Build a banked test case from a column

Profiles a column, computes its signature and candidate shortlist, and writes a self-describing JSON test case (values, profile, label, difficulty) to the bank directory. These cases are the replayable substrate for router training and regression testing.

Usage

fx_make_test_case(
  x,
  case_number,
  data_type = NA_character_,
  semantic_family = NA_character_,
  semantic_type = NA_character_,
  semantic_type_id = NA_character_,
  variant_id = NA_character_,
  depth_labeled = NA_integer_,
  label_confidence = NA_character_,
  label_method = "unlabeled",
  reviewer = NA_character_,
  relabeled = FALSE,
  relabel_reason = NA_character_,
  label_notes = NA_character_,
  difficulty = NA_character_,
  difficulty_rationale = NA_character_,
  noise_markers = character(0),
  corpus = "manual",
  corpus_version = NA_character_,
  original_header = NA_character_,
  original_label = NA_character_,
  table_id = NA_character_,
  parent_case_id = NA_character_,
  ontology = NULL,
  ontology_version = NA_character_,
  ontology_status = "in_ontology",
  proposed_parent = NA_character_,
  detector_exists = TRUE,
  hard_negative_for = character(0),
  order_is_signal = FALSE,
  mutations_allowed = character(0),
  mutations_forbidden = character(0),
  mutation_target = NA_character_,
  outdir = ".",
  n_candidates = 8
)

Arguments

x

A column (coerced to character).

case_number

Integer index used to mint the case id.

data_type, semantic_family, semantic_type, semantic_type_id, variant_id, depth_labeled

Ontology label fields (a later labelling pass may overwrite them).

label_confidence, label_method, reviewer, relabeled, relabel_reason, label_notes

Labelling provenance for the case's ground truth.

difficulty, difficulty_rationale, noise_markers

Difficulty annotation (difficulty is one of "easy", "medium", "hard", or NA).

corpus, corpus_version, original_header, original_label, table_id, parent_case_id

Source-corpus provenance recorded with the case.

ontology, ontology_version, ontology_status, proposed_parent, detector_exists, hard_negative_for

Ontology binding: the ontology to validate the label against, its version and status, and whether a detector exists for the labelled type.

order_is_signal, mutations_allowed, mutations_forbidden, mutation_target

Mutation contract governing how the case may be perturbed for training.

outdir

Directory the JSON case is written to.

n_candidates

Number of signature-distance candidates to store.

Value

Invisibly, a list with the written path and the case obj.

See also

fx_index_bank(), fx_bank_coverage()