Initial CRAN release.
fit_mixed_subjects_mml() and relatives). The estimator is anchored to the
human data and is asymptotically unbiased for the human item parameters at any
tuning weight.tune_lambda_ability_risk()), which
selects the tuning weight by direct 1-D optimization of propagated
ability-recovery risk (pass method = "grid" to scan a grid instead). Also
included: a theoretical PPI++ score diagnostic (tune_lambda_ppi_score()),
cross-fitted tuning (tune_lambda_ability_risk_crossfit(), the recommended
workflow for reported analyses), and experimental per-item tuning
(tune_lambda_ability_risk_item()). All non-experimental tuners use the
marginal-MML estimator by default; the frozen expected-count estimator remains
available via fit_fn but is discouraged.vcov() S3 method
(vcov_mixed_subjects_mml()), with ability scoring and item-parameter
uncertainty propagation (score_theta(), ability_risk()).R-CMD-check GitHub Actions workflow.predicted and generated data must be binary 0/1 responses in
the high-level fitting and PPI-score functions; the low-level quadrature
utilities accept fractional input.