[jcschaff]

Problem. POST /verify/omex answered COMPLETED with empty results whenever every simulator's run failed on biosimulations.org. OmexVerifyWorkflow skipped each child that had no run (continue), so neither the run ids nor the reasons reached the caller. A client can't tell that from a success. That's the "100% concordance for a run that compared nothing" case Novrus found in #126.

Reproduction (2026-10-22, anonymous, default five simulators).

• BioModels' own BIOMD0000000012 archive gave workflow compose-api-verification-3214309a-01f8-4a3d-a0a1-0151276e88c4: COMPLETED, with empty sims_run_info and comparison_statistics.
• Running one child by hand (COPASI 4.45.296, run 6ada2269041d31202094ca8f) ended FAILED with an empty error. The reason was only in GET /logs/{id}: "/root/archive.omex is not a valid COMBINE/OMEX archive … Content element BIOMD0000000012-biopax2.owl is invalid. Location is not a file". BioModels' manifests list files their zips don't contain, and BioSimulators validates strictly.

Changes (on top of #136):

• VerifyWorkflowOutput.failed_simulators is a new, additive field: one SimulatorFailure {simulator_version, biosimulations_run_id, error_message} per requested simulator that produced no results, in request order.
• poll_biosim_run_activity reads a failed run's reason from its logs, best effort, using error_from_run_logs: the archive-level exception, otherwise the first SED document's or task's.
• When no requested simulator produced results, the workflow ends FAILED, with workflow_error naming each simulator's reason. feat(verify): link OMEX hash to verification runs and return run metadata #136's write-once ledger then records FAILED instead of COMPLETED. The statistics activity is still scheduled, so workflows already in flight replay deterministically.
• The OpenAPI spec is regenerated.

For #126: VerificationStatus.vue already shows workflow_error on FAILED. It would also help to list failed_simulators, including on a partly successful COMPLETED, with links to /runs/{id}, and to leave failed simulators out of the concordance figures (details in this comment).

Tests.

• tests/biosim_verify/test_failed_simulators.py covers reading the reason from the real failed run's logs (recorded as tests/fixtures/local_data/biosim_failed_run_logs.json), the task-level fallback, failures reported in request order, the FAILED message, and that a partial success isn't a failure.
• Locally, ruff and mypy (biosim_server tests) are clean. tests/biosim_verify tests/biosim_runs tests/api gives 417 passed, 2 failed.
• The two failures, test_main.py::test_omex_verify_and_get_output and test_runs_verify_and_get_output, fail the same way on unmodified main with this machine's .env, and are skipped without it. So they come from the local environment, not this change.

🤖 Generated with Claude Code