@codeyam/codeyam-cli 0.1.0-staging.28f73cf → 0.1.0-staging.2a88920

package/analyzer-template/packages/ai/src/lib/dataStructure/helpers/deduplicateFunctionSchemas.ts

@@ -77,42 +77,52 @@ export default function deduplicateFunctionSchemas(
 
       // If multiple variants with arguments exist, find the one with the most robust structure
       if (withArgs.length > 1) {
-        // Count the number of related schema keys for each variant
-        const variantDepths = withArgs.map((variant) => {
-          // Find the end of the function call (the closing parenthesis)
+        // Pre-compute function call prefixes for all variants
+        const variantPrefixes = withArgs.map((variant) => {
           const closingParenIndex = variant.lastIndexOf(')');
-          const functionCallPrefix =
-            closingParenIndex > -1
-              ? variant.slice(0, closingParenIndex + 1)
-              : variant;
-
-          const relatedKeys = Object.keys(schema).filter((k) => {
-            // Check if this key is a child of the variant (starts with the function call)
-            return (
-              k.startsWith(functionCallPrefix + '.') ||
-              k.startsWith(functionCallPrefix + '[')
-            );
-          });
-
           return {
             key: variant,
-            depth: relatedKeys.length,
-            maxDepth: Math.max(
-              0,
-              ...relatedKeys.map((k) => {
-                // Count the nesting depth by counting dots and brackets after the variant
-                const remaining = k.slice(functionCallPrefix.length);
-                return (
-                  (remaining.match(/\./g) || []).length +
-                  (remaining.match(/\[/g) || []).length
-                );
-              }),
-            ),
+            prefix:
+              closingParenIndex > -1
+                ? variant.slice(0, closingParenIndex + 1)
+                : variant,
+            depth: 0,
+            maxDepth: 0,
           };
         });
 
+        // Single pass through schema keys to count related keys for ALL variants at once.
+        // This is O(schemaSize × variants) but done in one pass instead of
+        // O(variants × schemaSize) with separate passes per variant.
+        const schemaKeys = Object.keys(schema);
+        for (const k of schemaKeys) {
+          for (const variantInfo of variantPrefixes) {
+            // Check if key starts with prefix followed by . or [
+            const prefixLen = variantInfo.prefix.length;
+            if (k.length > prefixLen) {
+              const nextChar = k[prefixLen];
+              if (
+                (nextChar === '.' || nextChar === '[') &&
+                k.startsWith(variantInfo.prefix)
+              ) {
+                variantInfo.depth++;
+                // Count nesting depth by counting . and [ characters (faster than regex)
+                const remaining = k.slice(prefixLen);
+                let nestingDepth = 0;
+                for (let i = 0; i < remaining.length; i++) {
+                  const c = remaining[i];
+                  if (c === '.' || c === '[') nestingDepth++;
+                }
+                if (nestingDepth > variantInfo.maxDepth) {
+                  variantInfo.maxDepth = nestingDepth;
+                }
+              }
+            }
+          }
+        }
+
         // Sort by depth (number of related keys) and then by max depth (nesting level)
-        variantDepths.sort((a, b) => {
+        variantPrefixes.sort((a, b) => {
           if (b.depth !== a.depth) {
             return b.depth - a.depth;
           }
@@ -120,7 +130,7 @@ export default function deduplicateFunctionSchemas(
         });
 
         // Keep the one with the most robust structure, remove others
-        const keepVariant = variantDepths[0].key;
+        const keepVariant = variantPrefixes[0].key;
         withArgs.forEach((variant) => {
           if (variant !== keepVariant) {
             keysToRemove.add(variant);

package/analyzer-template/packages/ai/src/lib/dataStructure/helpers/fillInSchemaGapsAndUnknowns.ts

@@ -205,14 +205,88 @@ export default function fillInSchemaGapsAndUnknowns(
   }
 }
 
+/**
+ * Pre-built indexes that can be reused across multiple calls to fillInDirectSchemaGapsAndUnknowns.
+ * Build once with buildSchemaIndexes(), then pass to all calls processing the same schema.
+ */
+export interface SchemaIndexes {
+  functionKeysMapping: Record<string, string[]>;
+  prefixMaxDepth: Map<string, number>;
+}
+
+/**
+ * Build indexes for a schema that can be reused across multiple calls.
+ * This avoids rebuilding indexes (~1-2 seconds for 18k keys) on every call.
+ *
+ * @param schema - The schema to build indexes from (typically the full dependency schema)
+ * @returns Indexes that can be passed to fillInDirectSchemaGapsAndUnknowns
+ */
+export function buildSchemaIndexes(
+  schema: Record<string, string>,
+): SchemaIndexes {
+  // Build functionKeysMapping + prefixMaxDepth in a single pass
+  // Avoid repeated split/join work by caching parts and building prefixes incrementally.
+  const functionKeysMapping: Record<string, string[]> = {};
+  const prefixMaxDepth = new Map<string, number>();
+  const partsCache = new Map<string, string[]>();
+
+  for (const key in schema) {
+    const parts = getCachedParts(partsCache, key);
+    const lastPart = parts[parts.length - 1];
+    const prefixes = buildPrefixParts(parts);
+
+    if (
+      key.endsWith(')') ||
+      schema[key] === 'function' ||
+      lastPart === 'length'
+    ) {
+      const allButLastPart = parts.length > 1 ? prefixes[parts.length - 2] : '';
+      functionKeysMapping[allButLastPart] ||= [];
+      functionKeysMapping[allButLastPart].push(lastPart);
+    }
+
+    for (let i = 0; i < prefixes.length; i++) {
+      const prefix = prefixes[i];
+      const existingMax = prefixMaxDepth.get(prefix) ?? 0;
+      if (parts.length > existingMax) {
+        prefixMaxDepth.set(prefix, parts.length);
+      }
+    }
+  }
+
+  return { functionKeysMapping, prefixMaxDepth };
+}
+
 export function fillInDirectSchemaGapsAndUnknowns({
   scopeName,
   schema,
+  mergedSchema,
   attempts = 0,
+  // Pre-built indexes from buildSchemaIndexes() - use for cross-call caching
+  prebuiltIndexes,
+  // Internal: pre-computed indexes to avoid rebuilding on recursion
+  _functionKeysMapping,
+  _prefixMaxDepth,
+  _partsCache,
+  _prefixPartsCache,
+  _knownTypeCache,
+  _guessTypeCache,
+  _nameHintCache,
 }: {
   scopeName?: string;
   schema: Record<string, string>;
+  mergedSchema?: Record<string, string>;
   attempts?: number;
+  /** Pre-built indexes from buildSchemaIndexes() - pass to reuse across multiple calls */
+  prebuiltIndexes?: SchemaIndexes;
+  // Internal: pre-computed indexes to avoid rebuilding on recursion (used by recursive calls)
+  _functionKeysMapping?: Record<string, string[]>;
+  _prefixMaxDepth?: Map<string, number>;
+  _partsCache?: Map<string, string[]>;
+  _prefixPartsCache?: Map<string, string[]>;
+  _knownTypeCache?: Map<string, string | undefined>;
+  _guessTypeCache?: Map<string, string | undefined>;
+  _nameHintCache?: Map<string, string | undefined>;
 }) {
   try {
     const existingSchema = (path: any) => {
@@ -223,96 +297,204 @@ export function fillInDirectSchemaGapsAndUnknowns({
       }
     };
 
-    const functionKeysMapping = [...Object.keys(schema)]
-      .filter((key) => {
-        if (key.endsWith(')') || schema[key] === 'function') return true;
-        const parts = splitOutsideParenthesesAndArrays(key);
-        if (parts[parts.length - 1] === 'length') return true;
-      })
-      .reduce(
-        (acc, key) => {
-          const parts = splitOutsideParenthesesAndArrays(key);
-          const allButLastPart = joinParenthesesAndArrays(parts.slice(0, -1));
+    const partsCache = _partsCache ?? new Map<string, string[]>();
+    const prefixPartsCache = _prefixPartsCache ?? new Map<string, string[]>();
+    const knownTypeCache =
+      _knownTypeCache ?? new Map<string, string | undefined>();
+    const guessTypeCache =
+      _guessTypeCache ?? new Map<string, string | undefined>();
+    const nameHintCache =
+      _nameHintCache ?? new Map<string, string | undefined>();
+
+    const getParts = (path: string) => getCachedParts(partsCache, path);
+    const getPrefixParts = (path: string) => {
+      const cached = prefixPartsCache.get(path);
+      if (cached) return cached;
+      const prefixes = buildPrefixParts(getParts(path));
+      prefixPartsCache.set(path, prefixes);
+      return prefixes;
+    };
+
+    // Index resolution priority:
+    // 1. _functionKeysMapping/_prefixMaxDepth (from recursive calls within this invocation)
+    // 2. prebuiltIndexes (from cross-call caching - built once, reused across multiple calls)
+    // 3. Build new indexes (fallback - expensive for large schemas)
+    const functionKeysMapping =
+      _functionKeysMapping ??
+      prebuiltIndexes?.functionKeysMapping ??
+      (() => {
+        const mapping: Record<string, string[]> = {};
+        for (const key in schema) {
+          const parts = getParts(key);
           const lastPart = parts[parts.length - 1];
-          acc[allButLastPart] ||= [];
-          acc[allButLastPart].push(lastPart);
-          return acc;
-        },
-        {} as Record<string, string[]>,
-      );
+          if (
+            key.endsWith(')') ||
+            schema[key] === 'function' ||
+            lastPart === 'length'
+          ) {
+            const prefixes = getPrefixParts(key);
+            const allButLastPart =
+              parts.length > 1 ? prefixes[parts.length - 2] : '';
+            mapping[allButLastPart] ||= [];
+            mapping[allButLastPart].push(lastPart);
+          }
+        }
+        return mapping;
+      })();
+
+    // Build prefix index for O(1) lookups in checkIfObjectOrFunction
+    // Maps each prefix to the max depth of keys that have this prefix
+    // This replaces O(n) scans with O(1) lookups, fixing the O(n²) bottleneck
+    let prefixMaxDepth =
+      _prefixMaxDepth ?? prebuiltIndexes?.prefixMaxDepth ?? null;
+    if (!prefixMaxDepth) {
+      prefixMaxDepth = new Map<string, number>();
+      const targetSchema = mergedSchema ?? schema;
+      for (const key in targetSchema) {
+        const keyParts = getParts(key);
+        const prefixes = buildPrefixParts(keyParts);
+        // Record each prefix of this key with its max depth
+        for (let i = 0; i < prefixes.length; i++) {
+          const prefix = prefixes[i];
+          const existingMax = prefixMaxDepth.get(prefix) ?? 0;
+          if (keyParts.length > existingMax) {
+            prefixMaxDepth.set(prefix, keyParts.length);
+          }
+        }
+      }
+    }
 
-    let changeMade = false;
-    const guessTypeForPath = (path: string, defaultType = 'string') => {
-      let knownType = checkIfKnownType(path, functionKeysMapping);
-      if (!knownType) {
-        knownType = checkIfObjectOrFunction(path, schema);
+    // O(1) replacement for checkIfObjectOrFunction
+    const checkIfObjectOrFunctionFast = (path: string): string | undefined => {
+      const maxDepth = prefixMaxDepth!.get(path);
+
+      if (maxDepth === undefined) return undefined; // No key starts with this path
+
+      const pathParts = getParts(path);
+      // Check if path's last part ends with ')' → function
+      const lastPart = pathParts[pathParts.length - 1];
+      if (lastPart?.endsWith(')')) {
+        return 'function';
       }
 
-      if (knownType && schema[path] !== knownType) {
-        return knownType;
+      // Check if there are deeper keys
+      if (maxDepth > pathParts.length) {
+        return 'object';
       }
 
-      const keyParts = splitOutsideParenthesesAndArrays(path);
-      const lastPart = keyParts[keyParts.length - 1];
+      return undefined;
+    };
 
+    let changeMade = false;
+    const getKnownType = (path: string) => {
+      if (knownTypeCache.has(path)) return knownTypeCache.get(path);
+      const known = checkIfKnownType(path, functionKeysMapping);
+      knownTypeCache.set(path, known);
+      return known;
+    };
+    const guessTypeFromName = (lastPart: string, defaultType: string) => {
+      if (nameHintCache.has(lastPart)) {
+        return nameHintCache.get(lastPart) ?? defaultType;
+      }
+
+      let hint: string | undefined;
       if (isLikelyBoolean(lastPart)) {
-        return 'boolean';
+        hint = 'boolean';
       } else if (isLikelyDate(lastPart)) {
-        return 'date';
+        hint = 'date';
       } else if (isLikelyAction(lastPart)) {
-        return 'function';
+        hint = 'function';
       } else if (pluralize.isPlural(lastPart)) {
-        return 'number';
-      } else {
-        return defaultType;
+        // Default to number instead of array - safer for JSX rendering
+        // Arrays cause "Objects are not valid as a React child" errors when rendered directly
+        hint = 'number';
+      }
+
+      nameHintCache.set(lastPart, hint);
+      return hint ?? defaultType;
+    };
+    const guessTypeForPath = (path: string, defaultType = 'string') => {
+      const cacheKey = `${path}|${defaultType}`;
+      if (guessTypeCache.has(cacheKey)) {
+        return guessTypeCache.get(cacheKey);
+      }
+
+      let knownType = getKnownType(path);
+      if (!knownType) {
+        // Use optimized O(1) prefix lookup instead of O(n) scan
+        knownType = checkIfObjectOrFunctionFast(path);
       }
+
+      if (knownType) {
+        guessTypeCache.set(cacheKey, knownType);
+        return knownType;
+      }
+
+      const keyParts = getParts(path);
+      const lastPart = keyParts[keyParts.length - 1];
+      const guessed = guessTypeFromName(lastPart, defaultType);
+      guessTypeCache.set(cacheKey, guessed);
+      return guessed;
     };
 
     for (const key in schema) {
       if (!schema[key]) schema[key] = 'unknown';
       if (schema[key].includes(' | unknown')) {
-        schema[key] = schema[key].replace(' | unknown', '');
-        changeMade = true;
+        const cleaned = schema[key].replace(' | unknown', '');
+        if (schema[key] !== cleaned) {
+          schema[key] = cleaned;
+          changeMade = true;
+        }
       } else if (schema[key].startsWith('unknown | ')) {
         // Handle "unknown | undefined" -> "string | undefined"
         const remainingType = schema[key].substring('unknown | '.length);
         const newType = guessTypeForPath(key, 'string');
         if (newType) {
-          schema[key] = `${newType} | ${remainingType}`;
-          changeMade = true;
+          const replacement = `${newType} | ${remainingType}`;
+          if (schema[key] !== replacement) {
+            schema[key] = replacement;
+            changeMade = true;
+          }
         }
       } else if (schema[key] === 'unknown') {
         const newType = guessTypeForPath(key, 'string');
-        if (newType) {
+        if (newType && schema[key] !== newType) {
           schema[key] = newType;
           changeMade = true;
         }
       }
 
-      const keyParts = splitOutsideParenthesesAndArrays(key);
+      const keyParts = getParts(key);
+      const prefixParts = getPrefixParts(key);
       for (let i = 0; i < keyParts.length; i++) {
-        const subPathParts = keyParts.slice(0, i + 1);
-        const lastSubPathPart = subPathParts[subPathParts.length - 1];
-
-        const previousSubPathParts = subPathParts.slice(0, i);
-        if (previousSubPathParts.length === 0) continue;
-
-        const previousSubPath = joinParenthesesAndArrays(previousSubPathParts);
-
-        if (
-          cleanOutBoundary(lastSubPathPart).match(/\[\d*\]/) &&
-          !lastSubPathPart.match(/signature\[\d+\]/)
-        ) {
-          if (schema[previousSubPath] !== 'array') {
-            schema[previousSubPath] = 'array';
-            changeMade = true;
+        if (i === 0) continue;
+        const lastSubPathPart = keyParts[i];
+        const previousSubPath = prefixParts[i - 1];
+        const isSignatureIndex =
+          lastSubPathPart.includes('signature[') &&
+          SIGNATURE_INDEX_REGEX.test(lastSubPathPart);
+
+        if (lastSubPathPart.includes('[')) {
+          const cleaned = cleanOutBoundary(lastSubPathPart);
+          if (cleaned.match(/\[\d*\]/) && !isSignatureIndex) {
+            // Fix 39: Don't overwrite explicit 'object' types with 'array'
+            // This handles spurious [] paths from components like JsonNode that handle
+            // both arrays and objects. When metadata is explicitly typed as 'object',
+            // paths like metadata[] (from dynamic iteration) should not change it.
+            if (
+              schema[previousSubPath] !== 'array' &&
+              schema[previousSubPath] !== 'object'
+            ) {
+              schema[previousSubPath] = 'array';
+              changeMade = true;
+            }
           }
         }
 
         const isFunction =
           lastSubPathPart.endsWith(')') ||
           lastSubPathPart === 'functionCallReturnValue' ||
-          lastSubPathPart.match(/signature\[\d+\]/);
+          isSignatureIndex;
 
         if (
           !isFunction &&
@@ -320,8 +502,7 @@ export function fillInDirectSchemaGapsAndUnknowns({
             schema[previousSubPath],
           )
         ) {
-          const newValue =
-            checkIfKnownType(previousSubPath, functionKeysMapping) ?? 'object';
+          const newValue = getKnownType(previousSubPath) ?? 'object';
 
           if (
             !schema[previousSubPath] ||
@@ -336,8 +517,7 @@ export function fillInDirectSchemaGapsAndUnknowns({
         if (isFunction && !existingSchema(previousSubPath)) {
           schema[previousSubPath] = previousSubPath.endsWith(')')
             ? 'function'
-            : (checkIfKnownType(previousSubPath, functionKeysMapping) ??
-              'object');
+            : (getKnownType(previousSubPath) ?? 'object');
           changeMade = true;
         }
 
@@ -347,10 +527,12 @@ export function fillInDirectSchemaGapsAndUnknowns({
           const functionReturnValuePath = `${previousSubPath}.functionCallReturnValue`;
 
           // Extract the method name from lastSubPathPart (e.g., "then()" -> "then")
-          const methodName = lastSubPathPart.split('(')[0];
-          const isPromiseMethod = ['then', 'catch', 'finally'].includes(
-            methodName,
-          );
+          const parenIndex = lastSubPathPart.indexOf('(');
+          const methodName =
+            parenIndex === -1
+              ? lastSubPathPart
+              : lastSubPathPart.slice(0, parenIndex);
+          const isPromiseMethod = PROMISE_METHODS.has(methodName);
           const isPreviousAsync = schema[previousSubPath] === 'async-function';
 
           // Don't set to 'function' if this is a promise method on an async function
@@ -373,10 +555,49 @@ export function fillInDirectSchemaGapsAndUnknowns({
       fillInDirectSchemaGapsAndUnknowns({
         scopeName,
         schema,
+        mergedSchema,
         attempts: ++attempts,
+        prebuiltIndexes,
+        // Pass pre-computed indexes to avoid rebuilding on each recursive call
+        _functionKeysMapping: functionKeysMapping,
+        _prefixMaxDepth: prefixMaxDepth,
+        _partsCache: partsCache,
+        _prefixPartsCache: prefixPartsCache,
+        _knownTypeCache: knownTypeCache,
+        _guessTypeCache: guessTypeCache,
+        _nameHintCache: nameHintCache,
       });
     }
 
+    // Remove .length paths when the parent is typed as 'array' or 'string' (or nullable variants).
+    // These are built-in properties that were used for type inference but shouldn't appear
+    // in the final schema (they create { length: N } objects instead of proper arrays).
+    // However, if the parent could be an 'object' or other type, keep .length as it may be
+    // a custom property.
+    for (const key of Object.keys(schema)) {
+      const parts = getParts(key);
+      if (parts[parts.length - 1] === 'length') {
+        const prefixes = getPrefixParts(key);
+        const parentPath = parts.length > 1 ? prefixes[parts.length - 2] : '';
+        const parentType = schema[parentPath];
+        if (parentType) {
+          // Split union types and filter out nullable parts
+          const typeParts = parentType.split(' | ').map((t) => t.trim());
+          const nonNullableTypes = typeParts.filter(
+            (t) => t !== 'undefined' && t !== 'null',
+          );
+          // Only remove if ALL non-nullable types are array or string
+          // This ensures we don't remove .length from potential custom objects
+          if (
+            nonNullableTypes.length > 0 &&
+            nonNullableTypes.every((t) => t === 'array' || t === 'string')
+          ) {
+            delete schema[key];
+          }
+        }
+      }
+    }
+
     return schema;
   } catch (error) {
     console.info(
@@ -394,6 +615,35 @@ export function fillInDirectSchemaGapsAndUnknowns({
   }
 }
 
+const SIGNATURE_INDEX_REGEX = /signature\[\d+\]/;
+const PROMISE_METHODS = new Set(['then', 'catch', 'finally']);
+
+function getCachedParts(cache: Map<string, string[]>, path: string): string[] {
+  const cached = cache.get(path);
+  if (cached) return cached;
+  const parts = splitOutsideParenthesesAndArrays(path);
+  cache.set(path, parts);
+  return parts;
+}
+
+function buildPrefixParts(parts: string[]): string[] {
+  if (parts.length === 0) return [];
+  const prefixes = new Array(parts.length);
+  let prefix = '';
+  for (let i = 0; i < parts.length; i++) {
+    const part = parts[i];
+    if (i === 0) {
+      prefix = part;
+    } else if (part.startsWith('[') || part.startsWith('(')) {
+      prefix += part;
+    } else {
+      prefix += `.${part}`;
+    }
+    prefixes[i] = prefix;
+  }
+  return prefixes;
+}
+
 function checkIfObjectOrFunction(path: string, schema: ScopeNode['schema']) {
   const pathParts = splitOutsideParenthesesAndArrays(path);
   for (const structureKey in schema) {
@@ -446,6 +696,17 @@ const STRONG_ARRAY_METHODS = new Set([
   'with',
 ]);
 
+/**
+ * Check if a method call argument looks like a string literal.
+ * String literals start with ' or " (e.g., includes('foo'), indexOf("bar"))
+ */
+function hasStringLiteralArgument(methodCall: string): boolean {
+  const parenIndex = methodCall.indexOf('(');
+  if (parenIndex === -1) return false;
+  const arg = methodCall.slice(parenIndex + 1).trim();
+  return arg.startsWith("'") || arg.startsWith('"') || arg.startsWith('`');
+}
+
 function checkIfKnownType(
   key: string,
   functionAndAttributesKeysMapping: Record<string, string[]>,
@@ -454,24 +715,41 @@ function checkIfKnownType(
   let isString: boolean | undefined;
   let hasStrongArrayEvidence = false;
   let hasOnlyAmbiguousArrayMethods = true;
+  let hasLengthAccess = false;
+  let hasIncludesOrIndexOfWithVariable = false;
+
+  const entries = functionAndAttributesKeysMapping[key];
+  if (!entries || entries.length === 0) return;
 
-  for (const functionOrAttributeName of functionAndAttributesKeysMapping[key] ??
-    []) {
+  for (const functionOrAttributeName of entries) {
     // Extract the method name without arguments for checking against ambiguous list
     const methodName = functionOrAttributeName.split('(')[0].trim();
 
+    const knownType = knownMethodCalls(functionOrAttributeName);
     const couldBeArray =
-      knownMethodCalls(functionOrAttributeName).includes('array') ||
-      functionOrAttributeName === 'length';
+      knownType.includes('array') || functionOrAttributeName === 'length';
     const couldBeString =
-      knownMethodCalls(functionOrAttributeName).includes('string') ||
-      functionOrAttributeName === 'length';
+      knownType.includes('string') || functionOrAttributeName === 'length';
+
+    // Track .length access
+    if (functionOrAttributeName === 'length') {
+      hasLengthAccess = true;
+    }
 
-    // Check if this is strong evidence of an array
+    // Track methods that exist on both arrays and strings (like includes, indexOf)
+    // Check if .includes() or .indexOf() is called with a variable argument (not a string literal).
+    // Pattern: arr.includes(item) -> likely array (checking if item exists)
+    // Pattern: str.includes('substring') -> likely string (checking for substring)
     if (
-      STRONG_ARRAY_METHODS.has(methodName) ||
-      functionOrAttributeName === 'length'
+      (methodName === 'includes' || methodName === 'indexOf') &&
+      functionOrAttributeName.includes('(') &&
+      !hasStringLiteralArgument(functionOrAttributeName)
     ) {
+      hasIncludesOrIndexOfWithVariable = true;
+    }
+
+    // Check if this is strong evidence of an array (unambiguous array methods)
+    if (STRONG_ARRAY_METHODS.has(methodName)) {
       hasStrongArrayEvidence = true;
     }
 
@@ -495,10 +773,26 @@ function checkIfKnownType(
     }
   }
 
+  // Strong array evidence (unambiguous methods like .map, .filter) -> definitely array
+  if (isArray && hasStrongArrayEvidence) {
+    return 'array';
+  }
+
+  // When .includes() or .indexOf() is called with a variable argument AND we have .length,
+  // prefer array. This pattern (arr.length > 0 && arr.includes(item)) is common for arrays.
+  // But str.includes('literal') is common for strings - that case is handled by hasStringLiteralArgument.
+  if (isArray && hasLengthAccess && hasIncludesOrIndexOfWithVariable) {
+    return 'array';
+  }
+
+  // When both array and string are possible, prefer string.
+  // Strings have .length, .slice(), .includes(), etc. just like arrays.
+  // We only infer 'array' if we have STRONG evidence (like .map, .filter, .reduce).
   if (isString) return 'string';
-  // Only infer array if we have strong evidence OR not only ambiguous methods
-  // This prevents router.push() from being incorrectly inferred as an array
-  if (isArray && (hasStrongArrayEvidence || !hasOnlyAmbiguousArrayMethods)) {
+
+  // Only infer array if we have strong evidence (already handled above)
+  // or non-ambiguous array methods that don't exist on strings
+  if (isArray && !hasOnlyAmbiguousArrayMethods) {
     return 'array';
   }
 }