ML Lab

Full model comparison — leaderboards, overfit gaps, SHAP features, holdout scatter, embryo calibration.

Loading ML reports…

Best model

—

Within 2× of true price

—

holdout accuracy

Training rows

—

0 features

Models benchmarked

seeds —

Pareto frontier — R² vs training time

BoostingTreeKernelLinearNeuralEnsembleOther

Median R² by model family

Boosting and kernel methods dominate. Linear baselines and stacked ensembles follow. Neural nets overfit a small dataset.

Currency normalization (FX)

Could not load /meta/fx_rates — the API may be missing enrichment/currency.py.

Artifact freshness

Report	Built at	Age	Training rows	On disk