Recover unresolved grants against a grant-side dictionary

Runs the cascade over the rows a BMF run could not clear, against a dictionary from pf_grant_dictionary(), and applies an acceptance rule.

Usage

pf_recover_from_dictionary(
  status,
  dict,
  recipients,
  accept = c("agree", "all", "none"),
  passes = c(pf_staged_passes("lean", concat = TRUE), pf_recovery_passes(state = TRUE)),
  verbose = TRUE
)

Arguments

status

Status frame from pf_run_batches() or pf_match_grants(), carrying grant_id, status and ein.

dict

A bundle from pf_grant_dictionary().

recipients

The prepared recipient frame the run was matched from.

accept

"agree", "all" or "none".

passes

Pass list for the dictionary run.

verbose

Print the recovery counts.

Value

status with dict_status, dict_ein, dict_score and dict_agrees added, status and ein updated on accepted rows, and recovered flagging them. The non-accepted dictionary YES rows are on the queue attribute; attr(, "counts") carries the tally.

The acceptance rule

accept = "agree" (the default) promotes a row only when the dictionary's EIN is the one the BMF run already held as its own top candidate. Those were recoverable all along and the margin gate would not release them — 40.9% of the recoveries measured on the benchmark. The rest are a new claim from a reference whose labels this pipeline produced, so they go to the queue attribute rather than into the status frame. accept = "all" promotes every dictionary YES and is for measurement, not production; accept = "none" annotates and promotes nothing.

See also

pf_grant_dictionary()