Skip to content
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension


Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
2 changes: 1 addition & 1 deletion evals/README.md
Original file line number Diff line number Diff line change
Expand Up @@ -36,7 +36,7 @@ Each run:

## When to run

When something the evals depend on changes, not on a schedule. Between changes the inputs are pinned: the skill text is in the repo, the model and effort are in `promptfooconfig.yaml`, and promptfoo, the Agent SDK and its bundled Claude Code are in `package-lock.json`. A repeat run mostly re-measures sampling noise, at about $5.50 for a cold run of both skills on both models (functional-design about $2.50, performance about $2.90); a run soon after another reuses the prompt cache and costs less.
When something the evals depend on changes, not on a schedule. Between changes the inputs are pinned: the skill text is in the repo, the model and effort are in `promptfooconfig.yaml`, and promptfoo, the Agent SDK and its bundled Claude Code are in `package-lock.json`. A repeat run mostly re-measures sampling noise, at about $7.50 for a cold run of both skills on both models (functional-design about $4.70, performance about $2.80); a run soon after another reuses the prompt cache and costs less.

- A skill with evals changes: before the PR merges, and before and after a rewrite.
- The cases or `promptfooconfig.yaml` change.
Expand Down
Loading
Loading