Context
report.py hardcodes the effect column as _COL_EFFECT = "formality" (see _SWEEP_COLUMNS and _read_sweep_rows). Any sweep whose effect column is sentiment or verbosity — both shipped concepts — cannot be reported without editing the module. The effect column should be selectable.
Proposal
Thread an effect_column parameter through the CSV readers and build_report, defaulting to "formality" so current behaviour is unchanged, and expose a --effect-col CLI flag on steer-report.
Acceptance criteria
- Replace the hardcoded
_COL_EFFECT use in _read_sweep_rows with a parameter; load_dose_curve, load_layer_curve, and build_report accept effect_column: str = "formality".
- Add
--effect-col to steer-report (default formality), passed through to build_report.
- A clear
ValueError naming the available columns when the requested effect column is absent from the CSV (reuse the existing missing-columns check path).
- CPU-only unit test: a CSV with a
sentiment effect column reports correctly when effect_column="sentiment", and a missing column raises the named error. No model download.
ruff check ., ruff format --check ., mypy src, pytest -q all pass.
Good first issue: the change is bounded to two functions plus one CLI flag, with existing patterns to copy (the --json flag and the missing-columns check).
Context
report.pyhardcodes the effect column as_COL_EFFECT = "formality"(see_SWEEP_COLUMNSand_read_sweep_rows). Any sweep whose effect column issentimentorverbosity— both shipped concepts — cannot be reported without editing the module. The effect column should be selectable.Proposal
Thread an
effect_columnparameter through the CSV readers andbuild_report, defaulting to"formality"so current behaviour is unchanged, and expose a--effect-colCLI flag onsteer-report.Acceptance criteria
_COL_EFFECTuse in_read_sweep_rowswith a parameter;load_dose_curve,load_layer_curve, andbuild_reportaccepteffect_column: str = "formality".--effect-coltosteer-report(defaultformality), passed through tobuild_report.ValueErrornaming the available columns when the requested effect column is absent from the CSV (reuse the existing missing-columns check path).sentimenteffect column reports correctly wheneffect_column="sentiment", and a missing column raises the named error. No model download.ruff check .,ruff format --check .,mypy src,pytest -qall pass.Good first issue: the change is bounded to two functions plus one CLI flag, with existing patterns to copy (the
--jsonflag and the missing-columns check).