position ends at tick 204,200 (204200 / -204200, depending on token order). That is about a $4.5k launch cap with thin liquidity. 499 pools (about 19 a day), 100% eventually pulled, but 52% reach 5x first.
   - At +60 s (real fills): mean 2.72x (A, 240 trades), 3.44x (B, 64), 2.89x (C, 13). $1,215 / $1,299 / $488 a day at $50. 24 of 26 days positive.
   - At +30 s: 2.73x / 3.04x / 3.65x.
   - Median trade 4.4-4.7x: this is a race between the 5x take-profit and the creator's pull, and the 5x usually comes first. The median gap from the 5x crossing to the pull is about 1 h. Only 1% of pulls happen within 12 s of the crossing.
3. THE CATCH (why this is not ready): in the 204,200-preset pools, 98.8% of buy volume comes from fleet wallets (buy recipients that appear in 20+ new pools; the top ones buy ~3,000 times across 77-207 pools). Outside wallets are 0.9% of buy volume. The pump to 5x is the operator's own volume bot. This is the same shape as the killed v4 flip.
   - Different from v4: pulls do not react to outside buys here. P(pull within 10 s) after a $10-500 outside buy is 0.74% (n=405), against 67-88% on the v4 launchpad. Fleet buys: 0.13%.
   - Still unknown: can an outside wallet actually SELL these tokens, and at what price? On v3 the sell recipient is the router (0xcaf681a66d020601342297493863e78c959e5cb2), so sellers are invisible until PP Data's tx.from table (R1, v3from.db) lands. ETA: newest pools tonight, complete 9 Oct about 10:00.

## The 10 iterations
| # | angle (first principles / winners vs rest / unused data) | result | verdict |
|---|---|---|---|
| 1 | Blind fresh v3 entry at +10/30/60/120/300/900 s, all exits (spot model, then tick walk) | +30/+60 s with stop 0.5 / 5x / 4 h: 1.18-1.23x (spot), 1.22-1.26x (tick walk) in A and B; no target loses (pump-and-dump) | lead -> split by venue tier |
| 2 | Fee tier (unused field) | fee 100: 1.13-1.22x (tick walk, without the preset); 1%: 0.88-0.97x; 0.3%: 0.54-0.60x; 0.05%: 0.63-0.68x | 1% / 0.3% / 0.05% KILLED |
| 3 | Early buy flow winners vs rest (buys $, counts, biggest buy, concentration) | "a few real buys of $300-1,500 by +60 s" looked like 2.2x, but it is a proxy for the launch preset (winners' launch cap $4.5k vs $9.6k, liquidity/cap 0.23 vs 0.92) | merged into #4 |
| 4 | Launch preset (first Mint tick range, never used before) | tick 204,200 preset: 2.7-3.4x in every period, real fills | A2-V3P204, stage 2 passed |
| 5 | Creator history (prior pools of the creator) | 4,318 creators for 4,930 pools: almost all are one-shot, no history to use | KILLED (no signal) |
| 6 | Token birth gap, hour of day | no stable difference between halves | KILLED |
| 7 | Outside money proxy 1: do pulls react to buys? (v4 failure mode) | after age 60 s, no reaction: P(pull within 10 s) after an outside buy 0.03% (other fee-100 pools) and 0.74% (preset) | pass (proxy) |
| 8 | Race: 5x crossing vs pull | median gap about 1 h; fills 0-12 s later still 5.2-5.3x | pass |
| 9 | Who buys (buy recipients, wallet in 20+ pools = fleet) | preset pools 98.8% fleet; other fee-100 pools 55% outside, 24% mid, 21% fleet | flag on A2-V3P204 |
| 10 | Entry time and exits for the preset | +10 to +120 s all work; +300 s fades (1.2x); +900 s loses. 5x target beats 2x/3x; no target loses | rule frozen at +60 s, 5x |

## Candidate rules (frozen, not tuned further)
- A2-V3P204: v3 WETH pool, fee 100, first Mint tick range ends at +/-204,200 (launch preset, known at the first swap). Buy $50 at first swap +66 s if not pulled. Stop 0.5x (live value, swaps of 0.01+ ETH), take profit when the cap reaches 5x the entry cap, 4 h time stop, pull = sell what is left. About 12-19 trades a day.
- A2-V3L100: same launcher, fee 100, all other presets, buy at first swap +36 s, same exits. Tick walk: 1.22x / 1.13x / 1.67x (A/B/C), about 80 trades a day, about +$6-11 a trade at $50. Outside wallets are 55% of buy volume here, so it is the "real market" part.

Not the same as WPOND (killed: v3 WETH fee-100, buy +4 s, sell after 15 min, 74 trades, 65 pulled). Most pulls come in the first minute (median 53 s after the first swap), so +4 s walks into them; entering after 30-60 s on survivors and taking 5x is what works.

Extra exit checks (tick walk, preset, +60 s): 3x target 2.02x / 2.24x / 2.23x; 10x target 3.2x / 4.0x / 5.3x (on the wider $-band variant, not adopted, kept for stage 3 as a side check); no target 0.56x / 0.61x (loses: these coins are pumped and then pulled). +120 s entry: 2.41x / 3.42x / 2.94x. +60 s pond (other presets): 1.17x / 1.13x / 1.84x, $615 / $601 / $2,365 a day. 1% fee pools at +60 s: 0.89x / 0.96x / 0.89x.

## Next (stage 3, read only, no wallet)
- Outside money test with R1 signers (v3from.db) on both rules: real outside wallets (signer in fewer than 20 pools) that bought at +30-120 s; did they sell, at what x, and how often did they get nothing? Pass bar: outside wallets that bought in the entry window and sold get at least 1.3x on average (preset) or 1.05x (pond), and no sign that outside sells fail.
- Forward: the same rule on gold data after 8 Oct 13:00 (hourly append), logged daily.
- A1 cross-check with bt.py on the frozen rule files: /root/fwa2/v3r/p204_60.json (preset) and /root/fwa2/v3r/f30_t5.json split by pool (pond; trades tr_f30_t5.csv, script split.py); +60 s pond variant f100x_60.json is running.

Files: /root/fwa2/v3r (rules, verdicts o_*.json, trades tr_*.csv), /root/fwa2/edge/v3fresh.jsonl (spot table), v3fleet.out, f100_mint.json (launch presets).