@deepcitation/deepcitation-js 1.1.18 → 1.1.19

package/lib/parsing/normalizeCitation.d.ts

@@ -1,4 +1,4 @@
-export declare const removeCitations: (pageText: string, leaveValueBehind?: boolean) => string;
+export declare const removeCitations: (pageText: string, leaveKeySpanBehind?: boolean) => string;
 export declare const removePageNumberMetadata: (pageText: string) => string;
 export declare const removeLineIdMetadata: (pageText: string) => string;
 export declare const getCitationPageNumber: (startPageKey?: string | null) => number | null;

package/lib/parsing/normalizeCitation.js

@@ -1,9 +1,9 @@
-export const removeCitations = (pageText, leaveValueBehind) => {
-    const citationRegex = /<cite\s+(?:fileId|attachmentId)='(\w{0,25})'\s+start_page[\_a-zA-Z]*='page[\_a-zA-Z]*(\d+)_index_(\d+)'\s+full_phrase='((?:[^'\\]|\\.)*)'\s+line(?:_ids|Ids)='([^']+)'(?:\s+(value|reasoning)='((?:[^'\\]|\\.)*)')?\s*\/>/g;
-    return pageText.replace(citationRegex, (match, attachmentId, pageNumber, index, fullPhrase, lineIds, value) => {
+export const removeCitations = (pageText, leaveKeySpanBehind) => {
+    const citationRegex = /<cite\s+(?:fileId|attachmentId)='(\w{0,25})'\s+start_page[\_a-zA-Z]*='page[\_a-zA-Z]*(\d+)_index_(\d+)'\s+full_phrase='((?:[^'\\]|\\.)*)'\s+key_span='((?:[^'\\]|\\.)*)'\s+line(?:_ids|Ids)='([^']+)'(?:\s+(value|reasoning)='((?:[^'\\]|\\.)*)')?\s*\/>/g;
+    return pageText.replace(citationRegex, (match, attachmentId, pageNumber, index, fullPhrase, keySpan, lineIds, value) => {
         //it is still value= so we need to remove the value=
-        if (leaveValueBehind) {
-            return value?.replace(/value=['"]|['"]/g, "") || "";
+        if (leaveKeySpanBehind) {
+            return keySpan?.replace(/key_span=['"]|['"]/g, "") || "";
         }
         else {
             return "";
@@ -45,19 +45,29 @@ const normalizeCitationContent = (input) => {
     // Replace ></cite> with /> for consistency
     normalized = normalized.replace(/><\/cite>/g, "/>");
     const canonicalizeCiteAttributeKey = (key) => {
-        if (key === "fullPhrase" || key === "full_phrase")
+        const lowerKey = key.toLowerCase();
+        if (lowerKey === "fullphrase" || lowerKey === "full_phrase")
             return "full_phrase";
-        if (key === "lineIds" || key === "line_ids")
+        if (lowerKey === "lineids" || lowerKey === "line_ids")
             return "line_ids";
-        if (key === "startPageKey" ||
-            key === "start_pageKey" ||
-            key === "start_page_key")
+        if (lowerKey === "startpagekey" ||
+            lowerKey === "start_pagekey" ||
+            lowerKey === "start_page_key")
             return "start_page_key";
-        if (key === "fileID" || key === "fileId" || key === "file_id" || key === "attachmentId" || key === "attachment_id")
-            return "file_id";
-        if (key === "keySpan" || key === "key_span")
+        if (lowerKey === "fileid" ||
+            lowerKey === "file_id" ||
+            lowerKey === "attachmentid" ||
+            lowerKey === "attachment_id")
+            return "attachment_id";
+        if (lowerKey === "keyspan" || lowerKey === "key_span")
             return "key_span";
-        return key;
+        if (lowerKey === "reasoning" || lowerKey === "value")
+            return lowerKey;
+        if (lowerKey === "timestamps" ||
+            lowerKey === "timestamp" ||
+            lowerKey === "timestamps")
+            return "timestamps";
+        return lowerKey;
     };
     // Helper to decode HTML entities (simple implementation, expand if needed)
     const decodeHtmlEntities = (str) => {
@@ -146,16 +156,16 @@ const normalizeCitationContent = (input) => {
         const startPageKeys = keys.filter((k) => k.startsWith("start_page"));
         const ordered = [];
         // Shared first
-        if (attrs.file_id)
-            ordered.push("file_id");
+        if (attrs.attachment_id)
+            ordered.push("attachment_id");
         if (hasTimestamps) {
-            // AV citations: fileId, full_phrase, timestamps, (optional reasoning/value), then any extras
+            // AV citations: attachment_id, full_phrase, timestamps, (optional reasoning/value), then any extras
             if (attrs.full_phrase)
                 ordered.push("full_phrase");
             ordered.push("timestamps");
         }
         else {
-            // Document citations: fileId, start_page*, full_phrase, key_span, line_ids, (optional reasoning/value), then any extras
+            // Document citations: attachment_id, start_page*, full_phrase, key_span, line_ids, (optional reasoning/value), then any extras
             if (startPageKeys.includes("start_page_key"))
                 ordered.push("start_page_key");
             startPageKeys

package/lib/parsing/parseCitation.js

@@ -94,9 +94,9 @@ export const parseCitation = (fragment, mdAttachmentId, citationCounterRef, isVe
     // 4: full_phrase content (escaped)
     // 5: key_span content (escaped)
     // 6: line_ids content
-    // 6: Optional Key (value|reasoning)
-    // 7: Optional Value content (escaped)
-    const citationRegex = /<cite\s+file(?:_id|Id)='(\w{0,25})'\s+start_page[\_a-zA-Z]*='page[\_a-zA-Z]*(\d+)_index_(\d+)'\s+full_phrase='((?:[^'\\]|\\.)*)'\s+key_span='((?:[^'\\]|\\.)*)'\s+line(?:_ids|Ids)='([^']+)'(?:\s+(value|reasoning)='((?:[^'\\]|\\.)*)')?\s*\/>/g;
+    // 7: Optional Key (value|reasoning)
+    // 8: Optional Value content (escaped)
+    const citationRegex = /<cite\s+(?:attachment_id|attachmentId|file_id|fileId)='(\w{0,25})'\s+start_page[\_a-zA-Z]*='page[\_a-zA-Z]*(\d+)_index_(\d+)'\s+full_phrase='((?:[^'\\]|\\.)*)'\s+key_span='((?:[^'\\]|\\.)*)'\s+line(?:_ids|Ids)='([^']+)'(?:\s+(value|reasoning)='((?:[^'\\]|\\.)*)')?\s*\/>/g;
     const citationMatches = [...middleCite.matchAll(citationRegex)];
     const match = citationMatches?.[0];
     const pageNumber = match?.[2] ? parseInt(match?.[2]) : undefined;
@@ -133,7 +133,7 @@ export const parseCitation = (fragment, mdAttachmentId, citationCounterRef, isVe
     // 3: timestamps content
     // 4: Optional Key (value|reasoning)
     // 5: Optional Value content (escaped)
-    const avCitationRegex = /<cite\s+file(?:_id|Id)='(\w{0,25})'\s+full_phrase='((?:[^'\\]|\\.)*)'\s+timestamps='([^']+)'(?:\s+(value|reasoning)='((?:[^'\\]|\\.)*)')?\s*\/>/g;
+    const avCitationRegex = /<cite\s+(?:attachment_id|attachmentId|file_id|fileId)='(\w{0,25})'\s+full_phrase='((?:[^'\\]|\\.)*)'\s+timestamps='([^']+)'(?:\s+(value|reasoning)='((?:[^'\\]|\\.)*)')?\s*\/>/g;
     const avCitationMatches = [...middleCite.matchAll(avCitationRegex)];
     const avMatch = avCitationMatches?.[0];
     let timestamps;

package/lib/prompts/citationPrompts.d.ts

@@ -1,5 +1,5 @@
-export declare const CITATION_MARKDOWN_SYNTAX_PROMPT = "\nCitation syntax to use within Markdown:\n\u2022 To support any ideas or information that requires a citation from the provided content, use the following citation syntax:\n<cite file_id='attachment_id' start_page_key='page_number_PAGE_index_INDEX' full_phrase='the verbatim text of the terse phrase inside <file_text />; remember to escape quotes and newlines inside the full_phrase to remain as valid JSON' key_span='the verbatim 1-3 words within full_phrase that best support the citation' line_ids='2-6' reasoning='the terse logic used to conclude the citation' />\n\n\u2022 Very important: for page numbers, only use the page number and page index info from the page_number_PAGE_index_INDEX format (e.g. <page_number_1_index_0>) and never from the contents inside the page.\n\u2022 start_page_key, full_phrase, and line_ids are required for each citation.\n\u2022 Infer line_ids, as we only provide the first, last, and every 5th line. When copying a previous <cite />, use the full info from the previous citation without changing the start_page_key, line_ids, or any other <cite /> attributes.\n\u2022 Use refer to line_ids inclusively, and use a range (or single) for each citation, split multiple sequential line_ids into multiple citations.\n\u2022 These citations will be replaced and displayed in-line as a numeric element (e.g. [1]), the markdown preceding <cite /> should read naturally with only one <cite /> per sentence with rare exceptions for two <cite /> in a sentence. <cite /> often present best at the end of the sentence, and are not grouped at the end of the document.\n\u2022 The full_phrase should be the exact verbatim text of the phrase or paragraph from the source document to support the insight or idea.\n\u2022 We do NOT put the full_phrase inside <cite ...></cite>; we only use full_phrase inside the full_phrase attribute.\n";
-export declare const AV_CITATION_MARKDOWN_SYNTAX_PROMPT = "\n\u2022 To support any ideas or information that requires a citation from the provided content, use the following citation syntax:\n<cite file_id='attachment_id' full_phrase='the verbatim text of the phrase; remember to escape quotes and newlines inside the full_phrase to remain as valid JSON' timestamps='HH:MM:SS.SSS-HH:MM:SS.SSS' reasoning='the logic connecting the form section requirements to the supporting source citation' />\n\u2022 These citations are displayed in-line or in the relevant list item, and are not grouped at the end of the document.\n";
+export declare const CITATION_MARKDOWN_SYNTAX_PROMPT = "\nCitation syntax to use within Markdown:\n\u2022 To support any ideas or information that requires a citation from the provided content, use the following citation syntax:\n<cite attachment_id='attachment_id' start_page_key='page_number_PAGE_index_INDEX' full_phrase='the verbatim text of the terse phrase inside <attachment_text />; remember to escape quotes and newlines inside the full_phrase to remain as valid JSON' key_span='the verbatim 1-3 words within full_phrase that best support the citation' line_ids='2-6' reasoning='the terse logic used to conclude the citation' />\n\n\u2022 Very important: for page numbers, only use the page number and page index info from the page_number_PAGE_index_INDEX format (e.g. <page_number_1_index_0>) and never from the contents inside the page.\n\u2022 start_page_key, full_phrase, and line_ids are required for each citation.\n\u2022 Infer line_ids, as we only provide the first, last, and every 5th line. When copying a previous <cite />, use the full info from the previous citation without changing the start_page_key, line_ids, or any other <cite /> attributes.\n\u2022 Use refer to line_ids inclusively, and use a range (or single) for each citation, split multiple sequential line_ids into multiple citations.\n\u2022 These citations will be replaced and displayed in-line as a numeric element (e.g. [1]), the markdown preceding <cite /> should read naturally with only one <cite /> per sentence with rare exceptions for two <cite /> in a sentence. <cite /> often present best at the end of the sentence, and are not grouped at the end of the document.\n\u2022 The full_phrase should be the exact verbatim text of the phrase or paragraph from the source document to support the insight or idea.\n\u2022 We do NOT put the full_phrase inside <cite ...></cite>; we only use full_phrase inside the full_phrase attribute.\n";
+export declare const AV_CITATION_MARKDOWN_SYNTAX_PROMPT = "\n\u2022 To support any ideas or information that requires a citation from the provided content, use the following citation syntax:\n<cite attachment_id='attachment_id' full_phrase='the verbatim text of the phrase; remember to escape quotes and newlines inside the full_phrase to remain as valid JSON' timestamps='HH:MM:SS.SSS-HH:MM:SS.SSS' reasoning='the logic connecting the form section requirements to the supporting source citation' />\n\u2022 These citations are displayed in-line or in the relevant list item, and are not grouped at the end of the document.\n";
 export interface WrapSystemPromptOptions {
     /** The original system prompt to wrap with citation instructions */
     systemPrompt: string;
@@ -134,4 +134,5 @@ export declare const CITATION_AV_BASED_JSON_OUTPUT_FORMAT: {
             description: string;
         };
     };
+    required: string[];
 };

package/lib/prompts/citationPrompts.js

@@ -1,7 +1,7 @@
 export const CITATION_MARKDOWN_SYNTAX_PROMPT = `
 Citation syntax to use within Markdown:
 • To support any ideas or information that requires a citation from the provided content, use the following citation syntax:
-<cite file_id='attachment_id' start_page_key='page_number_PAGE_index_INDEX' full_phrase='the verbatim text of the terse phrase inside <file_text />; remember to escape quotes and newlines inside the full_phrase to remain as valid JSON' key_span='the verbatim 1-3 words within full_phrase that best support the citation' line_ids='2-6' reasoning='the terse logic used to conclude the citation' />
+<cite attachment_id='attachment_id' start_page_key='page_number_PAGE_index_INDEX' full_phrase='the verbatim text of the terse phrase inside <attachment_text />; remember to escape quotes and newlines inside the full_phrase to remain as valid JSON' key_span='the verbatim 1-3 words within full_phrase that best support the citation' line_ids='2-6' reasoning='the terse logic used to conclude the citation' />
 
 • Very important: for page numbers, only use the page number and page index info from the page_number_PAGE_index_INDEX format (e.g. <page_number_1_index_0>) and never from the contents inside the page.
 • start_page_key, full_phrase, and line_ids are required for each citation.
@@ -13,7 +13,7 @@ Citation syntax to use within Markdown:
 `;
 export const AV_CITATION_MARKDOWN_SYNTAX_PROMPT = `
 • To support any ideas or information that requires a citation from the provided content, use the following citation syntax:
-<cite file_id='attachment_id' full_phrase='the verbatim text of the phrase; remember to escape quotes and newlines inside the full_phrase to remain as valid JSON' timestamps='HH:MM:SS.SSS-HH:MM:SS.SSS' reasoning='the logic connecting the form section requirements to the supporting source citation' />
+<cite attachment_id='attachment_id' full_phrase='the verbatim text of the phrase; remember to escape quotes and newlines inside the full_phrase to remain as valid JSON' timestamps='HH:MM:SS.SSS-HH:MM:SS.SSS' reasoning='the logic connecting the form section requirements to the supporting source citation' />
 • These citations are displayed in-line or in the relevant list item, and are not grouped at the end of the document.
 `;
 /**
@@ -121,7 +121,7 @@ export const CITATION_JSON_OUTPUT_FORMAT = {
         },
         fullPhrase: {
             type: "string",
-            description: "The verbatim text of the terse phrase inside <file_text /> to support the citation (if there is a detected OCR correction, use the corrected text)",
+            description: "The verbatim text of the terse phrase inside <attachment_text /> to support the citation (if there is a detected OCR correction, use the corrected text)",
         },
         keySpan: {
             type: "string",
@@ -164,4 +164,5 @@ export const CITATION_AV_BASED_JSON_OUTPUT_FORMAT = {
             description: "The timestamp of the audio or video frame including milliseconds formatted as: HH:MM:SS.SSS",
         },
     },
+    required: ["attachmentId", "startPageKey", "fullPhrase", "timestamps"],
 };

package/lib/prompts/promptCompression.js

@@ -14,17 +14,20 @@ function buildSafePrefixMap(ids, prompt) {
             const digitCount = (prefix.match(/\d/g) || []).length;
             const letterCount = (prefix.match(/[a-zA-Z]/g) || []).length;
             if (prefix.length < MIN_PREFIX_LENGTH ||
-                (digitCount > 0 && letterCount < MIN_CHARACTERS_PER_PREFIX_WITH_AT_LEAST_ONE_DIGIT) ||
-                (digitCount === 0 && letterCount < MIN_CHARACTERS_PER_PREFIX_WITH_NO_DIGITS)) {
+                (digitCount > 0 &&
+                    letterCount < MIN_CHARACTERS_PER_PREFIX_WITH_AT_LEAST_ONE_DIGIT) ||
+                (digitCount === 0 &&
+                    letterCount < MIN_CHARACTERS_PER_PREFIX_WITH_NO_DIGITS)) {
                 continue;
             }
             // 1) Unique among IDs
-            if (ids.some(other => other !== id && other.startsWith(prefix))) {
+            if (ids.some((other) => other !== id && other.startsWith(prefix))) {
                 continue;
             }
             // 2) Only appears in prompt as part of the full ID
             const esc = (s) => s.replace(/[-\/\\^$*+?.()|[\]{}]/g, "\\$&");
-            const prefixCount = (prompt.match(new RegExp(esc(prefix), "g")) || []).length;
+            const prefixCount = (prompt.match(new RegExp(esc(prefix), "g")) || [])
+                .length;
             const fullCount = (prompt.match(new RegExp(esc(id), "g")) || []).length;
             if (prefixCount !== fullCount) {
                 continue;
@@ -91,18 +94,29 @@ export function decompressPromptIds(compressed, prefixMap) {
         const escPrefix = prefix.replace(/[-\/\\^$*+?.()|[\]{}]/g, "\\$&");
         text = text.replace(new RegExp(escPrefix, "g"), full);
     }
-    //this is for citation attachmentId, file_id, or fileId
-    if (entries.length === 1 && (text.includes("file_id='") || text.includes('file_id="'))) {
-        const fullId = entries[0][1];
-        text = text.replace(/file_id='[^']*'|file_id="[^"]*"/g, `file_id='${fullId}'`);
-    }
-    else if (entries.length === 1 && (text.includes("fileId='") || text.includes('fileId="'))) {
-        const fullId = entries[0][1];
-        text = text.replace(/fileId='[^']*'|fileId="[^"]*"/g, `fileId='${fullId}'`);
-    }
-    else if (entries.length === 1 && (text.includes("attachmentId='") || text.includes('attachmentId="'))) {
-        const fullId = entries[0][1];
-        text = text.replace(/attachmentId='[^']*'|attachmentId="[^"]*"/g, `attachmentId='${fullId}'`);
+    // Handle cases where the LLM may output ID in a different attribute format
+    // We look for common ID attribute patterns and replace compressed prefixes within them
+    // Note: fileId variants are supported for backwards compatibility with legacy citations
+    const idAttributeKeys = [
+        "attachmentId",
+        "attachment_id",
+        "attachment_ID",
+        "attachmentID",
+        "fileId",
+        "file_id",
+        "file_ID",
+        "fileID",
+        "fileid",
+    ];
+    // For each prefix, look for it within ID attribute values and replace with full ID
+    for (const [prefix, full] of entries) {
+        const escPrefix = prefix.replace(/[-\/\\^$*+?.()|[\]{}]/g, "\\$&");
+        const keyPattern = idAttributeKeys.join("|");
+        const quotePattern = "([\"'`])";
+        // Match: attributeName = 'prefix' or attributeName="prefix" etc.
+        // Only replace the prefix part, preserving the attribute name and quotes
+        const re = new RegExp(`(${keyPattern})(\\s*=\\s*)${quotePattern}${escPrefix}\\3`, "g");
+        text = text.replace(re, `$1$2$3${full}$3`);
     }
     const newLength = text?.length;
     const diff = originalLength - newLength;

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@deepcitation/deepcitation-js",
-    "version": "1.1.18",
+    "version": "1.1.19",
     "description": "DeepCitation JavaScript SDK for deterministic AI citation verification",
     "type": "module",
     "private": false,