Benchby BizLegal AI

Sample measurement report · anonymized

What you receive.

The structure below is a full Bench report in miniature — computed by the same deterministic engine that produces client reports, on a worked example of ten evaluations. Named results are always private; published data is aggregated or anonymized.

Illustrative dataset — replaced by our own self-benchmark after its first run

Headline metrics · 10 evaluations

17.6/25
Mean rubric score
70%
Accuracy
20%
Hallucination rate
20%
Critical errors
70%
Citation reliability
20%
Missing-law rate

Error taxonomy

Where it fails, classified.

Hallucinated authority2
Omission2
Bad reasoning1
Misstatement1
Unreliable citation1
Wrong jurisdiction1

Dimension means · 0–5

The shape of the system.

jurisdictional4.10
correctness3.30
completeness3.20
reasoning3.40
hallucination3.60

Per-area breakdown

Jurisdictional accuracy by practice area.

Practice areaItemsMean score (0–25)
casp318.7
market abuse18
scope222
stablecoin reserves215.5
whitepaper218.5

Corrections

Every failure comes back fixed.

Excerpt · item s-05 · severity: criticalillustrative

Finding: The output cited a “MiCA Article 112 reserve attestation regime” that does not exist, presented with an invented EBA guideline number.

Gold-standard correction: Reserve-of-assets obligations for ART issuers sit in Articles 36–38 of Regulation (EU) 2023/1114 — composition, custody and segregation of the reserve — with own-funds requirements in Article 35. The correction states the applicable articles and the operative rule, verified by an EU-regulatory practitioner.

Request your measurementHow these numbers are computed