Skip to contents

Wraps pf_bmf_context(), pf_candidate_frame(), and optionally pf_purpose_tiebreak(). Produces the sheet a human or an LLM reviews.

Usage

pf_adjudicate(
  x,
  bmf,
  ids = NULL,
  k = 5,
  tiebreak = FALSE,
  org_text = NULL,
  verbose = TRUE
)

Arguments

x

A pf_match (from pf_match_grants()) or a pf_linked batch run.

bmf

Processed BMF, for candidate and foundation context.

ids

Grant ids to include. Defaults to every MAYBE.

k

Candidates per grant.

tiebreak

FALSE, TRUE for defaults, or a pf_tiebreak_config().

org_text

Output of pf_build_org_text(); required when tiebreak is not FALSE.

verbose

Print progress.

Value

A candidate-level data frame, one row per (grant, candidate).

Details

The tiebreak is off by default. It gains 1 grant and loses 0 of the 6 changes we can check against adjudicated truth, out of 75 it makes on a 5,000-grant run — real in direction, unestablished in magnitude. See dev/PURPOSE-MATCH-FINDINGS.md §10 before switching it on.