Skip to content

Regression verification

The broad candidate run used:

uv run pytest tests/boundary/trace tests/property/test_trace.py \
  tests/property/test_trace_query.py tests/boundary/analysis \
  tests/boundary/report tests/boundary/cli tests/boundary/web -q --tb=short

Result: 609 passed, 11 failed in 197.46 seconds. All eleven failures were reproduced at baseline 5e7b06b, using the candidate's Python environment and identical restored local ontology/example inputs. The five frontend-rendering tests initially skipped in the baseline worktree until its Node dependencies were shared with the candidate; all five then reproduced their failures.

The failures concern an outdated purchase-intent anchor key, tests for the removed brief-listing interface, an existing frontend 502 response assertion, zero-byte Parquet inputs in corpus endpoint fixtures, and an anchor embedding model expectation that differs from the stored anchor. No optimization-related failure remained in this run.

Baseline failure cases

  • tests/boundary/cli/test_preflight.py::test_a_pinned_scale_can_actually_be_scored_with
  • tests/boundary/web/test_api.py::test_ontologies_and_briefs_come_from_the_engine_checkout
  • tests/boundary/web/test_interface.py::test_the_first_screen_reads_the_workspace_the_engine_derived
  • tests/boundary/web/test_interface.py::test_lists_arrive_the_way_the_pages_read_them
  • tests/boundary/web/test_interface.py::test_every_brief_the_engine_holds_survives_the_intake_form
  • tests/boundary/web/test_interface.py::test_a_form_with_an_awkward_description_is_still_a_brief_the_engine_reads
  • tests/boundary/web/test_interface.py::test_a_brief_loaded_into_intake_keeps_its_competitors_and_filters
  • tests/boundary/web/test_interface_reads_through_the_api.py::test_a_refusal_is_never_reduced_to_its_status_code
  • tests/boundary/web/test_launch_any_study.py::test_the_corpus_endpoint_says_which_shards_are_cached_and_which_sources_exist
  • tests/boundary/web/test_launch_any_study.py::test_the_corpus_endpoint_returns_no_filesystem_path
  • tests/boundary/web/test_launch_any_study.py::test_the_anchors_endpoint_lists_each_version_with_its_verdict

Follow-up checks

  • 46 passed after the invalid-persona-argument compatibility fix: snapshot boundary checks, shared study analysis, trace summaries and anomaly rules.
  • 59 passed for the schema leaf-rule checks.
  • All forty measured benchmark observations have matching output fingerprints between baseline and candidate. The final-code replay also matches all four cases and retains one partition load per finished world.

This records existing suite failures explicitly rather than changing unrelated product behavior or weakening tests as part of the trace optimization.