analytical-method-validation

Run plan_validation.py with the framework, measured attribute, and technique to get the required characteristics and a protocol skeleton. Q2(R2) Table 1 derives requirements from the attribute (assay, impurity, identity), and acceptance criteria must be stated before data collection since the guideline supplies almost no numeric limits.

The coefficient of determination rises with range and is insensitive to curvature, so a curve with r² of 0.983 can still show 9% back-calculated error at the low end. The check_response.py script uses the lack-of-fit F test against pure error and a residual runs test, which is what Q2(R2) 3.2.2.1 actually requires.

Chromatographic assays use ±15% accuracy and precision tolerances (±20% at LLOQ), while ligand binding assays use ±20% (±25% at LLOQ and ULOQ) plus a total-error criterion of 30% (40% at limits). The check_bioanalytical_run.py script requires an explicit --modality flag because applying the wrong set is a common finding.

Use compare_methods.py with a pre-stated equivalence margin; it runs TOST on paired differences and reports Deming and Passing-Bablok regression slopes. A non-significant t test is not evidence of equivalence, and ordinary least squares biases the slope because both procedures carry measurement error.

No, all scripts use only the Python 3.11+ standard library. Statistical distributions (t, chi-square, F) are computed from first principles via the regularized incomplete beta and gamma functions, so results are reproducible in any conforming interpreter without network access.

No, the scripts report statistics and flag findings against stated criteria but never conclude that a procedure is validated or accept a run. That judgment remains with the analyst, technical reviewer, quality unit, and regulator; exit codes (0, 1, 2) only gate workflows.