Returns a named list of summary-kind descriptors. Each descriptor has:
kind, title, expected_metrics (named list of type strings, informative
not enforced), emitted_by (free text), consumed_by_packs (character).
Details
Evidence coverage
Every summary kind a built-in workflow pack matches on is emitted by at least one shipped executor (or is documented as requiring a user executor). The table below maps the kinds packs most commonly ask for to the built-in node kind that produces them:
| summary kind | produced by built-in kind | notes |
hmc_diagnostics | cmdstanr_fit / brms_fit | HMC sampler diagnostics |
prior_predictive_check | prior_fit / brms_prior_fit | prior-predictive draws |
posterior_predictive_check | ppc | observed y + capped yrep draws |
loo_diagnostics | loo | PSIS-LOO Pareto-k |
loo_pit_calibration | loo_pit | randomized PIT + PIET severity |
sbc_result | sbc | ESS-bounded ranks + valid bins |
comparison_results / stacking_weights | compare | model comparison |
prior_spec | user executor | record the prior |
projpred_selection | user executor | use projpred |
causal_* / pad_* | user executors | supply causal/PAD evidence |
A test (test-summary-vocabulary-coverage.R) asserts every literal
summary_kinds = "<kind>" referenced by a pack source is present here.
New summary kinds used by packs require a vocabulary entry that identifies
a built-in or user executor as the source.