[linyuchao123]

Why are these changes needed?

Two tasks with the same ID can silently hide a regression. With baseline scores [True, True] and current scores [False, True], the overall pass rate drops to 50%, but RunResult.diff() maps tasks by ID and keeps only the last task, reporting 100% to 100% with no regressions.

Validate non-empty string IDs and uniqueness in Suite and RunResult. This also rejects collisions between explicit IDs and generated IDs, and ambiguous historical JSON runs loaded through load_run. Missing/null IDs still receive the existing sequential defaults; explicit IDs are never silently renamed. Errors identify the source, duplicated ID, and task indices.

Validation:

• New regression tests against the original implementation: 12 failed, 48 passed.
• With the fix: python -m pytest test/eval -q --disable-warnings: 332 passed, 6 skipped (optional dependencies).
• Ruff check/format and git diff --check passed for the changed files.
• No real model calls were made.

Related issue number

No directly corresponding open issue found. This is separate from #3401, which addresses unpaired scorer denominators rather than duplicate task identity.

Checks

• I've included any doc changes needed for https://docs.ag2.ai/.
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed.

AI assistance

AI assistance was used for investigation, implementation, regression tests, and this description. Local validation and comparison with the original implementation are recorded above. The human-ownership attestations below are left for the contributor to complete after personal review.

• I understand the changes in this PR and can explain them in my own words.
• I have verified that the PR description accurately reflects the actual diff.
• If AI assistance was used, I reviewed, tested, and validated the generated code/text before submitting.

[CLAassistant]

All committers have signed the CLA.