Celes Renata
a72f336ad1
Multi-stage evidence-grounded inference architecture replacing the monolithic 9B model extraction pipeline. CPU-first specialist services handle routine extraction while the 9B vLLM model is preserved for semantic adjudication of ambiguous cases. Key components: - Capability-aware inference gateway (OpenAI-compatible + Ollama) - Endpoint registry with DB migrations and REST API - Sentence-aware document segmenter (property tests) - Deterministic financial parsing with offset integrity - Symbol resolution with ambiguity detection - Specialist service (GLiNER2, dynamic batching, K8s deployment) - Company-specific sentiment (FinBERT, calibration) - Retrieval-based novelty and duplicate detection - Confidence calibration pipeline - Deterministic routing engine (property tests) - 9B adjudication layer with VRAM gating - Stock-specific impact model (features, labels, baseline, trained) - Pipeline orchestrator (state machine, queues, leases, feature flags) - Bounded parallelism (async workers, semaphore, load shedding) - Observability (tracing, metrics, alerts) - Compatibility adapter (v3→v2 golden mapping tests) - Shadow/canary promotion framework - Active learning and fine-tuning pipeline Test results: 1,161 tests pass, ruff lint clean. All 282 spec tasks completed.
86 lines
2.5 KiB
Python
86 lines
2.5 KiB
Python
"""Evidence coverage and unsupported-claim metrics.
|
|
|
|
Computes the proportion of extracted fields that have valid evidence support,
|
|
and identifies unsupported claims for audit and active learning.
|
|
"""
|
|
|
|
from __future__ import annotations
|
|
|
|
from dataclasses import dataclass, field
|
|
|
|
|
|
@dataclass(frozen=True)
|
|
class CoverageMetrics:
|
|
"""Evidence coverage statistics for a document's extraction.
|
|
|
|
Attributes:
|
|
total_fields: Total number of fields requiring evidence support.
|
|
supported_fields: Fields with at least one valid evidence span.
|
|
coverage_rate: Proportion of fields with valid evidence (0.0 to 1.0).
|
|
unsupported_claims: List of field identifiers lacking evidence.
|
|
unsupported_rate: Proportion of fields without valid evidence.
|
|
"""
|
|
|
|
total_fields: int
|
|
supported_fields: int
|
|
coverage_rate: float
|
|
unsupported_claims: list[str]
|
|
unsupported_rate: float
|
|
|
|
|
|
@dataclass
|
|
class FieldEvidence:
|
|
"""A field that requires evidence support."""
|
|
|
|
field_id: str
|
|
field_name: str
|
|
evidence_ids: list[str] = field(default_factory=list)
|
|
|
|
|
|
def compute_coverage(
|
|
fields: list[FieldEvidence], verified_span_ids: set[str]
|
|
) -> CoverageMetrics:
|
|
"""Compute evidence coverage metrics for a set of extraction fields.
|
|
|
|
A field is considered "supported" if it references at least one span ID
|
|
that passed offset verification (i.e., is in verified_span_ids).
|
|
|
|
Args:
|
|
fields: List of fields with their linked evidence IDs.
|
|
verified_span_ids: Set of span IDs that passed offset verification.
|
|
|
|
Returns:
|
|
CoverageMetrics with coverage rate and unsupported claims.
|
|
"""
|
|
total = len(fields)
|
|
if total == 0:
|
|
return CoverageMetrics(
|
|
total_fields=0,
|
|
supported_fields=0,
|
|
coverage_rate=1.0,
|
|
unsupported_claims=[],
|
|
unsupported_rate=0.0,
|
|
)
|
|
|
|
supported = 0
|
|
unsupported: list[str] = []
|
|
|
|
for f in fields:
|
|
# A field is supported if any of its evidence IDs are in the verified set
|
|
has_valid_evidence = any(eid in verified_span_ids for eid in f.evidence_ids)
|
|
if has_valid_evidence:
|
|
supported += 1
|
|
else:
|
|
unsupported.append(f.field_id)
|
|
|
|
coverage_rate = supported / total
|
|
unsupported_rate = 1.0 - coverage_rate
|
|
|
|
return CoverageMetrics(
|
|
total_fields=total,
|
|
supported_fields=supported,
|
|
coverage_rate=coverage_rate,
|
|
unsupported_claims=unsupported,
|
|
unsupported_rate=unsupported_rate,
|
|
)
|