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 v0.23

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
1023
Config hash
9fb593fcc0b77902
Created
05 Oct 22:38
Completed
05 Oct 22:40
Artifact
models/v0.23/model.json
sha256 d63e243d6458ce08…

Dataset & compute

Dataset
Solana Launch Dataset v15
9,754 snapshots · 1,957 launches · sha256 bbc271ff0196
Training runs
1
Compute
TRAIN training server (CPU)
GPU hours
—
no GPU used
CPU time
148.7 s
Provider cost
$0.00
nothing purchased

Benchmarks

held-out test set · change vs v0.22
Spotting crashes (−70% within 1h)
0.947
Spotting survivors (still traded after 1h)
0.962
Spotting 2× runs (within 1h)
0.857
Calling the 1h direction (down / flat / up)
53.1%
Spotting creator dumps (within 1h)
0.881
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=1164
collapse_6h
AUC 0.500
n=4
reach_2x_1h
AUC 0.917
n=1164
survival_1h
AUC 0.931
n=1164
survival_6h
—
n=4
survival_24h
—
n=0
migration_24h
—
n=0
trajectory_1h
acc 97.6%
n=1164
trajectory_6h
acc 25.0%
n=4
creator_exit_1h
AUC 0.835
n=267

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−0.018
Net SOL into the curve−0.011
Trades per wallet−0.004
Buy share of volume, last 5 minutes−0.003
Holder growth, last 5 minutes−0.002
Price change, last minute−0.001
Survival · 1h
Net SOL into the curve−0.146
Volume acceleration−0.017
Trades per wallet−0.008
Holders (from trades)−0.006
Buyers with a profitable record−0.003
Creator's current holding−0.003
Reaches 2× · 1h
Net SOL into the curve−0.050
Price change, last minute−0.025
Holders (from trades)−0.024
Holder growth, last 5 minutes−0.016
Creator's current holding−0.013
Number of trades−0.005
Creator exit · 1h
Creator's current holding−0.097
Net SOL into the curve−0.052
Age of the launch−0.020
Buy share of volume since launch−0.014
Creator's prior launches−0.007
Buy share of volume, last 5 minutes−0.006

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 22:38 · 00:02:25 · 8,590 examples · $0.00

Timeline

  1. 05 Oct 22:40
    TRAIN v0.23 not deployed; v0.19 stays live
    Reason: average score across 5 tests 0.842 vs 0.850 for the live v0.19.
  2. 05 Oct 22:40
    Benchmark completed: TRAIN v0.23
    Crash-spotting score regressed 0.949 → 0.937 vs live v0.19.
  3. 05 Oct 22:40
    Benchmark started: TRAIN v0.23
  4. 05 Oct 22:40
    Training completed: TRAIN v0.23
    00:02:25 · trained on 8590 examples · $0 compute cost
  5. 05 Oct 22:38
    Training started: TRAIN v0.23
    Solana Launch Dataset v15 · decision-tree ensemble · TRAIN training server
  6. 05 Oct 22:38
    TRAIN v0.23 queued
    Same configuration as v0.18 (best so far on the headline benchmark), retrained on the newest dataset.