Automatic search: a new configuration
Model Zero v0.46
Automatic search from v0.40: adds who funded the early buyers, the creator's wallet and real on-chain holders, plus every newer signal, with automatic signal discovery.
Completed
Definition
Signals used
… + funding links & on-chain holders
Technical details
Artifact
models/v0.46/model.json
sha256 74dd3c2459c3c433…
Dataset & compute
Dataset
Solana Launch Dataset v37 (f10)
124 snapshots · 28 launches · sha256 e41059f88e9b
Compute
Model Zero training server (CPU)
Benchmarks
held-out test set · change vs v0.45
Spotting crashes (−70% within 1h)
8 tests
not enough data yet
Spotting survivors (still traded after 1h)
8 tests
not enough data yet
Spotting 2× runs (within 1h)
8 tests
not enough data yet
Calling the 1h direction (down / flat / up)
8 tests
not enough data yet
Spotting creator dumps (within 1h)
7 tests
not enough data yet
Spotting crashes (within 6h)
8 tests
not enough data yet
Spotting survivors (6h)
8 tests
not enough data yet
Validation metrics (during training)
Signals this version discovered
No candidate improved predictions enough to keep.
0 candidate signals tested on held-out folds of the training launches. Cross-validated AUC → .
What this version relies on
held-out validation launches
This version was trained before Model Zero started measuring which signals each model relies on.
For each signal, how much the score drops when that signal is scrambled across launches. Model Zero learns these weights itself from the data; nobody hand-codes them.
Training runs
Attempt 1Succeeded
11 Oct 11:55 · 00:00:13 · 99 examples · $0.00
Timeline
11 Oct 11:55
Model Zero v0.46 not deployed; v0.40 stays live
Reason: no shared benchmarks yet; keeping the live model.
11 Oct 11:55
Benchmark completed: Model Zero v0.46
No locked test set yet (not enough held-out launches with observed outcomes). Benchmarks will be computed when it is locked.
11 Oct 11:55
Benchmark started: Model Zero v0.46
11 Oct 11:55
Model Zero v0.46 searched for new signals; none helped enough to keep
Every candidate was tested on held-out folds of the training launches.
11 Oct 11:55
Training completed: Model Zero v0.46
00:00:13 · trained on 99 examples · $0 compute cost
11 Oct 11:55
Training started: Model Zero v0.46
Solana Launch Dataset v37 (f10) · boosted decision trees · Model Zero training server (CPU)
11 Oct 11:55
Model Zero v0.46 queued
Automatic search from v0.40: adds who funded the early buyers, the creator's wallet and real on-chain holders, plus every newer signal, with automatic signal discovery.