Classifiers preserve input order and return one row per input. input_id is the
one-based position in that call; retain it when saving and resuming a batch.
status is ok, no_match or failed. An absent lookup membership is not a
failed request. category is the single selected label, or for a scored classifier
the highest-scoring label with its score in confidence; labels preserves every
label in a tibble with category, score and vendor.
Unknown scores are NA, not probabilities inferred from list membership.
A tibble with input identity, provider, evidence, category, confidence, labels, status, stage, error_code, retryable, http_status, model_id, model_revision, source_last_published, source_fetched_at and details.
provider identifies the classifier or list; evidence_type distinguishes page
content, archived content, name inference and list membership. details retains
source-specific records, including the collected page for piedomains results.
List lookups match the normalised host first and then its registrable domain
(public suffix plus one label); details$matched_key records which key matched.
Model and source identifiers are reported where known; missing dates stay missing.