Open in Markdown reader

Model / Tool Evaluation Card

Candidate model/tool/version:

Task class: drafting / research / spreadsheet analysis / code assistance / browser exploration / verdict review.

Allowed environment and data class:

Golden corpus version:

Route: fabricated mock / live measured trial. Mock results cannot approve a real model.

Input/prompt version and actor-access boundary (live route):

Sample counts and metric denominators: binary cases / known failures / known passes; score ambiguity and instruction-boundary cases separately.

For the supplied mock outputs, write not measured for task completion, evidence completeness, latency and cost. For live runs, link actual output and measurement records; unavailable values remain unknown.

Known failure modes:

Human review requirement:

Approved prompt/playbook version:

Approval decision and reviewer: