Define explicit rules for audit verdicts

Usage

audit_policy(version, rules, rationale)

Arguments

version: Nonempty version identifying the researcher's policy.
rules: Data frame with rule_id, scope (event, group, session), metric, operator (<, <=, >, >=, ==), threshold, verdict (PASS, WARNING, FAIL), min_coverage (0 to 1), and aggregation (all or any). Rows sharing a rule_id are AND conditions. Optional item_id, contrast_id, detail, and person_id columns restrict matched diagnostics; NA means unrestricted. Unrestricted rules exclude person-level metrics.
rationale: Explanation of the policy's normative basis.

Value

A policy object. Rules do not create a composite score. A rule whose conditions fail returns NOT_TRIGGERED; missing or incomplete evidence returns NOT_ASSESSED. These are separate from explicit PASS verdicts.

Examples

audit_policy("example-1", data.frame(
  rule_id = "concentration", scope = "session", metric = "max_speaker_share",
  operator = ">", threshold = .6, verdict = "WARNING", min_coverage = 1,
  aggregation = "any"), rationale = "Illustrative rule, not a validated threshold")
#> $version
#> [1] "example-1"
#> 
#> $rules
#>         rule_id   scope            metric operator threshold verdict
#> 1 concentration session max_speaker_share        >       0.6 WARNING
#>   min_coverage aggregation
#> 1            1         any
#> 
#> $rationale
#> [1] "Illustrative rule, not a validated threshold"
#> 
#> attr(,"class")
#> [1] "deliberation_policy"