Build the Tier-0 data integrity dashboard markdown for the report

Renders a markdown card that summarizes the run's anti-fabrication verification work: how many AI-attributed verbatim claims were checked, how many verified exactly vs fuzzy (with method breakdown), how many fabrications were dropped, and a relative path link to the fabrication_log.csv when fabrications occurred.

Usage

.build_tier0_dashboard(
  stats,
  fabrication_log_relpath = "fabrication_log.csv",
  config = NULL,
  fabrication_log_path = NULL,
  n_fabricated_caught = NULL
)

Arguments

stats

Named list returned by compute_quote_provenance_stats.

fabrication_log_relpath

Optional relative path to fabrication_log.csv (relative to the report HTML's directory). When NULL or no fabrications occurred, no link is rendered.

config

ThematicConfig object (or NULL) used for the Citations API bypass footnote.

fabrication_log_path

absolute path to fabrication_log.csv. When supplied, the dashboard counts pre-rejection fabrications from this file (the surviving population in stats is post-rejection, so it always reports 0 caught fabrications by itself). Pass NULL on legacy callers that don't have the path; the dashboard falls back to the surviving-population count.

n_fabricated_caught

explicit count of pre-rejection fabrications (from FabricationLog$state$n_logged). Overrides fabrication_log_path when supplied. Pass NULL to skip.

Value

A character string of markdown content (one card).

Details

This dashboard is the user-visible artifact of T0.1's universal verification contract – it makes the package's anti-fabrication work empirically inspectable from the report itself, addressing the transparency dimension of Jowsey et al. 2025's critique. When no verification was run (pre-T0.1 runs, or runs that skipped coding), the dashboard renders a "Verification not available" notice rather than silently omitting – absence of the badge would be its own integrity signal that should not be sent.