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

TRAIN v0.22

Same configuration as v0.18 (best so far on the headline benchmark), retrained on the newest dataset.
Completed

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
1022
Config hash
cb2c64c41b0ec51f
Created
05 Oct 21:22
Completed
05 Oct 21:24
Artifact
models/v0.22/model.json
sha256 11bd968eb03167ae…

Dataset & compute

Dataset
Solana Launch Dataset v14
9,235 snapshots · 1,853 launches · sha256 83676a982760
Training runs
1
Compute
TRAIN training server (CPU)
GPU hours
—
no GPU used
CPU time
136.9 s
Provider cost
$0.00
nothing purchased

Benchmarks

held-out test set · change vs v0.21
Spotting crashes (−70% within 1h)
0.948
Spotting survivors (still traded after 1h)
0.969
Spotting 2× runs (within 1h)
0.853
Calling the 1h direction (down / flat / up)
52.9%
Spotting creator dumps (within 1h)
0.880
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 0.994
n=1094
collapse_6h
—
n=0
reach_2x_1h
AUC 0.919
n=1094
survival_1h
AUC 0.945
n=1094
survival_6h
—
n=0
survival_24h
—
n=0
migration_24h
—
n=0
trajectory_1h
acc 97.3%
n=1094
trajectory_6h
—
n=0
creator_exit_1h
AUC 0.832
n=250

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−0.012
Net SOL into the curve−0.004
Buy share of volume, last 5 minutes−0.004
Trades per wallet−0.003
Trades, last 5 minutes−0.002
Holder growth, last 5 minutes−0.001
Survival · 1h
Net SOL into the curve−0.144
Buy share of volume since launch−0.012
Volume acceleration−0.007
Share of tiny (<$3) trades−0.005
Top-10 share of all held tokens−0.003
Creator's current holding−0.002
Reaches 2× · 1h
Net SOL into the curve−0.060
Holders (from trades)−0.023
Price change, last minute−0.020
Holder growth, last 5 minutes−0.011
Number of trades−0.003
Supply bought in the first minute−0.003
Creator exit · 1h
Creator's current holding−0.073
Net SOL into the curve−0.069
Age of the launch−0.042
Buy share of volume since launch−0.011
Volume since launch−0.008
Creator's prior launches−0.004

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 21:22 · 00:02:13 · 8,141 examples · $0.00

Timeline

  1. 05 Oct 21:24
    TRAIN v0.22 not deployed; v0.19 stays live
    Reason: average score across 5 tests 0.841 vs 0.850 for the live v0.19.
  2. 05 Oct 21:24
    Benchmark completed: TRAIN v0.22
    Crash-spotting score regressed 0.949 → 0.942 vs live v0.19.
  3. 05 Oct 21:24
    Benchmark started: TRAIN v0.22
  4. 05 Oct 21:24
    Training completed: TRAIN v0.22
    00:02:13 · trained on 8141 examples · $0 compute cost
  5. 05 Oct 21:22
    Training started: TRAIN v0.22
    Solana Launch Dataset v14 · decision-tree ensemble · TRAIN training server
  6. 05 Oct 21:22
    TRAIN v0.22 queued
    Same configuration as v0.18 (best so far on the headline benchmark), retrained on the newest dataset.
  7. 05 Oct 21:22
    Enough new data for another training run
    5,448 new labelled snapshots since Solana Launch Dataset v9.