[maximusunc]

This switches the harness and the Information Radiator to translator-testing-model 0.6.0 (TranslatorSRI/TranslatorTestingModel#43). CI will fail at install until translator-testing-model==0.6.0 is on PyPI.

Match an expected output on any of its CURIEs

Test assets name their expected output with output_ids, a list.

• Pass/fail: each CURIE is judged exactly as a single CURIE was before, then combined. TopAnswer and Acceptable pass if any CURIE passes; BadButForgivable and NeverShow fail if any CURIE fails.
• Rank and score: reported from the best-ranked match. Before, a CURIE bound by several results reported the last one.
• Split answers: each agent's result records found_curies, the CURIEs that came back, best first. More than one means Translator split the answer.
• Normalization: every CURIE is normalized, and duplicates are counted once.
• Radiator schema (radiator_schema version 2): AssetResult.output_curie becomes output_curies, and AgentResult gains found_curies. Version 1 records, such as saved runs, are still read. Migration 0007 converts existing rows.
• Zebrunner: the OutputCurie label comma-separates several CURIEs.

--env: the run says which environment it tests

Suites are the same in every environment, and 0.6.0 drops test_env from test cases.

• test-harness now requires --env {dev,ci,test,prod}. It drives the services queried, the node normalizer, the TRAPI version check, CI's cache bypass, and the environment Zebrunner and the radiator record.
• test-harness-cycle and test-harness-sweep take --env and pass it to every run.
• Helm: a cycle entry tests the environment it's named for (or its env), and every entry defaults to mvp_tests, pathfinder_tests and performance_tests. The sweep needs sweep.env.
• run-step.sh keeps --env. Compose has a HARNESS_ENV setting.

The radiator: one history per shared suite

Runs recorded under the old per-environment copies are filed under the shared suite, keeping the old name in runs.source_suite:

Old names	Shared suite
sprint_6_tests, dev_refactor_tests, test_integration, prod_integration	mvp_tests
pathfinder_tests_TEST, pathfinder_tests_PROD	pathfinder_tests

A history line is still suite, env, target and query type, so each environment's history, trends and diffs carry on from its own old copy. Migration 0008 renames the runs already stored, and ingest renames any that arrive later, such as a run saved before the switch. A run's page says what it was recorded as.

Testing

• Harness: 199 passing, including new tests for each multi-CURIE rule and for --env. A 20,000-case comparison against the old analysis found no pass/fail difference for single-CURIE assets.
• Radiator: 102 passing against Postgres, and alembic check is clean. Migrations 0007 and 0008 were tested up and down on sample rows.
• Not run:
  • A live run against Translator: the sandbox I worked in blocks the SmartAPI registry.
  • A render of the Helm charts: CI's helm-lint job will do that.

Rollout

1. Release translator-testing-model 0.6.0.
2. Deploy the radiator; migrations 0007 and 0008 run on startup.
3. In one gap between cycles, deploy this harness, with each cycle and sweep config passing its environment, and merge the NCATSTranslator/Tests migration.

---

Generated by Claude Code