Everything Model Zero does, in order

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

It had stalled (a failed run (retrying)); the watchdog restarted that step.

Solana Launch Dataset v31 · boosted decision trees · Model Zero training server (CPU)

Attempt 2. The failed attempt stays in the history.

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Reason: training process terminated before finishing (no final status written). Cost: $0.00. Awaiting a controlled retry.

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Solana Launch Dataset v31 · boosted decision trees · Model Zero training server (CPU)

62,359 snapshots from 12,495 launches.

Automatic search from v0.39: min samples leaf 10 → 40, with automatic signal discovery.

58,422 new labelled snapshots since Solana Launch Dataset v9.

Recorded as an outcome label (never used as a feature).

average score across 10 tests 0.876 vs 0.876 for the live v0.38

Spotting crashes (−70% within 1h) regressed 0.972 → 0.971 vs live v0.38.

Every candidate was tested on held-out folds of the training launches.

01:47:26 · trained on 53525 examples · $0 compute cost

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Trained on 34,809 market observations; live: 0.888 average score vs 0.888 for the previous version.

Solana Launch Dataset v30 · boosted decision trees · Model Zero training server (CPU)

Same configuration as v0.38 (best so far on the headline benchmark), retrained on the newest dataset.

57,237 new labelled snapshots since Solana Launch Dataset v9.

average score across 10 tests 0.876 vs 0.867 for the live v0.37

Spotting crashes (−70% within 1h) regressed 0.972 → 0.972 vs live v0.37.

Every candidate was tested on held-out folds of the training launches.

03:40:00 · trained on 53525 examples · $0 compute cost

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Solana Launch Dataset v30 · boosted decision trees · Model Zero training server (CPU)

61,174 snapshots from 12,258 launches.

Adds launch context known at the moment of launch: time of day, weekday, name and ticker style, and which launch tool created the coin with that tool's track record on earlier launches.

57,237 new labelled snapshots since Solana Launch Dataset v9.

average score across 10 tests 0.867 vs 0.849 for the live v0.33

Spotting crashes (−70% within 1h): 0.972 (n=3543).

00:32:38 · trained on 47355 examples · $0 compute cost

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

It had stalled (a failed run (retrying)); the watchdog restarted that step.

Solana Launch Dataset v29 · boosted decision trees · Model Zero training server (CPU)

Attempt 2. The failed attempt stays in the history.

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Reason: training process terminated before finishing (no final status written). Cost: $0.00. Awaiting a controlled retry.

Recorded as an outcome label (never used as a feature).

Solana Launch Dataset v29 · boosted decision trees · Model Zero training server (CPU)

3543 held-out snapshots from 1639 launches. No model can train on these.

54,194 snapshots from 10,862 launches.

It had stalled (dataset preparation); the watchdog restarted that step.

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Trained on 31,693 market observations; live: 0.891 average score vs 0.887 for the previous version.

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).

Recorded as an outcome label (never used as a feature).