Spaces:
Configuration error
Configuration error
File size: 1,496 Bytes
f1ba785 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 | """Conservative field-specific accept/retry/abstain policy."""
from .contracts import RankedCandidate, SemanticClass
from .reliability import AGREEMENT_CLASSES, CHECKSUM_CLASSES, MIN_RELIABILITY
BEAM_MARGIN_FLOOR = 0.05
RECOVERY_MARK = "nid_recovery"
# Character floors stay as a second screen after the reliability bound.
# They are not calibration.
THRESHOLDS: dict[SemanticClass, float] = {
"nid": 97.0,
"pan": 97.0,
"registration_number": 94.0,
"tax_id": 94.0,
"account_number": 94.0,
"date": 92.0,
"amount": 92.0,
"percentage": 90.0,
"generic_numeric": 92.0,
}
def decide(
candidate: RankedCandidate | None,
semantic_class: SemanticClass,
*,
agreement: bool = False,
reliability_lower_bound: float = 0.0,
) -> str:
if candidate is None or not candidate.schema_valid or candidate.checksum_valid is False:
return "abstain"
if candidate.beam_margin < BEAM_MARGIN_FLOOR:
return "retry"
if RECOVERY_MARK in candidate.preprocessing_variant:
return "retry"
if reliability_lower_bound < MIN_RELIABILITY:
return "retry"
if candidate.min_character_confidence < THRESHOLDS[semantic_class]:
return "retry"
if semantic_class in CHECKSUM_CLASSES:
return "accepted" if candidate.checksum_valid is True else "retry"
if semantic_class in AGREEMENT_CLASSES:
return "accepted" if agreement else "retry"
return "retry"
|