Pre-launch· Token not live yet — no rewards or treasury until launchWhat this means
← Model
Adds global fees paid

TRAIN v0.21

Adds global fees paid : total fees and Jito tips, fees per minute, fees relative to volume, transactions per paying wallet and failed-transaction share — then lets automatic discovery choose which change-signals to add on top.
Completed

Definition

Model type
Decision-tree ensemble
Signals used
… + global fees
Technical details
max_depth
3
subsample
0.8
n_estimators
300
learning_rate
0.03
min_samples_leaf
10
Random seed
1021
Config hash
779894d25dc21fc8
Created
05 Oct 19:21
Completed
05 Oct 19:23
Artifact
models/v0.21/model.json
sha256 52f68b0630c93eb5…

Dataset & compute

Dataset
Solana Launch Dataset v13
2,018 snapshots · 406 launches · sha256 1cbbdf250876
Training runs
1
Compute
TRAIN training server (CPU)
GPU hours
—
no GPU used
CPU time
1,253.8 s
Provider cost
$0.00
nothing purchased

Benchmarks

held-out test set · change vs v0.20
Spotting crashes (−70% within 1h)
0.924
Spotting survivors (still traded after 1h)
0.944
Spotting 2× runs (within 1h)
0.732
Calling the 1h direction (down / flat / up)
51.3%
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 0.984
n=210
collapse_6h
—
n=0
reach_2x_1h
—
n=210
survival_1h
AUC 0.836
n=210
survival_6h
—
n=0
survival_24h
—
n=0
migration_24h
—
n=0
trajectory_1h
acc 97.1%
n=210
trajectory_6h
—
n=0
creator_exit_1h
AUC 0.921
n=37

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

Signals this version discovered

45 candidate signals tested on held-out folds of the training launches. Cross-validated AUC 0.827 → 0.860.

What this version relies on

held-out validation launches
Collapse risk · 1h
Creator's prior launches−0.012
Volatility, last 5 minutes−0.005
Trades per wallet−0.005
Price change, last minute−0.004
Share of tiny (<$3) trades−0.003
Trades, last 5 minutes−0.002
Survival · 1h
Net SOL into the curve−0.162
Volume acceleration−0.070
Distinct trading wallets−0.011
X link set at launch−0.010
Volume per wallet−0.008
Volume, last minute−0.007
Creator exit · 1h
Creator's current holding−0.067
Net SOL into the curve−0.021
Creator's initial buy−0.020
Volume, last minute−0.018
Creator's prior launches−0.018
Volume, last 5 minutes−0.015

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 19:21 · 00:02:20 · 1,808 examples · $0.00

Timeline

  1. 05 Oct 19:23
    TRAIN v0.21 not deployed; v0.19 stays live
    Reason: average score across 1 tests 0.439 vs 0.584 for the live v0.19.
  2. 05 Oct 19:23
    Benchmark completed: TRAIN v0.21
    Crash-spotting score: not enough held-out data yet (n=67).
  3. 05 Oct 19:23
    Benchmark started: TRAIN v0.21
  4. 05 Oct 19:23
    TRAIN v0.21 discovered 3 useful new signals
    Change between snapshots: share of tiny (<$3) trades; Change between snapshots: buyers with a profitable record; Change between snapshots: price change, last minute. Score on held-out training launches 0.827 → 0.860.
  5. 05 Oct 19:23
    Training completed: TRAIN v0.21
    00:02:20 · trained on 1808 examples · $0 compute cost
  6. 05 Oct 19:21
    Training started: TRAIN v0.21
    Solana Launch Dataset v13 · decision-tree ensemble · TRAIN training server
  7. 05 Oct 19:21
    TRAIN v0.21 queued
    Adds global fees paid : total fees and Jito tips, fees per minute, fees relative to volume, transactions per paying wallet and failed-transaction share — then lets automatic discovery choose which change-signals to add on top.
  8. 05 Oct 19:21
    Enough new data for another training run
    4,758 new labelled snapshots since Solana Launch Dataset v9.