import type { KnowledgeExtractionRecord } from './types.js';
/**
 * The current extractor generation. Stamped onto every extraction's metadata at
 * write time (`extractorVersion`). Bump this when the extraction pipeline improves
 * so that the retained raw-artifact lake is re-processed: an extraction produced by
 * an older generation is treated as stale-by-version and re-extracted from its
 * stored artifact, even when its prior text was non-empty. This is what turns the
 * retained lake into a compounding asset. Extractions written before versioning
 * carry no stamp and resolve to version 0, so they re-extract once.
 */
export declare const KNOWLEDGE_EXTRACTOR_VERSION = 1;
export declare function readKnowledgeExtractorVersion(metadata: Record<string, unknown>): number;
export declare const KNOWLEDGE_MAX_STRUCTURE_SEARCH_TEXT_CHARS: number;
export declare const KNOWLEDGE_MIN_BINARY_SAMPLE_CHARS = 120;
export declare const KNOWLEDGE_BINARY_SAMPLE_CHARS = 4096;
export declare const KNOWLEDGE_BINARY_EXTENDED_RATIO_THRESHOLD = 0.18;
export declare const KNOWLEDGE_BINARY_USEFUL_RATIO_THRESHOLD = 0.78;
export declare const KNOWLEDGE_BINARY_PUNCTUATION_RATIO_THRESHOLD = 0.42;
export declare const KNOWLEDGE_BINARY_WHITESPACE_RATIO_THRESHOLD = 0.08;
export declare function knowledgeExtractionNeedsRefresh(extraction: KnowledgeExtractionRecord | null, currentExtractorVersion?: number): boolean;
export declare function readKnowledgeSearchText(record: Record<string, unknown>): string | undefined;
export declare function hasUsefulKnowledgeExtractionText(value: string | undefined): boolean;
export declare function looksLikeRawPdfPayload(value: string): boolean;
export declare function looksBinaryLikeText(value: string): boolean;
//# sourceMappingURL=extraction-policy.d.ts.map