scripts/test_evals_lift_report_schema.py

• schema · function · L21-L22 — def schema()
• validator · function · L26-L28 — def validator(schema)
• _valid_report · function · L31-L58 — def _valid_report()
• test_valid_report_passes · function · L61-L62 — def test_valid_report_passes(validator)
• test_is_valid_draft_2020_12_schema · function · L65-L68 — def test_is_valid_draft_2020_12_schema(schema): # Must not raise.
• test_rejects_missing_required_field · function · L71-L75 — def test_rejects_missing_required_field(validator)
• test_rejects_empty_caveats · function · L78-L81 — def test_rejects_empty_caveats(validator)
• test_rejects_additional_top_level_property · function · L84-L87 — def test_rejects_additional_top_level_property(validator)
• test_rejects_per_task_missing_aggregate_metric · function · L90-L93 — def test_rejects_per_task_missing_aggregate_metric(validator)
• test_trivial_accept_all_mutation_proves_constraints_load_bearing · function · L96-L108 — def test_trivial_accept_all_mutation_proves_constraints_load_bearing(): # Swap the real schema for {} (accept-all). A known-bad report (empty # caveats + missing run_id) THEN validates, proving the real schema's # required/minItems constraints are doing the work.