Summarise a cascade run as a report — np_run

Assembles a human-readable report from an np_cascade() result: provenance (version, config, inputs), the matching process (per-pass cascade table, how many candidates were generated, how matches were formed), the outcome (YES / MAYBE / NO with margins and vetoes), operations (timings, files written), and a pointer to the review hand-off for the MAYBE queue.

Usage

np_run_report(
  res,
  inputs = NULL,
  timings = NULL,
  outputs = NULL,
  reference = NULL,
  file = NULL,
  title = "npmatch run report"
)

Arguments

res: An np_tiered result from np_cascade() (carrying its stages, pairs, and config attributes).
inputs: Optional data frame or named list describing the source files used (e.g. columns file, vintage, rows, md5). Rendered verbatim.
timings: Optional named numeric vector of stage durations in seconds (e.g. c(normalize = 610, cascade = 2900)).
outputs: Optional character vector of output file paths written.
reference: Label for the reference used, e.g. "unified (active+inactive)".
file: Optional path to write the Markdown report to.
title: Report title.

Value

The Markdown report as a length-one character string (invisibly if file is given). A structured list of the computed pieces is attached as attr(., "parts").

Details

Most of the content is read straight off the result object — attr(res, "stages") (the per-pass summary) and attr(res, "pairs") (the scored candidate union) — so it works on any cascade result without a labelled truth set. Supply inputs, timings, and outputs to record provenance and operational context that the result object cannot know.