Where: tests/ directory - no test_models.py or test_figures.py exist at all, despite faircode/models.py and faircode/figures.py both having real, untested failure paths.
The gap:
faircode/models.py's build_model with an unrecognized model_name raises a bare KeyError from dict indexing - no test exercises this, and no test verifies hyperparameters/random_state are actually propagated correctly to each of the three model families for a valid name either.faircode/figures.py's plot_strategy_comparison has an explicit if subset.empty: raise ValueError(...) guard that no test triggers, and its main() CLI entry point (argument parsing, the --metric flag, a custom results_dir) is exercised nowhere directly - only indirectly through test_benchmark.py's write_report tests, which never reach the empty-subset or CLI paths.
Why it matters: both are real, reachable code paths (a malformed audit.yaml's model field, or figures.py run standalone via its documented CLI) with zero coverage - a regression in either would ship silently.
Suggested fix: add tests/test_models.py (valid model_name per family + hyperparameter/random_state propagation, and an unknown-name error case) and tests/test_figures.py (the empty-subset ValueError, and a main() smoke test with --metric and a custom output directory).
Where:
tests/directory - notest_models.pyortest_figures.pyexist at all, despitefaircode/models.pyandfaircode/figures.pyboth having real, untested failure paths.The gap:
faircode/models.py'sbuild_modelwith an unrecognizedmodel_nameraises a bareKeyErrorfrom dict indexing - no test exercises this, and no test verifies hyperparameters/random_stateare actually propagated correctly to each of the three model families for a valid name either.faircode/figures.py'splot_strategy_comparisonhas an explicitif subset.empty: raise ValueError(...)guard that no test triggers, and itsmain()CLI entry point (argument parsing, the--metricflag, a customresults_dir) is exercised nowhere directly - only indirectly throughtest_benchmark.py'swrite_reporttests, which never reach the empty-subset or CLI paths.Why it matters: both are real, reachable code paths (a malformed
audit.yaml'smodelfield, orfigures.pyrun standalone via its documented CLI) with zero coverage - a regression in either would ship silently.Suggested fix: add
tests/test_models.py(validmodel_nameper family + hyperparameter/random_statepropagation, and an unknown-name error case) andtests/test_figures.py(the empty-subsetValueError, and amain()smoke test with--metricand a custom output directory).