Engineering Database · Software
AI Evaluation Harness
SW-002ReleasedArtificial IntelligenceRevised 2026-03-02
Reproducible evaluation harness for language and embedding models with structured artefact output.
OVERVIEW
Overview
Description
Deterministic evaluation harness producing signed, reproducible artefacts.
Current phase
Released — v1 in use across AI Lab.
Objectives
- Reproducibility across runs
- Structured artefact output
Engineering challenges
- Model non-determinism
- Dataset provenance