DoAIRight
Benchmark Methodology

Exactly how a score is computed

Every dimension below is read live from the shipped scoring engine — the same catalog that scores every run. Nothing here is illustrative.

This page documents the scoring engine (the math), not a dataset or a leaderboard. A separate public Apache-2.0 benchmark repository with an authored, versioned public dataset is a planned follow-up that does not exist yet — this page shows “here is exactly how a score is computed,” not “here is a public leaderboard.”

Depth
claritydeterministic

Readability of the response (Flesch-style), computed from the text alone.

concisenessdeterministic

Filler/redundancy density — penalizes padding without penalizing genuine detail.

structuredeterministic

Paragraph/list/heading structure appropriate to the response's length and content.

grammardeterministic

A deterministic grammar-proxy check (no LLM call).

Semantic
semantic_similaritydeterministic

Cosine similarity between the response and its reference materials, via a pinned embedding model.

retrieval_relevancedeterministic

Cosine similarity between the retrieval query and the reference materials actually used.

Grounding
faithfulnessjudged

Share of claims in the response that are entailed by the reference materials (claim-level entailment via an LLM judge).

hallucinationjudged

Share of claims that are contradicted or unsupported by the reference materials (inverse of faithfulness's coverage).

attributionjudged

Whether supported claims can be traced to a specific reference span.

context_usejudged

How much of the available reference material the response actually draws on.

Quality
relevancejudged

LLM-judged: does the response actually answer the prompt.

completenessjudged

LLM-judged: does the response cover what the prompt asked for.

accuracyjudged

LLM-judged: factual correctness of claims not checked by the deterministic grounding pipeline.

well_formednessjudged

LLM-judged: is the response well-formed for its intended format (e.g. valid structure for a requested format).

Safety
harm_absencejudged

Multi-label harm/safety classification. Currently always N/A — no safety classifier is deployed yet (honestly labeled, never a fabricated 0).