CLI reference¶
Run ragproof COMMAND --help for every option. All commands return a non-zero exit code for validation failures, evaluation errors that cannot be tolerated, or failed regression gates.
| Command | Purpose |
|---|---|
init |
Create a starter YAML configuration and JSONL dataset. |
validate |
Validate configuration and dataset structure without calling the RAG API. |
run |
Evaluate the selected dataset and write a versioned run artifact. |
compare |
Compare baseline and current runs; exit 1 when a gate fails. |
report |
Render HTML, Markdown, CSV, JUnit XML, or SARIF. |
probe |
Call an HTTP adapter and suggest response mappings without printing secrets or response content. |
dataset-lint |
Check schema, duplicate IDs/questions, and near duplicates. |
dataset-manifest |
Print a stable dataset fingerprint and summary. |
benchmark-manifest-lint |
Verify benchmark paths, hashes, schemas, and license evidence. |
trend |
Summarize multiple runs with bootstrap confidence intervals. |
bisect |
Find the first run file that crosses a threshold. |
threshold-recommend |
Recommend conservative gates from a run history. |
judge-check |
Check an optional judge endpoint and structured score parsing. |
judge-calibrate |
Measure judge calibration against supplied golden labels. |
Core workflow¶
ragproof validate -c ragproof.yaml
ragproof run -c ragproof.yaml -o runs/current.json
ragproof compare \
--baseline runs/baseline.json \
--current runs/current.json \
--threshold "recall@5=0.70" \
--max "error_rate<=0.05" \
--max-relative-drop "citation_coverage=5%"
ragproof report runs/current.json -o runs/report.html
Gate families¶
| Option | Meaning |
|---|---|
--threshold metric=value |
Minimum absolute value for higher-is-better metrics. |
--max metric<=value |
Maximum absolute value for lower-is-better metrics. |
--min-delta metric=value |
Minimum allowed current-minus-baseline change. |
--max-relative-drop metric=5% |
Maximum allowed proportional drop. |
--group-threshold dimension:value:metric=value |
Minimum threshold for one tag or difficulty slice. |
--group-max dimension:value:metric<=value |
Maximum threshold for one slice. |
--min-coverage field=value |
Minimum availability for a field or metric. |
--require-field field |
Require complete availability. |
--min-sample-count count |
Reject undersized current runs. |
--policy path |
Load the same gate model from YAML or JSON. |
Baseline and current artifacts must have compatible provenance unless --allow-provenance-mismatch is explicitly supplied.
Dataset tools¶
ragproof dataset-lint dataset.jsonl
ragproof dataset-manifest dataset.jsonl --json
ragproof benchmark-manifest-lint examples/benchmark-manifest.json
See the dataset reference for supported fields and input formats.