@juspay/neurolink 9.88.6 → 9.88.7

package/dist/lib/utils/csvProcessor.js

@@ -337,6 +337,20 @@ function calculateDataQualityScore(columns, warnings, totalRows) {
 /**
  * Analyze all columns in parsed CSV data
  */
+/**
+ * Confidence for a CSV processing result, weighted by data quality.
+ *
+ * The result used to report a static `confidence: 100` regardless of how clean
+ * the data was. This ties confidence to `dataQualityScore` so it actually
+ * varies with quality factors (ragged rows, high null rate, type
+ * inconsistency): a cleanly-parsed CSV stays at 100, a messy one is pulled
+ * halfway toward its quality score. It never drops below 50 for a CSV that
+ * parsed — the *format* is certain, only the *data* is imperfect.
+ */
+function csvResultConfidence(dataQualityScore) {
+    const clamped = Math.max(0, Math.min(100, dataQualityScore));
+    return Math.round((100 + clamped) / 2);
+}
 function analyzeColumns(rows) {
     if (rows.length === 0) {
         return {
@@ -574,7 +588,7 @@ export class CSVProcessor {
                 content: limitedCSV,
                 mimeType: "text/csv",
                 metadata: {
-                    confidence: 100,
+                    confidence: csvResultConfidence(dataQualityScore),
                     size: content.length,
                     rowCount,
                     totalLines: limitedLines.length,
@@ -648,7 +662,7 @@ export class CSVProcessor {
             content: formatted,
             mimeType: "text/csv",
             metadata: {
-                confidence: 100,
+                confidence: csvResultConfidence(dataQualityScore),
                 size: content.length,
                 rowCount,
                 columnCount,

package/dist/lib/utils/fileDetector.js

@@ -2166,13 +2166,38 @@ class ContentHeuristicStrategy {
             const hasUniformLengths = stdDev / avgLength < 0.75;
             return hasReasonableLengths && noBinaryChars && hasUniformLengths;
         }
-        // Count delimiters per line and check consistency
-        const delimRegex = delimiter === "|" ? /\|/g : new RegExp(delimiter, "g");
-        const counts = lines.map((line) => (line.match(delimRegex) || []).length);
+        // Count delimiters per line and check consistency. Delimiters inside a
+        // double-quoted field are field content, not column separators (RFC 4180) —
+        // a naive count inflates rows with quoted delimiters (e.g. `"Smith, John"`),
+        // which used to make consistency collapse and reject a valid CSV.
+        const counts = lines.map((line) => ContentHeuristicStrategy.countDelimitersOutsideQuotes(line, delimiter));
         const firstCount = counts[0];
         const consistentLines = counts.filter((c) => c === firstCount).length;
         return consistentLines / lines.length >= 0.8;
     }
+    /**
+     * Count occurrences of `delimiter` in `line` that fall OUTSIDE double-quoted
+     * fields, honoring RFC-4180 escaped quotes (`""`). Used by CSV detection so a
+     * delimiter embedded in a quoted value is not mistaken for a column break.
+     */
+    static countDelimitersOutsideQuotes(line, delimiter) {
+        let count = 0;
+        let inQuotes = false;
+        for (let i = 0; i < line.length; i++) {
+            const ch = line[i];
+            if (ch === '"') {
+                if (inQuotes && line[i + 1] === '"') {
+                    i++; // escaped quote inside a quoted field — skip the pair
+                    continue;
+                }
+                inQuotes = !inQuotes;
+            }
+            else if (ch === delimiter && !inQuotes) {
+                count++;
+            }
+        }
+        return count;
+    }
     looksLikeJSON(text) {
         // hasJsonMarkers now does full validation including JSON.parse
         return hasJsonMarkers(text);

package/dist/utils/csvProcessor.js

@@ -337,6 +337,20 @@ function calculateDataQualityScore(columns, warnings, totalRows) {
 /**
  * Analyze all columns in parsed CSV data
  */
+/**
+ * Confidence for a CSV processing result, weighted by data quality.
+ *
+ * The result used to report a static `confidence: 100` regardless of how clean
+ * the data was. This ties confidence to `dataQualityScore` so it actually
+ * varies with quality factors (ragged rows, high null rate, type
+ * inconsistency): a cleanly-parsed CSV stays at 100, a messy one is pulled
+ * halfway toward its quality score. It never drops below 50 for a CSV that
+ * parsed — the *format* is certain, only the *data* is imperfect.
+ */
+function csvResultConfidence(dataQualityScore) {
+    const clamped = Math.max(0, Math.min(100, dataQualityScore));
+    return Math.round((100 + clamped) / 2);
+}
 function analyzeColumns(rows) {
     if (rows.length === 0) {
         return {
@@ -574,7 +588,7 @@ export class CSVProcessor {
                 content: limitedCSV,
                 mimeType: "text/csv",
                 metadata: {
-                    confidence: 100,
+                    confidence: csvResultConfidence(dataQualityScore),
                     size: content.length,
                     rowCount,
                     totalLines: limitedLines.length,
@@ -648,7 +662,7 @@ export class CSVProcessor {
             content: formatted,
             mimeType: "text/csv",
             metadata: {
-                confidence: 100,
+                confidence: csvResultConfidence(dataQualityScore),
                 size: content.length,
                 rowCount,
                 columnCount,

package/dist/utils/fileDetector.js

@@ -2166,13 +2166,38 @@ class ContentHeuristicStrategy {
             const hasUniformLengths = stdDev / avgLength < 0.75;
             return hasReasonableLengths && noBinaryChars && hasUniformLengths;
         }
-        // Count delimiters per line and check consistency
-        const delimRegex = delimiter === "|" ? /\|/g : new RegExp(delimiter, "g");
-        const counts = lines.map((line) => (line.match(delimRegex) || []).length);
+        // Count delimiters per line and check consistency. Delimiters inside a
+        // double-quoted field are field content, not column separators (RFC 4180) —
+        // a naive count inflates rows with quoted delimiters (e.g. `"Smith, John"`),
+        // which used to make consistency collapse and reject a valid CSV.
+        const counts = lines.map((line) => ContentHeuristicStrategy.countDelimitersOutsideQuotes(line, delimiter));
         const firstCount = counts[0];
         const consistentLines = counts.filter((c) => c === firstCount).length;
         return consistentLines / lines.length >= 0.8;
     }
+    /**
+     * Count occurrences of `delimiter` in `line` that fall OUTSIDE double-quoted
+     * fields, honoring RFC-4180 escaped quotes (`""`). Used by CSV detection so a
+     * delimiter embedded in a quoted value is not mistaken for a column break.
+     */
+    static countDelimitersOutsideQuotes(line, delimiter) {
+        let count = 0;
+        let inQuotes = false;
+        for (let i = 0; i < line.length; i++) {
+            const ch = line[i];
+            if (ch === '"') {
+                if (inQuotes && line[i + 1] === '"') {
+                    i++; // escaped quote inside a quoted field — skip the pair
+                    continue;
+                }
+                inQuotes = !inQuotes;
+            }
+            else if (ch === delimiter && !inQuotes) {
+                count++;
+            }
+        }
+        return count;
+    }
     looksLikeJSON(text) {
         // hasJsonMarkers now does full validation including JSON.parse
         return hasJsonMarkers(text);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@juspay/neurolink",
-  "version": "9.88.6",
+  "version": "9.88.7",
   "packageManager": "pnpm@10.15.1",
   "description": "TypeScript AI SDK with 24+ LLM providers behind one consistent API. MCP-native (58+ servers), voice TTS/STT/realtime, RAG, agents, memory, context compaction. OpenAI · Anthropic · Gemini · Bedrock · Azure · Ollama · DeepSeek · NVIDIA NIM and more.",
   "author": {