Reports and benchmark releases.

This page will hold public evaluation reports, benchmark releases, calibration notes, corrections, and field guidance. It is structured now so the first release has a clear home.

Forthcoming artifacts

Benchmark release report Scope, methods, model/system behavior, key findings, limitations, and interpretation guidance.

Evaluation reports System-specific or cohort-level reports tied to named versions, configurations, and claims limits.

Calibration notes Reviewer agreement, synthetic judge limitations, adjudication process, and remaining validity gaps.

Field guidance Plain-language reports for faith institutions making adoption, procurement, or policy decisions.

Report principles

Every public result should name the benchmark version and evaluated configuration.
Reports should distinguish evidence from endorsement.
Limitations and correction paths should be visible, not buried.
Participant relationships and funding constraints should be disclosed by category where relevant.