launch-kit evals: the count is a script and the tier is a procedure - #19
Merged
Conversation
evals/count.mjs runs vos validate --json over every kit a run produced and prints the one deterministic line a results file ends on (N problems across M assets on K kits), exit 1 on any problem. Re-run on the two adopter kits it reproduces the published number, 0 across 22. evals/launch-kit/RUN.md is the procedure for a model tier: the fixtures, the skill installed project-scope for the agent under test, the blind A/B against a no-skill copy (the baseline every scenario file asked for), the non-interactive claude -p and codex exec commands, the results file shape. The scenarios header, the results README and the repo README point at it.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The launch-kit eval loop had a frozen scenario set and a deterministic number, but the number was computed by hand and "run it on a second tier" had no written procedure. This PR gives both a mechanism.
evals/count.mjs: the count as a dependency-free script overvos validate <kit.json> --json. Walks run directories for kits, prints one line per kit with its problems and one total in the results-file form (N problems across M assets on K kits), exits 1 on any problem so a judge or a CI step cannot pass a run by not looking. Re-run on the two adopter kits it reproduces the published number,0 problems across 22 assets on 2 kits; a tampered manifest fails it in words.evals/launch-kit/RUN.md: the procedure for a tier. Fixtures, the skill installed project-scope for the agent under test, the blind A/B against a no-skill copy (the baseline every scenario file asked for and no run had recorded), the non-interactiveclaude -pandcodex execcommands, the results file shape ending on the count line.node scripts/check.mjspasses.