dimension,criterion,points,full_credit,partial_credit,zero_credit
Numeric accuracy,Exact/near-exact output,30,All required numeric fields are within benchmark tolerance.,Minor rounding differences do not change interpretation.,"Wrong statistic, probability, interval, sign, tail, or method."
Method specification,Variant disclosed,12,"Names Welch/pooled, Yates/no correction, sample/population SD, tail, CI method, and other relevant choices.",Some choices are inferable but not explicit.,Default variant is hidden or mismatched.
Invalid input handling,Unsafe inputs rejected,12,"Rejects impossible probabilities, negative counts, mismatched vectors, insufficient n, and undefined variance cases.",Rejects most but not all invalid cases.,Returns plausible-looking numeric output for invalid input.
Boundary behavior,"Handles zero, one, perfect association, and extreme tails",10,Returns mathematically defined boundary results and clearly flags undefined cases.,Output is usable with minor warnings missing.,"Crashes, silently clips, or reports misleading values."
Transparency,Formula and assumptions visible,10,"Formula, variable definitions, assumptions, and calculation path are visible.",Only a short explanation is provided.,Black-box result.
Precision,Sufficient significant digits,8,Shows enough digits to reproduce decisions and allows copying full precision.,Rounds reasonably but hides full precision.,Rounding changes conclusions or makes verification impossible.
Interpretation,Decision wording is statistically correct,8,"Separates evidence, magnitude, and causation; reports correct tail and null.",Mostly correct but incomplete.,"Misinterprets p-values, confidence, association, or causality."
Reproducibility,Inputs and output can be retained,5,"Shareable URL, export, or copyable input/output is available.",Manual copying is possible.,Result cannot be reproduced.
Accessibility,Keyboard/mobile/readability,3,Usable by keyboard and mobile with readable labels.,Minor accessibility friction.,Core operation inaccessible.
Versioning,Method/version/date identified,2,Calculator version or reviewed date is visible.,Site update date only.,No versioning signal.