Skip to content

Use strict accuracy data and expand labeled cases #320

Description

@1sgtpepper

Problem

The accuracy suite uses a hand-parsed YAML-like format that can silently accept incomplete
input. Its 18 cases contain expected diagnostics from only eight of the 38 catalog codes,
which is not enough evidence for per-rule release thresholds.

Expected behavior

Replace the hand parser with one strict versioned format supported by the standard library
or an existing dependency. Reject unknown and missing fields. Expand independently labeled
positive and negative cases with rule, source format, scientific domain, provenance, and
license or synthetic status.

Run every case through the public analysis path and keep expectations human-authored. Do
not add a canary differential or precision/recall gate until enough independently labeled
data and baseline variance exist. Align future corpus work with #198#201.

Metadata

Metadata

Assignees

No one assigned

    Labels

    area:testsTest fixtures, golden outputs, and test coverage.enhancementNew feature or requestp2Medium-priority issue.

    Projects

    No projects

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions