Records a calibration on a run and folds its manifest_patch into the run's
design, so the run now carries its validated inference and its study
agent_manifest() hash changes (the apparatus is no longer a bare
model-conditioned run; it is a calibrated estimate). The run's claim_type
is set to "calibrated_inference", which suppresses the model-conditioned
caveat in report().
Arguments
- run
An object accepted by
as_agent_run().- cal
An
agent_calibrationfromagent_calibrate().
Examples
if (FALSE) { # \dontrun{
run <- as_agent_run(a)
cal <- agent_calibrate(preds, gold = g, method = "ppi", estimand = "proportion")
run <- attach_calibration(run, cal)
} # }