@heripo/model 0.1.7 → 0.1.8

package/dist/index.cjs.map

@@ -1 +1 @@
-{"version":3,"sources":["../src/index.ts"],"sourcesContent":["export type * from './docling-document';\nexport type * from './hanja-assessment';\nexport type * from './processed-document';\nexport type * from './token-usage-report';\nexport type * from './document-process-result';\n"],"mappings":";;;;;;;;;;;;;;;;AAAA;AAAA;","names":[]}
+{"version":3,"sources":["../src/index.ts"],"sourcesContent":["export type * from './docling-document';\nexport type * from './processed-document';\nexport type * from './token-usage-report';\nexport type * from './document-process-result';\nexport type * from './ocr-strategy';\n"],"mappings":";;;;;;;;;;;;;;;;AAAA;AAAA;","names":[]}

package/dist/index.d.cts

@@ -109,43 +109,6 @@ interface DoclingDocument {
     pages: Record<string, DoclingPage>;
 }
 
-/**
- * Result of Hanja (KCJ) quality assessment
- *
- * Evaluates OCR quality of Korean-Chinese-Japanese (KCJ/KCJ) characters
- * in the document by sampling pages and comparing with Vision LLM.
- */
-interface HanjaAssessment {
-    /**
-     * Whether the document should be re-parsed using VLM pipeline
-     * due to significant KCJ character corruption
-     */
-    needsVlmReparse: boolean;
-    /**
-     * Severity of KCJ character corruption
-     * - 'none': No KCJ characters found or no corruption detected
-     * - 'minor': Some corruption but still usable
-     * - 'severe': Significant corruption requiring VLM re-parse
-     */
-    severity: 'none' | 'minor' | 'severe';
-    /**
-     * Total number of text pages considered as candidates for assessment
-     */
-    kcjPageCount: number;
-    /**
-     * Number of pages actually sampled for quality assessment
-     */
-    sampledPageCount: number;
-    /**
-     * Ratio of corrupted characters (0.0 ~ 1.0)
-     */
-    corruptedRatio: number;
-    /**
-     * Human-readable reason for the assessment result
-     */
-    reason: string;
-}
-
 /**
  * Caption information
  *
@@ -734,4 +697,26 @@ interface DocumentProcessResult {
     usage: TokenUsageReport;
 }
 
-export type { Caption, Chapter, ComponentUsageReport, DoclingBBox, DoclingBaseNode, DoclingBody, DoclingDocument, DoclingGroupItem, DoclingOrigin, DoclingPage, DoclingPageImage, DoclingPictureItem, DoclingProv, DoclingReference, DoclingTableCell, DoclingTableData, DoclingTableItem, DoclingTextItem, DocumentProcessResult, HanjaAssessment, ModelUsageDetail, PageRange, PhaseUsageReport, ProcessedDocument, ProcessedFootnote, ProcessedImage, ProcessedTable, ProcessedTableCell, TextBlock, TokenUsageReport, TokenUsageSummary };
+/**
+ * Result of the OCR strategy sampling phase.
+ * Determines whether to use ocrmac (standard Docling pipeline)
+ * or VLM (direct vision language model processing) for a given document.
+ */
+interface OcrStrategy {
+    /** Selected OCR method */
+    method: 'ocrmac' | 'vlm';
+    /** OCR language weights for ocrmac (e.g., ['ko-KR', 'en-US'] or ['zh-Hant', 'ko-KR']) */
+    ocrLanguages?: string[];
+    /** BCP 47 language tags detected during sampling (e.g., ['ko-KR', 'en-US']) */
+    detectedLanguages?: string[];
+    /** Human-readable explanation of the decision */
+    reason: string;
+    /** Number of pages that were sampled for the decision */
+    sampledPages: number;
+    /** Total number of pages in the document */
+    totalPages: number;
+    /** 1-based page numbers where Korean-Hanja mixed script was detected in text layer */
+    koreanHanjaMixPages?: number[];
+}
+
+export type { Caption, Chapter, ComponentUsageReport, DoclingBBox, DoclingBaseNode, DoclingBody, DoclingDocument, DoclingGroupItem, DoclingOrigin, DoclingPage, DoclingPageImage, DoclingPictureItem, DoclingProv, DoclingReference, DoclingTableCell, DoclingTableData, DoclingTableItem, DoclingTextItem, DocumentProcessResult, ModelUsageDetail, OcrStrategy, PageRange, PhaseUsageReport, ProcessedDocument, ProcessedFootnote, ProcessedImage, ProcessedTable, ProcessedTableCell, TextBlock, TokenUsageReport, TokenUsageSummary };

package/dist/index.d.ts

@@ -109,43 +109,6 @@ interface DoclingDocument {
     pages: Record<string, DoclingPage>;
 }
 
-/**
- * Result of Hanja (KCJ) quality assessment
- *
- * Evaluates OCR quality of Korean-Chinese-Japanese (KCJ/KCJ) characters
- * in the document by sampling pages and comparing with Vision LLM.
- */
-interface HanjaAssessment {
-    /**
-     * Whether the document should be re-parsed using VLM pipeline
-     * due to significant KCJ character corruption
-     */
-    needsVlmReparse: boolean;
-    /**
-     * Severity of KCJ character corruption
-     * - 'none': No KCJ characters found or no corruption detected
-     * - 'minor': Some corruption but still usable
-     * - 'severe': Significant corruption requiring VLM re-parse
-     */
-    severity: 'none' | 'minor' | 'severe';
-    /**
-     * Total number of text pages considered as candidates for assessment
-     */
-    kcjPageCount: number;
-    /**
-     * Number of pages actually sampled for quality assessment
-     */
-    sampledPageCount: number;
-    /**
-     * Ratio of corrupted characters (0.0 ~ 1.0)
-     */
-    corruptedRatio: number;
-    /**
-     * Human-readable reason for the assessment result
-     */
-    reason: string;
-}
-
 /**
  * Caption information
  *
@@ -734,4 +697,26 @@ interface DocumentProcessResult {
     usage: TokenUsageReport;
 }
 
-export type { Caption, Chapter, ComponentUsageReport, DoclingBBox, DoclingBaseNode, DoclingBody, DoclingDocument, DoclingGroupItem, DoclingOrigin, DoclingPage, DoclingPageImage, DoclingPictureItem, DoclingProv, DoclingReference, DoclingTableCell, DoclingTableData, DoclingTableItem, DoclingTextItem, DocumentProcessResult, HanjaAssessment, ModelUsageDetail, PageRange, PhaseUsageReport, ProcessedDocument, ProcessedFootnote, ProcessedImage, ProcessedTable, ProcessedTableCell, TextBlock, TokenUsageReport, TokenUsageSummary };
+/**
+ * Result of the OCR strategy sampling phase.
+ * Determines whether to use ocrmac (standard Docling pipeline)
+ * or VLM (direct vision language model processing) for a given document.
+ */
+interface OcrStrategy {
+    /** Selected OCR method */
+    method: 'ocrmac' | 'vlm';
+    /** OCR language weights for ocrmac (e.g., ['ko-KR', 'en-US'] or ['zh-Hant', 'ko-KR']) */
+    ocrLanguages?: string[];
+    /** BCP 47 language tags detected during sampling (e.g., ['ko-KR', 'en-US']) */
+    detectedLanguages?: string[];
+    /** Human-readable explanation of the decision */
+    reason: string;
+    /** Number of pages that were sampled for the decision */
+    sampledPages: number;
+    /** Total number of pages in the document */
+    totalPages: number;
+    /** 1-based page numbers where Korean-Hanja mixed script was detected in text layer */
+    koreanHanjaMixPages?: number[];
+}
+
+export type { Caption, Chapter, ComponentUsageReport, DoclingBBox, DoclingBaseNode, DoclingBody, DoclingDocument, DoclingGroupItem, DoclingOrigin, DoclingPage, DoclingPageImage, DoclingPictureItem, DoclingProv, DoclingReference, DoclingTableCell, DoclingTableData, DoclingTableItem, DoclingTextItem, DocumentProcessResult, ModelUsageDetail, OcrStrategy, PageRange, PhaseUsageReport, ProcessedDocument, ProcessedFootnote, ProcessedImage, ProcessedTable, ProcessedTableCell, TextBlock, TokenUsageReport, TokenUsageSummary };

package/package.json

@@ -2,7 +2,7 @@
   "name": "@heripo/model",
   "private": false,
   "type": "module",
-  "version": "0.1.7",
+  "version": "0.1.8",
   "description": "Document models and type definitions for heripo engine",
   "main": "dist/index.cjs",
   "module": "dist/index.js",