eval 0.0.5

Pure Dart LLM evaluation helpers for tests, including judge-based matchers, RAG scoring, and statistics.

0.0.5

• Increase timeout for evals to 10 minutes, expose timeout attribute

0.0.4

• Fix parseMarkdownBody so nested frontmatter collections are mutable. Lists and maps nested inside the parsed frontmatter previously came back as immutable YamlList/YamlMap instances and threw UnsupportedError on assignment.
• Export deepConvertYaml, a recursive helper that converts YamlMap/YamlList trees to mutable Map<String, dynamic> and List structures.

0.0.3

• Overhaul the README and public Dartdoc so eval(...) is documented as the primary workflow instead of raw test(...).
• Document the full exported matcher surface, including the previously omitted JSON array, schema-path, frontmatter schema, and RAG matchers.
• Refresh the bundled example to show an end-to-end eval(...) run with sync and async assertions.
• Reset internal eval run state after each run so expect(...) cleanly falls back to normal test behavior outside an active eval.

0.0.2

• Align package metadata and documentation with the published API.
• Fix async LLM and RAG matcher behavior so sync expect(...) usage fails with clear guidance instead of silently succeeding.
• Fix APICallQueue recovery so one failed request does not poison later queued calls.
• Preserve detailed evaluateRag() metadata including relevant context indices, unsupported claims, and joined metric reasons.
• Treat empty frontmatter as valid frontmatter and reject malformed YAML with closing delimiters.
• Distinguish missing paths from explicit null values in schema-based path matchers.

0.0.1

• Initial public release of the eval package.
• Added string, JSON, schema, frontmatter, distance, LLM-judge, and RAG matchers.
• Added aggregate statistics and prompt comparison helpers.
• Added the APICallService abstraction and the bundled Claude example service.