JOURDANLABS
CRUCIBLE

Open research
and validation
infrastructure.

CRUCIBLE is the JourdanLabs research division. It runs the benchmark program, maintains the COSMIC methodology playbook, and operates VANTAGE — the diagnostic suite that validates the other four divisions.

Crucible research artifact - open book with laboratory flask
What's in CRUCIBLE
VANTAGE artifact

VANTAGE

Flagship diagnostic suite. Sealed receipts, per-task scores, BCa confidence intervals.

RAVEN artifact

RAVEN

Open Source · Memory Validation

Seven-engine memory validation pipeline for AI agents. Runs after retrieval to filter contradictions, stale facts, and unranked noise.

Explore RAVEN →
BENCHMARKS artifact

BENCHMARKS

Six public benchmarks across pharmacovigilance, entity resolution, SOC, verification, semantic search, calibration.

METHODOLOGY artifact

METHODOLOGY

Seven non-negotiable principles. Sealed corpora, honest baselines, per-fix attribution, refusal as a feature.

Why Open Research?

CRUCIBLE benchmarks are public because the claim that COSMIC outperforms baselines in regulated domains is only credible if it can be verified. Engine implementations are proprietary — the corpus, scoring harnesses, and baseline code are not. Anyone can run the baselines. Anyone can verify the corpus SHA. Anyone can point their own pipeline at our scoring harness and compare.

This is how SuperGLUE, HELM, and other credible benchmark programs operate. We follow the same model.