Skip to content

API Evidentia Eval

evidentia-wiki-sync[bot] edited this page May 29, 2026 · 1 revision

evidentia-eval API

Auto-generated page. This public-surface index is generated from the package source by scripts/wiki/sync_api_docs.py. Do not edit it by hand; change the package and re-run the generator (uv run python scripts/wiki/sync_api_docs.py).

DFAH (Decision-Faithfulness Assessment Harness) determinism + faithfulness eval harness for Evidentia — dev-time AI-output quality gates

Install: pip install evidentia-eval. Import root: evidentia_eval.

Public API

The names exported by evidentia_eval.__all__ (re-exported at the package root):

Symbol Summary
CLAIM_EXTRACTION_PROMPT
DEFAULT_FAITHFULNESS_THRESHOLD
DEFAULT_SEMANTIC_MODEL
DEFAULT_SEMANTIC_THRESHOLD
DFAHarness Run determinism + replay checks against a generator function.
DeterminismResult One prompt's determinism outcome over N samples.
EvalResult Top-level harness output covering all prompts in one run.
EvalSample One prompt's inputs (immutable; audit-trail-stable).
FaithfulnessResult One claim's faithfulness outcome.
PromptFaithfulnessResult Per-prompt faithfulness aggregation (v0.8.4 P1 wiring).
ReplayResult One prompt's replay-equivalence outcome.
SemanticFaithfulnessNotAvailableError Raised when sentence-transformers is not installed.
determinism_score Compute :class:DeterminismResult from raw sample outputs.
extract_claims Extract atomic claims from generated text via LLM.
faithfulness_score Compute :class:FaithfulnessResult for one claim.
faithfulness_score_semantic Compute :class:FaithfulnessResult via sentence-embeddings.
hash_output SHA-256 hex digest of the normalized output.
normalize_for_determinism Apply the canonical normalization for determinism comparison.
replay_equivalent Build a :class:ReplayResult from two raw outputs.
sign_eval_result Write an EvalResult to disk + produce its Sigstore bundle.
verify_eval_result Verify a signed EvalResult JSON against its Sigstore bundle.

Submodules

Public submodules + subpackages of evidentia_eval (import directly, e.g. from evidentia_eval.claim_extraction import ...):

  • evidentia_eval.claim_extraction
  • evidentia_eval.faithfulness
  • evidentia_eval.faithfulness_semantic
  • evidentia_eval.harness
  • evidentia_eval.metrics
  • evidentia_eval.seeds
  • evidentia_eval.signing

Full signatures

This page is a public-surface index. For full, always-current type signatures and rendered docstrings, see the live API reference (built from the source with mkdocstrings) or read the source.

Clone this wiki locally