Pre-launch· Token not live yet — no rewards or treasury until launchWhat this means
← Model
Adds launch structure (bundles, snipers, dev)

TRAIN v0.18

Adds launch structure — bundled buys in the creation slot, snipers in the first 10 seconds and how much they've sold, first-minute buying, organic-volume checks (volume per wallet, back-and-forth wallets, tiny bump trades), KOL buys, and the dev's launches across all pump.fun coins.
Completedpreviously deployed

Definition

Model type
Decision-tree ensemble
Signals used
Market + wallets + launch structure
Technical details
max_depth
2
subsample
0.8
n_estimators
200
learning_rate
0.05
min_samples_leaf
10
Random seed
1018
Config hash
7340b8b11b56d9f3
Created
05 Oct 17:07
Completed
05 Oct 17:09
Artifact
models/v0.18/model.json
sha256 1a2ded7a90800f3f…

Dataset & compute

Dataset
Solana Launch Dataset v10
7,170 snapshots · 1,439 launches · sha256 94c2078b7766
Training runs
1
Compute
TRAIN training server (CPU)
GPU hours
—
no GPU used
CPU time
103.5 s
Provider cost
$0.00
nothing purchased

Benchmarks

held-out test set · change vs v0.17
Spotting crashes (−70% within 1h)
0.961
Spotting survivors (still traded after 1h)
0.960
Spotting 2× runs (within 1h)
0.855
Calling the 1h direction (down / flat / up)
56.0%
Spotting creator dumps (within 1h)
0.871
Spotting crashes (within 6h)
insufficient data
Spotting survivors (6h)
insufficient data
Calling the 6h direction
insufficient data
Spotting survivors (24h)
insufficient data
Spotting migrations (within 24h)
insufficient data

Validation metrics (during training)

collapse_1h
AUC 0.995
n=809
collapse_6h
—
n=0
reach_2x_1h
AUC 0.923
n=809
survival_1h
AUC 0.961
n=809
survival_6h
—
n=0
survival_24h
—
n=0
migration_24h
—
n=0
trajectory_1h
acc 97.7%
n=809
trajectory_6h
—
n=0
creator_exit_1h
AUC 0.797
n=202

Validation launches are separate from training launches and from the locked test set. Benchmarks above are the public numbers.

What this version relies on

held-out validation launches
Collapse risk · 1h
Volatility, last 5 minutes−0.014
Trades per wallet−0.005
Buy share of volume, last 5 minutes−0.002
Creator's current holding−0.001
Holder growth, last 5 minutes−0.001
Trades, last 5 minutes−0.001
Survival · 1h
Net SOL into the curve−0.126
Buy share of volume since launch−0.011
Creator's current holding−0.010
Volume acceleration−0.006
Share of tiny (<$3) trades−0.005
Buy volume from profitable wallets−0.005
Reaches 2× · 1h
Price change, last minute−0.035
Holders (from trades)−0.032
Net SOL into the curve−0.026
Holder growth, last 5 minutes−0.021
Creator's current holding−0.017
Buy share of volume since launch−0.012
Creator exit · 1h
Creator's current holding−0.071
Net SOL into the curve−0.062
Age of the launch−0.036
Buy share of volume since launch−0.016
Buy share of volume, last 5 minutes−0.009
Volume acceleration−0.009

For each signal, how much the score drops when that signal is scrambled across launches. TRAIN learns these weights itself from the data; nobody hand-codes them.

Training runs

Attempt 1Succeeded
05 Oct 17:07 · 00:01:39 · 6,361 examples · $0.00

Timeline

  1. 05 Oct 17:09
    TRAIN v0.18 deployed — now live
    average score across 5 tests 0.843 vs 0.822 for the live v0.09
  2. 05 Oct 17:09
    Benchmark completed: TRAIN v0.18
    Crash-spotting score: 0.956 (n=491).
  3. 05 Oct 17:09
    Benchmark started: TRAIN v0.18
  4. 05 Oct 17:09
    Training completed: TRAIN v0.18
    00:01:39 · trained on 6361 examples · $0 compute cost
  5. 05 Oct 17:07
    Training started: TRAIN v0.18
    Solana Launch Dataset v10 · decision-tree ensemble · TRAIN training server
  6. 05 Oct 17:07
    TRAIN v0.18 queued
    Adds launch structure — bundled buys in the creation slot, snipers in the first 10 seconds and how much they've sold, first-minute buying, organic-volume checks (volume per wallet, back-and-forth wallets, tiny bump trades), KOL buys, and the dev's launches across all pump.fun coins.