Skip to contents

Render a review frame as a static HTML report

Source: R/review_report.R
np_review_report.Rd

Turns the review queue from np_route() into a self-contained HTML report for human review: one section per source entity (UEI), each candidate laid out as a side-by-side "matching form" (id, name, nonprofit type, age, size, affiliate, address, point of contact, and the match-score block). Candidates are ordered with the pipeline's selected pick first, then by descending total_score.

Usage

np_review_report(
  x,
  output = "match-review-report.html",
  decision = "MAYBE",
  max_ueis = 0,
  only_missing_gt = FALSE,
  seed = 42,
  quiet = TRUE
)

Arguments

x

One of: an np_routing object from np_route() (its review element is used); an np_review frame or plain data frame in the review schema; or a path to a CSV of one.

output

Path for the HTML file to write. Default "match-review-report.html" in the working directory. A .html extension is added if missing.

decision

Which match_decision tiers to include: "ALL", "YES", "NO", or "MAYBE" (comma-separate to combine, e.g. "MAYBE,NO"). Default "MAYBE" — the human-review hand-off queue.

max_ueis

Cap on the number of UEI sections rendered; 0 or NA means no cap. When the pool is larger than max_ueis, a deterministic random sample (see seed) is drawn. Default 0.

only_missing_gt

Logical; when the frame carries ground-truth columns, restrict to UEIs whose ground-truth EIN is not among the candidates (recall misses). Ignored on frames without ground truth. Default FALSE.

seed

Sampling seed used when max_ueis truncates the pool. Default 42.

quiet

Suppress Quarto's rendering progress output. Default TRUE.

Value

The path to the written HTML file, invisibly.

Details

The report is read-only. If the frame also carries evaluation columns (is_gt_ein, gt_confidence, gt_outcome_class, labeled_by) — as produced by the benchmarking harness — the ground-truth candidate is highlighted and floated to the top of its section, and UEIs whose true EIN is missing from the candidate set are flagged. On an ordinary np_route() frame those columns are simply absent and the report renders without them.

Rendering requires the quarto package and a Quarto CLI installation (the same dependency the vignettes use). The bundled template lives at system.file("report", "match-review-report.qmd", package = "npmatch").

See also

np_route() for producing the review frame; np_run_report() for a Markdown run summary.

Examples

if (FALSE) { # \dontrun{
res     <- np_cascade(query, reference)
routing <- np_route(res, bmf = bmf_raw, sam = sam_raw)
np_review_report(routing, output = "review.html")

# just the recall misses on a labelled evaluation frame
np_review_report("EVAL-FRAME-MERGED.csv", decision = "ALL",
                 only_missing_gt = TRUE, output = "misses.html")
} # }