Package index

C - Construct role-task

Document how the LLM is being used and which metric family applies.

role()

Construct role-task mapping (the C step of CRAFT)

R - Report dual-track metrics

Reliability (six methods) paired with validity (six metrics).

reliab()

Inter-rater reliability (R step of CRAFT)

reliab_pairs()

Pairwise reliability across all rater columns

valid()

Validity metrics against a gold standard

dual()

Dual-track metrics: reliability + validity, side by side

A - Assess stability

Cross-prompt and cross-model comparison.

stab()

Stability across prompts or models (A step of CRAFT)

F - Field audit

Surface disagreements and low-confidence agreements; threshold sensitivity.

audit()

Rationale audit (F step of CRAFT)

disagree()

Summarize cases needing adjudication

tau_sens()

Sensitivity of label coverage to the confidence threshold

T - Translate to inference

Design-based corrected regression so misclassification uncertainty propagates.

dsl_fit()

Design-based supervised learning correction (T step of CRAFT)

dsl_cmp()

Compare naive vs. design-based corrected estimates

Reproducibility report

report()

Generate a CRAFT reproducibility report