dr_quality_rule()

Usage

dr_quality_rule(
  name = NULL,
  check = NULL,
  description = "",
  engine = c("native", "pointblank"),
  action = NULL,
  threshold = NULL,
  dimension = NULL,
  volatile = FALSE
)

dr_quality_counts(n_failed, n_total)

dr_pointblank_checks(
  name,
  build,
  policy = c("rule", "agent"),
  action = NULL,
  threshold = NULL,
  volatile = FALSE
)

Arguments

name

Rule name, or a one-sided formula as a shortcut. When omitted, formula rules use their expression as a label. Explicit names stay unchanged.

check

Function taking a table and returning a logical vector or dr_quality_counts(), or a one-sided row predicate such as ~ amount >= 0. A scalar function result is one aggregate test; a longer vector must have one value per row. Formula results are evaluated per row, including scalar predicates repeated for each row. Missing logical values count as failures. Lazy formulas run on the backend; arbitrary functions remain responsible for their own collection.

description

Rule description.

engine

Formula evaluation engine: "native" (default) or optional "pointblank". Both require logical row predicates and count missing values as failures. Pointblank interrogates a real agent against the normalized predicate, retaining its reports and check evidence. Ordinary functions use the native engine; use dr_pointblank_checks() for custom agents.

action

Optional publication policy: block, warn or quarantine. Quarantine removes every failing row before writing and retains it locally on the result. It supports native row formulas only. Standalone validation still blocks rejected rows until an execution path actually removes them.

threshold

Permitted fraction of failed test units, from zero to one. For quarantine, every rejected row is removed regardless of threshold.

dimension

Optional ODCS quality dimension.

volatile

Explicitly allow time-dependent or random checks. Such checks are labelled volatile, cannot authorize attestations and cannot use release caching.

n_failed, n_total

Failed and total test units.

build

Function creating a pointblank agent from a lazy table.

policy

"rule" preserves the explicit action / threshold gate. "agent" uses pointblank's per-step action levels: warnings permit publication, stop/error and critical states block. Native pointblank threshold rounding applies. An unconfigured, inactive or errored agent step always blocks. Old notify-only levels do not authorize publication.

Value

Examples

rule <- dr_quality_rule("positive", function(data) {
  counts <- dplyr::summarise(data, failed = sum(amount <= 0), total = dplyr::n())
  if (inherits(counts, "tbl_sql")) counts <- dplyr::collect(counts)
  dr_quality_counts(counts$failed, counts$total)
})
rule$name