Skip to content

Add llm evaluation against human accessibility audits - #5

Open
reiganarmstrong wants to merge 152 commits into
mainfrom
feat/model-evaluation
Open

reiganarmstrong wants to merge 152 commits into
mainfrom
feat/model-evaluation

Conversation

@reiganarmstrong

Copy link
Copy Markdown

Adds a CLI eval tool that compares findings from the va-dat tool and those in a provided human accessibility audit. This is especially useful for comparing the price to performance ratios between various llms in the va-dat tool. Details of the CLI tool are documented in the following links:

Document the CLI workflow, eligibility classifications, current Luna proof-of-concept status, and remaining work. Expand the domain glossary for the human-audit reference workbook and non-model eligibility classes.
Use Human finding mapping names for saved artifacts, local lookup paths, and manifest fields. Require approved mappings throughout execution and reporting, and remove tests for unsupported review behavior.

BREAKING CHANGE: Remove old CLI aliases and storage names. Saved runs using the previous format require fresh preparation.
Move normalized Programmatic findings into artifacts and add concise run READMEs with clear preview and live instructions.

Keep mapping preparation focused on CSVs and instructions, rebuilding approval evidence from the original audit and snapshots.
Move detailed evaluator documentation into its package README and provide a complete quick start in the root README. Clarify the human audited Excel file terminology in documentation and local variables, and include the editor environment setting.
@reiganarmstrong reiganarmstrong self-assigned this Oct 25, 2026
@reiganarmstrong
reiganarmstrong marked this pull request as ready for review October 25, 2026 20:08
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant