Predictions interface contract
The stable wrapped predictions boundary that lets validation and evaluation stay independent of the training or inference framework.
Glossary
Terms for the predictions boundary, protocol identity, maturity lanes, and comparison evidence.
Core terms
Wrapped predictions
A predictions payload with schema version, prediction entries, and metadata needed to interpret how the artifact was produced.
Protocol identity
The recorded dataset, split, metric settings, thresholds, and other evaluation choices that determine whether reports are comparable.
Export settings
Producer-side preprocessing and output settings recorded in meta.export_settings so prediction artifacts can be interpreted and compared.
Stable lane
Prediction validation and evaluation capabilities that form the product's supported core boundary.
Bridge lane
An adapter, exporter, or external training handoff that emits the stable predictions interface contract without replacing it.
Benchmark lane
A qualification path for runtime or backend parity used after stable prediction evaluation is already working.
Research lane
An opt-in experiment over a stable artifact whose output does not establish production readiness.
Skipped lane
A path that did not execute because a capability, artifact, runtime, or prerequisite was unavailable; it is not a passing result.
Comparable report
A report whose task, data, protocol, predictions provenance, and metric semantics align with the report it is compared against.