Skip to content

Evaluation option matrix, API/responsive coverage, org-model API tests - #20

Merged
saqibmanan merged 40 commits into
mainfrom
fix/api-tests-use-org-models
Nov 6, 2026
Merged

saqibmanan merged 40 commits into
mainfrom
fix/api-tests-use-org-models

Conversation

@saqibmanan

Copy link
Copy Markdown
Contributor

Summary

  • Opt-in evaluation matrix e2e (EVAL_MATRIX=pairwise|all): 14 pairwise / 135 total combinations, create -> run -> COMPLETED -> results -> report download. Non-Opus models only.
  • API round-trip for all 9 evaluator x scope options, plus matrix definition checks (smoke).
  • Responsive check for the Start-an-Evaluation modal (phone/tablet).
  • API tests now use existing org-1 models (never Opus); skips resolved.
  • Failure screenshots on fixture setup errors and in CI artifacts; session refresh verifies login and retries.
  • Stale xfails removed; ledger row 31 added.

Prod results (2026-11-05)

12 of 14 pairwise runs passed end to end. Two General-scope library runs stalled mid-run (audits 910/911) -> ParakhAI-Backend#134.

Notes

  • Matrix runs create real audits; completed ones are kept as evidence. Needs SANDBOX_ORG_SLUG.
  • The 135-case run was not executed.

@saqibmanan saqibmanan closed this Nov 5, 2026
@saqibmanan saqibmanan reopened this Nov 5, 2026
@saqibmanan
saqibmanan merged commit 8f37e3e into main Nov 6, 2026
14 checks passed
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