Specify a validation programme for a process-IRT model — irt_validation

Creates an evidence contract rather than executing a particular estimator. The object can be passed to stress-test and evidence-grading helpers.

Usage

irt_validation_spec(
  model_id,
  replications = 250L,
  parameters = NULL,
  metrics = c("bias", "rmse", "coverage", "interval_width", "convergence"),
  grouped_validation = c("device", "session", "site"),
  preprocessing_variants = NULL,
  misspecification_scenarios = NULL,
  thresholds = list(max_abs_bias = 0.1, max_rmse = 0.3, min_coverage = 0.9,
    max_failure_rate = 0.05, min_external_folds = 2L),
  seed = 20260808L,
  notes = NULL
)

Arguments

model_id: Stable model identifier.
replications: Number of simulation replications planned.
parameters: Parameter families expected to be recovered.
metrics: Recovery metrics to require.
grouped_validation: Grouping variables for transport validation.
preprocessing_variants: Optional named preprocessing variants.
misspecification_scenarios: Optional named scenarios.
thresholds: Named evidence thresholds.
seed: Reproducibility seed.
notes: Free-text scientific notes.