Analysis / validation and data coverage
Analytics
Inspect measured data coverage and model validation metrics. Estimates remain distinct from the held-out benchmark evidence.
Measured observations
0
Source dataset rows
Model sizes
0
Unique observed models
Observed stacks
0
Precision / quantization / backend
Context values
0
Observed token lengths
Observed scaling
Measured generation rate by parameter count. Each point is an empirical observation.
Held-out model validation
Saved evaluation metrics; validation method is leave-one-model-out. These are aggregate errors, not per-prediction intervals.
Evidence coverage
Breadth of empirical support for the current model artifacts.
0 benchmark rows
Raw measured data remains unmodified. Estimates point back to this source distribution.
0 observed stacks
Unseen combinations are model-based estimates, optionally context-calibrated against nearby rows.
Supporting rows, not confidence
The API exposes matching empirical row counts and aggregate held-out MAE; it does not claim a per-prediction probability.