← Về thư mục
- _measured · function · L17-L32 — def _measured(name, metric="accuracy", value=1.0, threshold=0.9, comparison=">=", passed=True, per_class=None)
- _pending · function · L35-L47 — def _pending(name): # Shape mirrors run_evals._pending_result: a placeholder aggregate_metric # with no threshold — the renderer must key off status, not metric absence.
- _render · function · L50-L52 — def _render(tasks)
- test_all_passed_verdict_line · function · L55-L61 — def test_all_passed_verdict_line()
- test_measured_row_renders_metric_value_threshold · function · L64-L66 — def test_measured_row_renders_metric_value_threshold()
- test_pending_row_uses_placeholder_columns · function · L69-L72 — def test_pending_row_uses_placeholder_columns()
- test_aggregate_failure_marks_row_and_verdict · function · L75-L78 — def test_aggregate_failure_marks_row_and_verdict()
- test_per_class_only_failure_still_fails_row · function · L81-L89 — def test_per_class_only_failure_still_fails_row(): # Aggregate passes but a per-class metric fails: the gate blocks (#328), # so the row and the verdict must not read as a clean pass.
- test_lower_is_better_comparison_passes_through · function · L92-L95 — def test_lower_is_better_comparison_passes_through()
- test_measured_rows_sort_before_pending · function · L98-L102 — def test_measured_rows_sort_before_pending()
- test_no_pending_omits_pending_suffix · function · L105-L107 — def test_no_pending_omits_pending_suffix()
- test_raw_json_folded_in_details · function · L110-L115 — def test_raw_json_folded_in_details()
- test_cli_main_writes_markdown · function · L118-L126 — def test_cli_main_writes_markdown(tmp_path, capsys)
- test_cli_main_usage_error · function · L129-L130 — def test_cli_main_usage_error()
- test_failure_signal_agrees_with_threshold_gate · function · L133-L153 — def test_failure_signal_agrees_with_threshold_gate(): # _task_failures deliberately MIRRORS (does not import) the gate's failure # rule — importing would couple display to the gate's flat key format. # This pin makes the mirror loud instead of silent: if a future gated axis # lands in _eval_threshold_gate but not here, this fails in CI rather than # the comment rendering green on a run the gate blocks.
- test_table_cells_escape_pipes_and_line_boundaries · function · L156-L162 — def test_table_cells_escape_pipes_and_line_boundaries(): # task_name / metric come from eval manifests; a
| or any line boundary # (\n, \r, \r\n) must not break the table or spoof extra columns/rows # (codex review P2 + re-review \r gap).