# J-002 early-second score (FW Analyst 1)

## READ THIS FIRST (7 Oct 17:50 Bangkok)
- **KILL (v1, 0 of 6 pass).** A score from everything known at +10 to +60 s (first-seconds buys/sells/WETH flow, cap, launch LP, fee tier, launch kit, creator and funder-root track record) does NOT pick the winners. On half B the top of the score makes $1 to $36 a day at $10 vs $106 to $213 for buying every coin in the band, and loses without the top 3.
- Weights learned on half A (fee tier, cap, root record, early WETH in) flip on half B: the same pattern as the crew test (FW-021). What looks predictive in one week is gone the next.
- Not covered: obs.eye attention (PP trench data, not on research-box) and buyer wallet ids (swap table has no sender). Those are the only PP items left untested.
- Bug fixed before verdict: v1 ranked ties by row order, which leaked time into the weights (fake Spearman 0.5 to 0.6). v2 uses tie-safe ranks; verdict is the same.

Built 07 Oct 17:33 Bangkok. Score = rank-weighted features known at t, weights and top-q cut fit on half A only, frozen, tested once on half B. PASS = beats buy-all-in-band on both halves at $10 AND beats random p95 on half B AND half B no top 3 beats +5 min control no top 3 ($+71). Control +5 min half B: $10 $+164, $1k $+14160.

| entry | exits | rule | n B | B $10 | B no top 3 | A $10 (buy all A) | B $1k | random p95 B $10 | verdict |
|---|---|---|---|---|---|---|---|---|---|
| +10 s | E0 | top 33% of score in $50k to $750k | 89 | +1 | -27 | +140 (buy all +155) | -436 | +92 | fail |
| +10 s | E2 stop 0.3x | top 50% of score in $50k to $750k | 108 | -18 | -47 | +129 (buy all +149) | -2231 | +104 | fail |
| +20 s | E0 | top 50% of score in $75k to $750k | 88 | +36 | -13 | +96 (buy all +129) | +2554 | +82 | fail |
| +20 s | E2 stop 0.3x | top 50% of score in $75k to $750k | 83 | +35 | -14 | +117 (buy all +148) | +2409 | +82 | fail |
| +30 s | E0 | top 50% of score in $75k to $750k | 90 | +32 | -14 | +94 (buy all +123) | +2179 | +108 | fail |
| +60 s | E0 | top 33% of score in $100k to $750k | 49 | +13 | -14 | +69 (buy all +93) | +912 | +61 | fail |

- +10 s E0: weights fit on half A (Spearman with log x, |w| >= 0.05): fee +0.46, logcap +0.35, root_hits +0.29, root_n +0.29, weth_in +0.25, root_pulls +0.21; half B buy-all in band $+162 (no top 3 $+23); score B worst 3 days $-164; random p95 $1k +4815
- +10 s E2 stop 0.3x: weights fit on half A (Spearman with log x, |w| >= 0.05): fee +0.48, logcap +0.30, root_hits +0.26, root_n +0.25, weth_in +0.20, root_pulls +0.18; half B buy-all in band $+166 (no top 3 $+23); score B worst 3 days $-324; random p95 $1k +5319
- +20 s E0: weights fit on half A (Spearman with log x, |w| >= 0.05): fee +0.49, logcap +0.35, root_n +0.34, root_hits +0.29, weth_in +0.27, root_pulls +0.27; half B buy-all in band $+181 (no top 3 $+42); score B worst 3 days $-166; random p95 $1k +3633
- +20 s E2 stop 0.3x: weights fit on half A (Spearman with log x, |w| >= 0.05): fee +0.51, root_n +0.31, logcap +0.30, root_hits +0.27, root_pulls +0.24, weth_in +0.22; half B buy-all in band $+185 (no top 3 $+46); score B worst 3 days $-187; random p95 $1k +3872
- +30 s E0: weights fit on half A (Spearman with log x, |w| >= 0.05): fee +0.51, logcap +0.35, root_n +0.33, root_hits +0.29, weth_in +0.27, root_pulls +0.25; half B buy-all in band $+213 (no top 3 $+53); score B worst 3 days $-175; random p95 $1k +4695
- +60 s E0: weights fit on half A (Spearman with log x, |w| >= 0.05): fee +0.43, logcap +0.38, root_n +0.30, weth_in +0.28, px_chg +0.26, root_hits +0.25; half B buy-all in band $+106 (no top 3 $+28); score B worst 3 days $-65; random p95 $1k +3894

Missing from PP list here: obs.eye attention (PP trench db) and buyer wallet ids (no tx sender in the swap table). Ledger FW-023.

Version: v2, 7 Oct 2026 17:50 Bangkok, FW Analyst 1. Latest: http://159.223.88.168/hx-c9ois/fw/lab/j002-score.md