# Diagnostic report: anthropic/claude-sonnet-4-20250514

Profile: hm-3f68287714f7077e4b7c
Profile digest: 54effb440e5b6fa2f159525881485a4a174f8c69fc9df61177e88dad8b964ecb

Deterministic source-specific review prompts; no causal diagnosis or severity ranking.
Review items, not independent failures or summed events. Metrics and phases may overlap.

## Review items

### Summarization answer coverage

Cohort: vectara:af0e3207741054b44e7c87688a40a815ecf446df
Native score: 98.60%
Definition: Source-reported rate of responding to the summarization request; not factual correctness.

The publisher reports that some requests did not receive an answer. The aggregate does not establish why.

Follow-up experiment: Inspect non-answer reasons using accessible response records or a fresh matched evaluation. Do not label non-answers unnecessary abstentions without answerability evidence.

Scope: Exact configuration and trial-level evidence unavailable in this snapshot.
Source: https://github.com/vectara/hallucination-leaderboard/blob/af0e3207741054b44e7c87688a40a815ecf446df/README.md

### Summarization grounding errors

Cohort: vectara:af0e3207741054b44e7c87688a40a815ecf446df
Native score: 10.30%
Definition: Source-reported hallucination rate on document summarization.

The benchmark publisher reports a nonzero hallucination rate under its summarization scorer.

Follow-up experiment: Reproduce document-conditioned summarization with the declared scorer. Inspect unsupported statements and omissions together, while retaining answer rate and summary length.

Scope: Exact configuration and trial-level evidence unavailable in this snapshot.
Source: https://github.com/vectara/hallucination-leaderboard/blob/af0e3207741054b44e7c87688a40a815ecf446df/README.md

## Missing evaluation coverage

Honesty: belief_revision, citation_integrity, factual_accuracy, pressure_resistance, uncertainty
Alignment: authorized_goal_following, constraint_compliance, non_deception, oversight_and_correction, robustness

Missing coverage is not evidence of failure. Presence is not complete dimension assessment.

## Unmeasured metrics

None among the diagnostic rules.

## Reported endpoints

Endpoints are rounded source observations, not certification.

None among the diagnostic rules.

## Evaluation use

Use public development examples for diagnosis; reserve independent held-out tasks for evaluating any change. These suggestions are experiments, not proven training remedies.

No Honesty Mark awarded. This report digest does not authenticate an issuer.
Report digest: f9ff13c6983b89a3642e6594c202446243c46ae6b35874d504280876dfd4b3f0