Glossary

API Profile
A Project record describing one reusable synchronous HTTP JSON inference operation and its closed request and response mappings.

Claim
The sealed terminal record of an Evaluation. Its outcome is completed, failed, or cancelled.

Evaluation
An assessment of a fixed Prediction Set using selected Performance metrics and supported Product modules.

Evaluation Preset
A reusable, task-typed selection of metrics, module methods, transforms, and thresholds. It does not own inputs or a Prediction Set.

Evidence
Content-identified metrics, tables, images, logs, and diagnostics generated during an Evaluation.

Import
Copying an external source into a Project as an independently owned Material.

Inference Source
The local Model or API Profile used by Prediction Generation.

Material
A content-identified Dataset, Model, or other input copied into and owned by a Project.

Maximum allowed degradation
The transform-level percentage limit used to decide each selected Robustness severity result.

Model
A Project record that owns one copied model file together with its settings and verification.

Performance metric
A top-level scientific measurement of clean-input predictions, such as F1, mean absolute error, or mean IoU.

Prediction Generation
The execution that produces clean and selected transformed predictions without calculating Evaluation metrics.

Prediction Set
A sealed, content-identified collection of predictions, coverage, provenance, and Stage Claims that compatible Evaluations may score.

Product module
One of the three fixed Evaluation perspectives: Robustness, Fairness, or Explainability.

Project
A visible folder containing the Materials, decisions, Claims, and Evidence for one body of evaluation work.

Verification
The independent decision that retained Project files and identities agree. It is not the same as worker success.