Module reliability 

Extraction reliability scoring

Computes a 0.0–1.0 reliability score for domain::ExtractionResult outputs so fallback chains can optimize for data quality, not only fetch success. See the module-level docs for the score interpretation table and the selection policy. Extraction reliability scoring

Computes a 0.0–1.0 reliability score for ExtractionResult outputs so fallback chains can optimize for data quality, not only fetch success.

Score components

The score is the weighted sum of three sub-scores, clamped to [0.0, 1.0]:

Sub-score	Default weight	Source
schema_completeness	0.70	successful_regions / total_regions
transformation_success	0.30	1 − transformation_failure_rate
retry_penalty	0.10	retry_count / max_retries (clamped)

retry_penalty is subtracted from the weighted sum — it represents a quality discount for results that took many retries to produce.

Interpretation bands

The continuous overall score is bucketed into three discrete bands so callers can branch on a single ReliabilityBand value:

Band	Score range	Interpretation
High	0.85..=1.00	Production-ready; trust the result.
Medium	0.50..0.85	Partial; treat as best-effort, retry if needed.
Low	0.00..0.50	Unreliable; prefer an alternative fallback.

Selection policy

The ScoreWeightedSelector helper ranks a list of (name, score) candidates and picks the highest-scoring one. Callers can apply a custom ScoringWeights to tune the importance of each sub-score for their target class.

Example

use stygian_plugin::domain::{ExtractionResult, IdempotencyKey};
use stygian_plugin::reliability::{ReliabilityScorer, ReliabilityBand};

let result = ExtractionResult::new(IdempotencyKey::new());
let score = ReliabilityScorer::new().score_extraction(&result, 0);
assert_eq!(score.band, ReliabilityBand::High); // empty template -> vacuously complete
assert!((score.overall - 1.0).abs() < f32::EPSILON);

Re-exports

pub use score::ReliabilityBand;

pub use score::ReliabilityScore;

pub use scorer::ReliabilityScorer;

pub use scorer::ScoringWeights;

pub use selector::ScoreWeightedSelector;

pub use selector::ScoredCandidate;

Modules

score

ReliabilityScore value type and its discrete interpretation band.

scorer

Reliability scorer: turns an ExtractionResult into a ReliabilityScore.

selector

Score-weighted selection over a list of (name, score) candidates.