codegraph/src/resolution/index.ts

/**
 * Reference Resolution Orchestrator
 *
 * Coordinates all reference resolution strategies.
 */

import * as fs from 'fs';
import * as path from 'path';
import { Language, Node, UnresolvedReference, Edge } from '../types';
import { QueryBuilder } from '../db/queries';
import {
  UnresolvedRef,
  ResolvedRef,
  ResolutionResult,
  ResolutionContext,
  FrameworkResolver,
  ImportMapping,
} from './types';
import { isVisibleAcrossFiles, matchReference, matchFunctionRef, matchDottedCallChain, matchScopedCallChain, matchMethodCall, sameLanguageFamily, crossesKnownFamily, dumpNameMatcherProfile, clearNameMatcherMemos } from './name-matcher';
import { resolveViaImport, resolvePhpImportedStaticCall, resolveJvmImport, extractImportMappings, extractReExports, loadCppIncludeDirs, isPhpIncludePathRef, isCobolCopybookRef, isNixPathImportRef, clearImportResolverMemos, resolveImportPath } from './import-resolver';
import { ResolverPool, minRefsForPool } from './resolver-pool';
import { detectFrameworks } from './frameworks';
import { synthesizeCallbackEdges } from './callback-synthesizer';
import { createYielder, type MaybeYield } from './cooperative-yield';
import { loadProjectAliases, type AliasMap } from './path-aliases';
import { loadGoModule, type GoModule } from './go-module';
import { loadWorkspacePackages, type WorkspacePackages } from './workspace-packages';
import { logDebug } from '../errors';
import { lexicalPathWithinRoot } from '../utils';
import type { ReExport } from './types';
import { LRUCache } from './lru-cache';
import { JS_BUILT_INS } from './js-builtins';

/** Node kinds that can declare supertypes (extends/implements). */
const SUPERTYPE_BEARING_KINDS = new Set<Node['kind']>([
  'class', 'struct', 'interface', 'trait', 'protocol', 'enum',
]);

/**
 * Languages whose chained static-factory/fluent calls defer to the conformance
 * second pass. Dotted-receiver languages resolve via matchDottedCallChain; the
 * `::`-receiver ones (Rust) via matchScopedCallChain.
 */
const CHAIN_LANGUAGES = new Set(['java', 'kotlin', 'csharp', 'swift', 'rust', 'go', 'scala', 'dart', 'objc', 'pascal']);
const SCOPED_CHAIN_LANGUAGES = new Set(['rust']);

/** The extractor's chained-receiver encoding: `<inner>().<method>`. */
const CHAIN_SHAPE = /^(.+)\(\)\.(\w+)$/;

/** PHP `$this->prop->method()` encoded as `this->prop.method` — no `()`, so CHAIN_SHAPE misses it. */
const PHP_PROP_SHAPE = /^this->\w+\.\w+$/;

/**
 * Cache size limits. Each per-resolver cache is bounded so memory
 * stays flat on large codebases (20k+ files). Sizes were chosen to
 * cover the working set for typical resolution batches without
 * exceeding a few hundred MB worst-case. Override via the env var
 * `CODEGRAPH_RESOLVER_CACHE_SIZE` (single integer applied to all
 * caches) when tuning for very large or very small projects.
 */
const DEFAULT_CACHE_LIMIT = 5_000;
function resolveCacheLimit(): number {
  const raw = process.env.CODEGRAPH_RESOLVER_CACHE_SIZE;
  if (!raw) return DEFAULT_CACHE_LIMIT;
  const parsed = Number.parseInt(raw, 10);
  if (Number.isFinite(parsed) && parsed > 0) return parsed;
  return DEFAULT_CACHE_LIMIT;
}

// Re-export types
export * from './types';

// Pre-built Sets for O(1) built-in lookups (allocated once, shared across all instances)
const REACT_HOOKS = new Set([
  'useState', 'useEffect', 'useContext', 'useReducer', 'useCallback',
  'useMemo', 'useRef', 'useLayoutEffect', 'useImperativeHandle', 'useDebugValue',
]);

const PYTHON_BUILT_INS = new Set([
  'print', 'len', 'range', 'str', 'int', 'float', 'list', 'dict', 'set', 'tuple',
  'open', 'input', 'type', 'isinstance', 'hasattr', 'getattr', 'setattr',
  'super', 'self', 'cls', 'None', 'True', 'False',
]);

const PYTHON_BUILT_IN_TYPES = new Set([
  'list', 'dict', 'set', 'tuple', 'str', 'int', 'float', 'bool',
  'bytes', 'bytearray', 'frozenset', 'object', 'super',
]);

const PYTHON_BUILT_IN_METHODS = new Set([
  'append', 'extend', 'insert', 'remove', 'pop', 'clear', 'sort', 'reverse', 'copy',
  'update', 'keys', 'values', 'items', 'get',
  'add', 'discard', 'union', 'intersection', 'difference',
  'split', 'join', 'strip', 'lstrip', 'rstrip', 'replace', 'lower', 'upper',
  'startswith', 'endswith', 'find', 'index', 'count', 'encode', 'decode',
  'format', 'isdigit', 'isalpha', 'isalnum',
  'read', 'write', 'readline', 'readlines', 'close', 'flush', 'seek',
]);

const GO_STDLIB_PACKAGES = new Set([
  'fmt', 'os', 'io', 'net', 'http', 'log', 'math', 'sort', 'sync',
  'time', 'path', 'bytes', 'strings', 'strconv', 'errors', 'context',
  'json', 'xml', 'csv', 'html', 'template', 'regexp', 'reflect',
  'runtime', 'testing', 'flag', 'bufio', 'crypto', 'encoding',
  'filepath', 'hash', 'mime', 'rand', 'signal', 'sql', 'syscall',
  'unicode', 'unsafe', 'atomic', 'binary', 'debug', 'exec', 'heap',
  'ring', 'scanner', 'tar', 'zip', 'gzip', 'zlib', 'tls', 'url',
  'user', 'pprof', 'trace', 'ast', 'build', 'parser', 'printer',
  'token', 'types', 'cgo', 'plugin', 'race', 'ioutil',
  // Kubernetes-common stdlib aliases
  'utilruntime', 'utilwait', 'utilnet',
]);

const GO_BUILT_INS = new Set([
  'make', 'new', 'len', 'cap', 'append', 'copy', 'delete', 'close',
  'panic', 'recover', 'print', 'println', 'complex', 'real', 'imag',
  'error', 'nil', 'true', 'false', 'iota',
  'int', 'int8', 'int16', 'int32', 'int64',
  'uint', 'uint8', 'uint16', 'uint32', 'uint64', 'uintptr',
  'float32', 'float64', 'complex64', 'complex128',
  'string', 'bool', 'byte', 'rune', 'any',
]);

const PASCAL_UNIT_PREFIXES = [
  'System.', 'Winapi.', 'Vcl.', 'Fmx.', 'Data.', 'Datasnap.',
  'Soap.', 'Xml.', 'Web.', 'REST.', 'FireDAC.', 'IBX.',
  'IdHTTP', 'IdTCP', 'IdSSL',
];

const PASCAL_BUILT_INS = new Set([
  'System', 'SysUtils', 'Classes', 'Types', 'Variants', 'StrUtils',
  'Math', 'DateUtils', 'IOUtils', 'Generics.Collections', 'Generics.Defaults',
  'Rtti', 'TypInfo', 'SyncObjs', 'RegularExpressions',
  'SysInit', 'Windows', 'Messages', 'Graphics', 'Controls', 'Forms',
  'Dialogs', 'StdCtrls', 'ExtCtrls', 'ComCtrls', 'Menus', 'ActnList',
  'WriteLn', 'Write', 'ReadLn', 'Read', 'Inc', 'Dec', 'Ord', 'Chr',
  'Length', 'SetLength', 'High', 'Low', 'Assigned', 'FreeAndNil',
  'Format', 'IntToStr', 'StrToInt', 'FloatToStr', 'StrToFloat',
  'Trim', 'UpperCase', 'LowerCase', 'Pos', 'Copy', 'Delete', 'Insert',
  'Now', 'Date', 'Time', 'DateToStr', 'StrToDate',
  'Raise', 'Exit', 'Break', 'Continue', 'Abort',
  'True', 'False', 'nil', 'Self', 'Result',
  'Create', 'Destroy', 'Free',
  'TObject', 'TComponent', 'TPersistent', 'TInterfacedObject',
  'TList', 'TStringList', 'TStrings', 'TStream', 'TMemoryStream', 'TFileStream',
  'Exception', 'EAbort', 'EConvertError', 'EAccessViolation',
  'IInterface', 'IUnknown',
]);

const C_BUILT_INS = new Set([
  // Standard C library functions
  'printf', 'fprintf', 'sprintf', 'snprintf', 'scanf', 'fscanf', 'sscanf',
  'malloc', 'calloc', 'realloc', 'free',
  'memcpy', 'memmove', 'memset', 'memcmp', 'memchr',
  'strlen', 'strcpy', 'strncpy', 'strcat', 'strncat', 'strcmp', 'strncmp',
  'strstr', 'strchr', 'strrchr', 'strtok', 'strdup',
  'fopen', 'fclose', 'fread', 'fwrite', 'fgets', 'fputs', 'fputc', 'fgetc',
  'feof', 'ferror', 'fflush', 'fseek', 'ftell', 'rewind',
  'exit', 'abort', 'atexit', 'atoi', 'atol', 'atof', 'strtol', 'strtoul', 'strtod',
  'qsort', 'bsearch',
  'abs', 'labs', 'rand', 'srand',
  'sin', 'cos', 'tan', 'sqrt', 'pow', 'log', 'log10', 'exp', 'ceil', 'floor', 'fabs',
  'time', 'clock', 'difftime', 'mktime', 'localtime', 'gmtime', 'strftime', 'asctime',
  'assert', 'errno',
  'perror', 'remove', 'rename', 'tmpfile', 'tmpnam',
  'getenv', 'system',
  'signal', 'raise',
  'setjmp', 'longjmp',
  'va_start', 'va_end', 'va_arg', 'va_copy',
  'NULL', 'EOF', 'BUFSIZ', 'FILENAME_MAX', 'RAND_MAX', 'EXIT_SUCCESS', 'EXIT_FAILURE',
  'size_t', 'ptrdiff_t', 'wchar_t', 'intptr_t', 'uintptr_t',
  'int8_t', 'int16_t', 'int32_t', 'int64_t',
  'uint8_t', 'uint16_t', 'uint32_t', 'uint64_t',
  'FILE',
  // POSIX additions commonly seen
  'stat', 'lstat', 'fstat', 'open', 'close', 'read', 'write', 'pipe',
  'fork', 'exec', 'waitpid', 'getpid', 'getppid', 'kill', 'sleep', 'usleep',
  'pthread_create', 'pthread_join', 'pthread_mutex_lock', 'pthread_mutex_unlock',
  'dlopen', 'dlsym', 'dlclose',
]);

const CPP_BUILT_INS = new Set([
  // iostream objects (often used without std:: prefix via using)
  'cout', 'cin', 'cerr', 'clog', 'endl', 'flush', 'ws',
  'std', // the namespace itself when used as std::something
  // Common C++ keywords that leak as references
  'nullptr', 'true', 'false', 'this', 'sizeof', 'alignof', 'typeid',
  'static_cast', 'dynamic_cast', 'reinterpret_cast', 'const_cast',
  'make_unique', 'make_shared', 'make_pair',
  'move', 'forward', 'swap',
]);

/**
 * Reference Resolver
 *
 * Orchestrates reference resolution using multiple strategies.
 */
export class ReferenceResolver {
  private projectRoot: string;
  private queries: QueryBuilder;
  private context: ResolutionContext;
  private frameworks: FrameworkResolver[] = [];
  // Chained static-factory/fluent call refs the first pass couldn't resolve,
  // collected in-memory and left pending in the DB until the post-pass
  // finishes, so a restart can recover the queue (#1577). Drained by
  // resolveChainedCallsViaConformance
  // once implements/extends edges exist, to resolve methods on a supertype the
  // receiver conforms to (#750).
  private deferredChainRefs: UnresolvedRef[] = [];
  // `this.<member>` function-as-value refs whose member is NOT on the
  // enclosing class itself — possibly inherited. Collected in-memory for the
  // same reason as deferredChainRefs and drained by
  // resolveDeferredThisMemberRefs once implements/extends edges exist (#808).
  private deferredThisMemberRefs: UnresolvedRef[] = [];
  private deferredRowIds = new Set<number>();
  // Per-`.razor`/`.cshtml`-file `@using` namespace set (own directives + folder
  // `_Imports.razor`, cascading to the project root). Used to disambiguate a
  // markup type ref to the right C# namespace.
  private razorUsingsCache = new Map<string, string[]>();
  // All per-resolver caches are LRU-bounded. Previously these were
  // unbounded Maps that grew with every distinct lookup and OOM'd on
  // codebases with 20k+ files (see issue: unbounded cache growth).
  private nodeCache: LRUCache<string, Node[]>; // per-file node cache
  private fileCache: LRUCache<string, string | null>; // per-file content cache
  private importMappingCache: LRUCache<string, ImportMapping[]>;
  private reExportCache: LRUCache<string, ReExport[]>;
  private nameCache: LRUCache<string, Node[]>; // name → nodes cache
  private lowerNameCache: LRUCache<string, Node[]>; // lower(name) → nodes cache
  private qualifiedNameCache: LRUCache<string, Node[]>; // qualified_name → nodes cache
  private fileLinesCache: LRUCache<string, string[] | null>; // file → split lines cache
  private methodMatchCache: LRUCache<string, Node[]>; // lang\0Type::method → matching method nodes
  // Per-(language, methodName) owner index for getMethodMatches: buckets a
  // method name's candidates by their qualifiedName's last two segments so a
  // (type, method) query is a lookup instead of an O(candidates) filter per
  // methodMatchCache miss. Derived purely from node rows (stable through the
  // resolution loop, same window nameCache relies on); dropped in clearCaches.
  private methodOwnerIndexCache = new Map<string, Map<string, Node[]>>();
  // Generation-tagged memo for getSupertypes. Supertype edges GROW during the
  // resolution loop (batch k persists its implements/extends edges BEFORE
  // batch k+1 fans out — the #1320 ordering), so a plain cache would freeze an
  // early batch's emptier answer and change later batches' outcomes. Within
  // one batch the edge state is fixed by that same ordering, so entries are
  // tagged with a generation that advances at every batch entry point
  // (resolveBatchYielding / resolveListForAdmission) — a stale-gen entry is
  // recomputed, making the memo behavior-identical to no memo at every point
  // in time. On the Swift compiler the unmemoized walk ran 971k times for
  // 565s of combined worker time (~581µs each, recursion-multiplied).
  private supertypeGen = 0;
  private supertypeMemo = new Map<string, { gen: number; supers: string[] }>();

  /** Invalidate the getSupertypes memo — call when resolved edges may have advanced. */
  private advanceSupertypeGeneration(): void {
    this.supertypeGen++;
    // Lazy invalidation via the gen tag; bound the map so a long run over many
    // batches doesn't accrete dead entries.
    if (this.supertypeMemo.size > 50_000) this.supertypeMemo.clear();
  }
  // Node kinds are a small fixed set (~24), so this is a plain Map, not an LRU.
  // getNodesByKind returns the FULL node list for a kind; it was previously
  // uncached — a per-ref `SELECT * FROM nodes WHERE kind=?` + row-mapping. Called
  // for every dotted call ref by the Spring resolver (constants) and every
  // `hook_` ref by the Drupal resolver (functions), that scan dominated
  // resolution on large repos (#1180). The node set is stable within a
  // resolution pass (same lifetime assumption as nameCache); clearCaches() resets
  // it between passes. Callers must treat the returned array as read-only.
  private nodesByKindCache = new Map<Node['kind'], Node[]>();
  private knownNames: Set<string> | null = null; // all known symbol names for fast pre-filtering
  private knownFiles: Set<string> | null = null;
  private cachesWarmed = false;
  // tsconfig/jsconfig path-alias map. `undefined` = not yet computed,
  // `null` = computed and absent. Treated as immutable for the
  // resolver's lifetime; callers re-create the resolver if config changes.
  private projectAliases: AliasMap | null | undefined = undefined;
  // go.mod module path. Same lazy/immutable convention as projectAliases.
  private goModule: GoModule | null | undefined = undefined;
  // Monorepo workspace member packages. Same lazy/immutable convention.
  private workspacePackages: WorkspacePackages | null | undefined = undefined;

  constructor(projectRoot: string, queries: QueryBuilder) {
    this.projectRoot = projectRoot;
    this.queries = queries;

    const limit = resolveCacheLimit();
    // The content cache is heavier (full file text), so we give it a
    // smaller budget than the metadata caches.
    const contentLimit = Math.max(64, Math.floor(limit / 5));
    this.nodeCache = new LRUCache(limit);
    this.fileCache = new LRUCache(contentLimit);
    this.importMappingCache = new LRUCache(limit);
    this.reExportCache = new LRUCache(limit);
    this.nameCache = new LRUCache(limit);
    this.lowerNameCache = new LRUCache(limit);
    this.qualifiedNameCache = new LRUCache(limit);
    // Split-lines arrays are heavier than content strings; refs arrive
    // file-ordered, so a small cache still hits nearly always.
    this.fileLinesCache = new LRUCache(contentLimit);
    this.methodMatchCache = new LRUCache(limit);

    this.context = this.createContext();
  }

  /**
   * Initialize the resolver (detect frameworks, etc.)
   */
  initialize(): void {
    this.frameworks = detectFrameworks(this.context);
    this.clearCaches();
  }

  /**
   * Run each framework resolver's cross-file finalization pass and persist
   * the returned node updates. Idempotent — safe to call after every indexAll
   * and every incremental sync. Returns the number of nodes updated.
   *
   * Caches are cleared before/after so the post-extract pass sees fresh DB
   * state and downstream queries see the updated names.
   */
  runPostExtract(): number {
    let updated = 0;
    this.clearCaches();
    for (const fw of this.frameworks) {
      if (!fw.postExtract) continue;
      try {
        const nodes = fw.postExtract(this.context);
        for (const node of nodes) {
          this.queries.updateNode(node);
          updated++;
        }
      } catch (err) {
        logDebug(`Framework '${fw.name}' postExtract failed`, {
          error: err instanceof Error ? err.message : String(err),
        });
      }
    }
    if (updated > 0) this.clearCaches();
    return updated;
  }

  /**
   * Pre-build lightweight caches for resolution.
   * Node lookups are now handled by indexed SQLite queries instead of
   * loading all nodes into memory (which caused OOM on large codebases).
   * We cache the set of known symbol names for fast pre-filtering.
   */
  warmCaches(): void {
    if (this.cachesWarmed) return;

    // Only cache the set of known file paths (lightweight string set)
    this.knownFiles = new Set(this.queries.getAllFilePaths());

    // Cache all distinct symbol names for fast pre-filtering (just strings, not full nodes)
    this.knownNames = new Set(this.queries.getAllNodeNames());

    this.cachesWarmed = true;
  }

  /**
   * warmCaches for the async resolution entry points: streams the distinct
   * name set with periodic yields instead of one synchronous `.all()`. On a
   * multi-million-node index the DISTINCT scan is a solid multi-second block
   * (measured up to 28s inside `codegraph sync` on the Linux kernel index),
   * long enough to matter to the #850 watchdog on slower hardware. Same
   * result, same memory — only the event loop keeps turning.
   */
  async warmCachesYielding(onYield: MaybeYield): Promise<void> {
    if (this.cachesWarmed) return;

    this.knownFiles = new Set(this.queries.getAllFilePaths());

    const names = new Set<string>();
    let scanned = 0;
    for (const name of this.queries.iterateNodeNames()) {
      names.add(name);
      if ((++scanned & 8191) === 0) await onYield();
    }
    this.knownNames = names;

    this.cachesWarmed = true;
  }

  /**
   * Clear internal caches
   */
  clearCaches(): void {
    this.nodeCache.clear();
    this.fileCache.clear();
    this.importMappingCache.clear();
    this.reExportCache.clear();
    this.nameCache.clear();
    this.lowerNameCache.clear();
    this.qualifiedNameCache.clear();
    this.fileLinesCache.clear();
    this.methodMatchCache.clear();
    this.methodOwnerIndexCache.clear();
    this.supertypeMemo.clear();
    this.supertypeGen++;
    this.nodesByKindCache.clear();
    this.knownNames = null;
    this.knownFiles = null;
    this.cachesWarmed = false;
    // The import-resolver's and name-matcher's per-context memos assume the
    // same stable window as the caches above — drop them together.
    if (this.context) {
      clearImportResolverMemos(this.context);
      clearNameMatcherMemos(this.context);
    }
  }

  /** `readFile` through the LRU content cache (null = read failed, also cached). */
  private readFileCached(filePath: string): string | null {
    if (this.fileCache.has(filePath)) {
      return this.fileCache.get(filePath)!;
    }
    const fullPath = path.join(this.projectRoot, filePath);
    try {
      const content = fs.readFileSync(fullPath, 'utf-8');
      this.fileCache.set(filePath, content);
      return content;
    } catch (error) {
      logDebug('Failed to read file for resolution', { filePath, error: String(error) });
      this.fileCache.set(filePath, null);
      return null;
    }
  }

  /**
   * Create the resolution context
   */
  private createContext(): ResolutionContext {
    return {
      getNodesInFile: (filePath: string) => {
        if (!this.nodeCache.has(filePath)) {
          this.nodeCache.set(filePath, this.queries.getNodesByFile(filePath));
        }
        return this.nodeCache.get(filePath)!;
      },

      getNodesByName: (name: string) => {
        const cached = this.nameCache.get(name);
        if (cached !== undefined) return cached;
        const result = this.queries.getNodesByName(name);
        this.nameCache.set(name, result);
        return result;
      },

      getMethodMatches: (typeName: string, methodName: string, language: Language) => {
        const key = `${language} ${typeName}::${methodName}`;
        const cached = this.methodMatchCache.get(key);
        if (cached !== undefined) return cached;
        let candidates = this.nameCache.get(methodName);
        if (candidates === undefined) {
          candidates = this.queries.getNodesByName(methodName);
          this.nameCache.set(methodName, candidates);
        }
        const want = `${typeName}::${methodName}`;
        let matches: Node[];
        if (typeName.includes('::') || methodName.includes(':')) {
          // Legacy linear filter for the shapes the owner index below can't
          // key exactly: a multi-segment typeName (the endsWith test then
          // spans more than two `::` segments) and ObjC selectors (whose
          // single/empty-keyword colons defeat the segment split). Tiny
          // populations; the per-key memo above still amortizes them.
          matches = [];
          for (const m of candidates) {
            if (m.kind !== 'method') continue;
            if (m.language !== language) continue;
            const qn = m.qualifiedName;
            if (qn === want || qn.endsWith(`::${want}`)) matches.push(m);
          }
        } else {
          // Owner index: the linear filter above is O(all same-named methods)
          // per CACHE MISS, and on overload-heavy landscapes the distinct
          // (type, method) key space is so large the per-key memo never
          // amortizes — Swift's `init` has tens of thousands of candidates
          // and the compiler repo measured 732µs per failing call, most of it
          // this scan (re-entered once per supertype recursion level, too).
          // Bucket each (language, methodName)'s candidates ONCE by the
          // qualifiedName's last two `::` segments — exactly the span the
          // `qn === want || qn.endsWith('::' + want)` predicate tests for a
          // segment-clean typeName — then every query is a map lookup.
          // Bucket insertion follows candidate order, so each bucket is
          // byte-identical to what the linear filter produced.
          const idxKey = `${language} ${methodName}`;
          let ownerIndex = this.methodOwnerIndexCache.get(idxKey);
          if (!ownerIndex) {
            ownerIndex = new Map<string, Node[]>();
            for (const m of candidates) {
              if (m.kind !== 'method') continue;
              if (m.language !== language) continue;
              const qn = m.qualifiedName;
              const i2 = qn.lastIndexOf('::');
              if (i2 < 0) continue; // single-segment qn can never match `T::m`
              const i1 = qn.lastIndexOf('::', i2 - 1);
              const bucketKey = i1 < 0 ? qn : qn.slice(i1 + 2);
              const bucket = ownerIndex.get(bucketKey);
              if (bucket) bucket.push(m);
              else ownerIndex.set(bucketKey, [m]);
            }
            this.methodOwnerIndexCache.set(idxKey, ownerIndex);
          }
          matches = ownerIndex.get(want) ?? [];
        }
        this.methodMatchCache.set(key, matches);
        return matches;
      },

      getNodesByQualifiedName: (qualifiedName: string) => {
        const cached = this.qualifiedNameCache.get(qualifiedName);
        if (cached !== undefined) return cached;
        const result = this.queries.getNodesByQualifiedNameExact(qualifiedName);
        this.qualifiedNameCache.set(qualifiedName, result);
        return result;
      },

      getNodesByKind: (kind: Node['kind']) => {
        const cached = this.nodesByKindCache.get(kind);
        if (cached !== undefined) return cached;
        const result = this.queries.getNodesByKind(kind);
        this.nodesByKindCache.set(kind, result);
        return result;
      },

      // Streamed, uncached — synthesizers scan-and-filter whole kinds, and
      // both the materialized array AND the per-kind cache retention are
      // O(nodes) memory (#1212). Per-ref resolvers keep the cached array
      // variant above.
      iterateNodesByKind: (kind: Node['kind']) => this.queries.iterateNodesByKind(kind),

      fileExists: (filePath: string) => {
        // Check pre-built known files set first (O(1))
        if (this.knownFiles) {
          const normalized = filePath.replace(/\\/g, '/');
          if (this.knownFiles.has(filePath) || this.knownFiles.has(normalized)) {
            return true;
          }
        }
        // Fall back to filesystem for files not yet indexed. `path.join` does
        // not clamp, and relative-import resolution hands us paths carrying
        // `../` segments, so the probe has to be contained (#1631): a path
        // outside the root can never be an indexed project file, and the
        // `knownFiles` check above already answered for everything that is.
        // Lexical containment only: this is a per-candidate hot path, and the
        // symlink half of `validatePathWithinRoot` costs two `realpathSync`
        // calls per probe (~70x slower here). It would also be wrong to apply
        // — indexing deliberately follows in-root symlinks whose targets live
        // outside the root (#935), so only the `../` escape is refused.
        const fullPath = lexicalPathWithinRoot(this.projectRoot, filePath);
        if (fullPath === null) return false;
        try {
          return fs.existsSync(fullPath);
        } catch (error) {
          logDebug('Error checking file existence', { filePath, error: String(error) });
          return false;
        }
      },

      readFile: (filePath: string) => this.readFileCached(filePath),

      getFileLines: (filePath: string) => {
        const cached = this.fileLinesCache.get(filePath);
        if (cached !== undefined) return cached;
        const source = this.readFileCached(filePath);
        const lines = source === null ? null : source.split(/\r?\n/);
        this.fileLinesCache.set(filePath, lines);
        return lines;
      },

      getProjectRoot: () => this.projectRoot,

      getAllFiles: () => {
        return this.queries.getAllFilePaths();
      },

      listDirectories: (relativePath: string) => {
        const target = relativePath === '.' || relativePath === ''
          ? this.projectRoot
          : path.join(this.projectRoot, relativePath);
        try {
          return fs
            .readdirSync(target, { withFileTypes: true })
            .filter((entry) => entry.isDirectory())
            .map((entry) => entry.name);
        } catch (error) {
          logDebug('Failed to list directory for resolution', {
            relativePath,
            error: String(error),
          });
          return [];
        }
      },

      getNodesByLowerName: (lowerName: string) => {
        const cached = this.lowerNameCache.get(lowerName);
        if (cached !== undefined) return cached;
        const result = this.queries.getNodesByLowerName(lowerName);
        this.lowerNameCache.set(lowerName, result);
        return result;
      },

      getNodeById: (id: string) => {
        return this.queries.getNodeById(id);
      },

      getSupertypes: (typeName: string, language) => {
        // Union the `implements`/`extends` targets of every same-named type node.
        // Matching by simple name (not id) reconciles a type declared in one node
        // (`KF::Builder`) with conformance declared in a separate extension node
        // (`KF.Builder: KFOptionSetter`) — both have name `Builder`.
        // Memoized per batch generation (see supertypeMemo): within a batch the
        // edge state is fixed, and the conformance walk re-queries the same
        // popular supertypes (Swift stdlib protocols especially) thousands of
        // times per batch.
        const memoKey = `${language} ${typeName}`;
        const hit = this.supertypeMemo.get(memoKey);
        if (hit && hit.gen === this.supertypeGen) return hit.supers;
        const typeNodes = this.context
          .getNodesByName(typeName)
          .filter((n) => SUPERTYPE_BEARING_KINDS.has(n.kind) && n.language === language);
        let supers: string[];
        if (typeNodes.length === 0) {
          supers = [];
        } else {
          const supertypes = new Set<string>();
          for (const tn of typeNodes) {
            for (const edge of this.queries.getOutgoingEdges(tn.id, ['implements', 'extends'])) {
              const target = this.queries.getNodeById(edge.target);
              if (target?.name && target.name !== typeName) supertypes.add(target.name);
            }
          }
          supers = [...supertypes];
        }
        this.supertypeMemo.set(memoKey, { gen: this.supertypeGen, supers });
        return supers;
      },

      getImportMappings: (filePath: string, language) => {
        const cacheKey = filePath;
        const cached = this.importMappingCache.get(cacheKey);
        if (cached) return cached;

        const content = this.context.readFile(filePath);
        if (!content) {
          this.importMappingCache.set(cacheKey, []);
          return [];
        }

        const mappings = extractImportMappings(filePath, content, language);
        this.importMappingCache.set(cacheKey, mappings);
        return mappings;
      },

      getProjectAliases: () => {
        if (this.projectAliases === undefined) {
          this.projectAliases = loadProjectAliases(this.projectRoot);
        }
        return this.projectAliases;
      },

      getGoModule: () => {
        if (this.goModule === undefined) {
          this.goModule = loadGoModule(this.projectRoot);
        }
        return this.goModule;
      },

      getWorkspacePackages: () => {
        if (this.workspacePackages === undefined) {
          this.workspacePackages = loadWorkspacePackages(this.projectRoot);
        }
        return this.workspacePackages;
      },

      getReExports: (filePath: string, language) => {
        const cached = this.reExportCache.get(filePath);
        if (cached) return cached;
        const content = this.context.readFile(filePath);
        if (!content) {
          this.reExportCache.set(filePath, []);
          return [];
        }
        // Re-exports are a JS/TS-only construct, and what matters is the
        // BARREL file's own language — not the consuming reference's. A
        // `.svelte`/`.vue` consumer threads its own language down the
        // re-export chase, which would make extractReExports() bail on a
        // `.ts` index barrel and silently break the chain (#629). Re-key
        // the parse on the barrel's extension so the chase works no matter
        // what kind of file imports through it.
        const isJsFamily = /\.(?:d\.ts|[cm]?tsx?|[cm]?jsx?|ets)$/i.test(filePath);
        const reExports = extractReExports(content, isJsFamily ? 'typescript' : language);
        this.reExportCache.set(filePath, reExports);
        return reExports;
      },

      getCppIncludeDirs: () => {
        return loadCppIncludeDirs(this.projectRoot);
      },
    };
  }

  /**
   * Resolve all unresolved references
   */
  resolveAll(
    unresolvedRefs: UnresolvedReference[],
    onProgress?: (current: number, total: number) => void
  ): ResolutionResult {
    // Pre-load all nodes into memory for fast lookups
    this.warmCaches();
    this.advanceSupertypeGeneration();

    const resolved: ResolvedRef[] = [];
    const unresolved: UnresolvedRef[] = [];
    const byMethod: Record<string, number> = {};

    // Convert to our internal format, using denormalized fields when available
    const refs: UnresolvedRef[] = unresolvedRefs.map((ref) => ({
      fromNodeId: ref.fromNodeId,
      referenceName: ref.referenceName,
      referenceKind: ref.referenceKind,
      line: ref.line,
      column: ref.column,
      filePath: ref.filePath || this.getFilePathFromNodeId(ref.fromNodeId),
      language: ref.language || this.getLanguageFromNodeId(ref.fromNodeId),
      rowId: ref.rowId,
    }));

    const total = refs.length;
    let lastReportedPercent = -1;

    for (let i = 0; i < refs.length; i++) {
      const ref = refs[i]!; // Array index is guaranteed to be in bounds
      const result = this.resolveOneTimed(ref);

      if (result) {
        resolved.push(result);
        byMethod[result.resolvedBy] = (byMethod[result.resolvedBy] || 0) + 1;
      } else {
        unresolved.push(ref);
      }

      // Report progress every 1% to avoid too many updates
      if (onProgress) {
        const currentPercent = Math.floor((i / total) * 100);
        if (currentPercent > lastReportedPercent) {
          lastReportedPercent = currentPercent;
          onProgress(i + 1, total);
        }
      }
    }

    // Final progress report
    if (onProgress && total > 0) {
      onProgress(total, total);
    }

    return {
      resolved,
      unresolved,
      stats: {
        total: refs.length,
        resolved: resolved.length,
        unresolved: unresolved.length,
        byMethod,
      },
    };
  }

  /**
   * Check if a reference name has any possible match in the codebase.
   * Uses the pre-built knownNames set to skip expensive resolution
   * for names that definitely don't exist as symbols.
   */
  private hasAnyPossibleMatch(name: string): boolean {
    if (!this.knownNames) return true; // no pre-filter available

    // Direct name match
    if (this.knownNames.has(name)) return true;

    // For qualified names like "obj.method" or "Class::method", check the parts
    const dotIdx = name.indexOf('.');
    if (dotIdx > 0) {
      const receiver = name.substring(0, dotIdx);
      const member = name.substring(dotIdx + 1);
      if (this.knownNames.has(receiver) || this.knownNames.has(member)) return true;
      // Also check capitalized receiver (instance-method resolution)
      const capitalized = receiver.charAt(0).toUpperCase() + receiver.slice(1);
      if (this.knownNames.has(capitalized)) return true;
      // JVM FQN: `com.example.foo.Bar` — the only useful segment is the
      // last one (`Bar`); the earlier check finds `example.foo.Bar` which
      // never matches a node name.
      const lastDot = name.lastIndexOf('.');
      if (lastDot > dotIdx) {
        const tail = name.substring(lastDot + 1);
        if (tail && this.knownNames.has(tail)) return true;
      }
    }
    const colonIdx = name.indexOf('::');
    if (colonIdx > 0) {
      const receiver = name.substring(0, colonIdx);
      const member = name.substring(colonIdx + 2);
      if (this.knownNames.has(receiver) || this.knownNames.has(member)) return true;
      // Multi-segment path `a::b::c` (a Rust/C++ module call like
      // `database::profiles::find`) — the only segment that names a symbol is
      // the last (`c`); `member` above is `b::c`, which never matches a node
      // name, so without this the pre-filter drops the ref before the Rust path
      // resolver ever sees it. Mirror the dotted-name leaf check above.
      const lastColon = name.lastIndexOf('::');
      if (lastColon > colonIdx) {
        const tail = name.substring(lastColon + 2);
        if (tail && this.knownNames.has(tail)) return true;
      }
    }

    // Lua/Luau method calls use a single `:` (`lg:log`); R uses `$` (`lg$log`).
    // Check the member (and receiver) around these separators too, so the ref
    // isn't dropped here before the method-call resolver ever sees it. The `:`
    // case is skipped when the name actually contains `::` (handled above).
    for (const sep of [':', '$']) {
      if (sep === ':' && name.includes('::')) continue;
      const sepIdx = name.indexOf(sep);
      if (sepIdx > 0) {
        const receiver = name.substring(0, sepIdx);
        const member = name.substring(sepIdx + 1);
        if (this.knownNames.has(member) || this.knownNames.has(receiver)) return true;
        const capitalized = receiver.charAt(0).toUpperCase() + receiver.slice(1);
        if (this.knownNames.has(capitalized)) return true;
      }
    }

    // For path-like references (e.g., "snippets/drawer-menu.liquid"), check the filename
    const slashIdx = name.lastIndexOf('/');
    if (slashIdx > 0) {
      const fileName = name.substring(slashIdx + 1);
      if (this.knownNames.has(fileName)) return true;
    }

    return false;
  }

  /**
   * Does `ref.referenceName` match an import declared in its containing
   * file? Used as a pre-filter escape so re-export chain resolution
   * still gets a chance when the name has no project-wide declaration.
   */
  private matchesAnyImport(ref: UnresolvedRef): boolean {
    const imports = this.context.getImportMappings(ref.filePath, ref.language);
    if (imports.length === 0) return false;
    for (const imp of imports) {
      if (
        imp.localName === ref.referenceName ||
        ref.referenceName.startsWith(imp.localName + '.')
      ) {
        return true;
      }
    }
    return false;
  }

  /**
   * Resolve a single reference
   */
  resolveOne(ref: UnresolvedRef): ResolvedRef | null {
    // Skip built-in/external references
    if (this.isBuiltInOrExternal(ref)) {
      return null;
    }

    // CFML component paths in inheritance (#1152): `extends="coldbox.system.web.
    // Controller"` names the supertype by its dot-separated path (or `extends=
    // "../base"` by relative file path) — the graph indexes the class under its
    // final segment only, so these die at the fast pre-filter below and never
    // resolved. Handled by a dedicated path-corroborated matcher, gated to
    // inheritance refs only (a dotted `calls` ref is a member-access chain, not
    // a component path). No fallthrough on miss: the full path string can only
    // ever mis-match downstream, and an unresolvable supertype usually lives in
    // an out-of-repo library (mxunit, testbox) — silent beats wrong.
    if (
      (ref.language === 'cfml' || ref.language === 'cfscript') &&
      (ref.referenceKind === 'extends' || ref.referenceKind === 'implements') &&
      (ref.referenceName.includes('.') || ref.referenceName.includes('/'))
    ) {
      return this.resolveCfmlComponentPath(ref);
    }

    // Fast pre-filter: skip if no symbol with this name exists anywhere
    // AND the name doesn't match a local import. The import escape is
    // necessary because re-export rename chains (`import { login }
    // from './barrel'` where the barrel has `export { signIn as login }
    // from './auth'`) intentionally call a name that has no
    // declaration anywhere — only the renamed upstream symbol does.
    // ArkTS chained-attribute refs carry a leading dot (`.titleStyle`) that
    // routes them to the decorator-gated matcher; the symbol itself is
    // indexed under the bare name, so the existence check strips the dot.
    // Nix static path imports (`import ./x.nix`) name a FILE, not a symbol —
    // they bypass the symbol-existence check and resolve via resolveViaImport.
    let existenceName =
      ref.language === 'arkts' && ref.referenceName.startsWith('.')
        ? ref.referenceName.slice(1)
        : ref.referenceName;
    // Erlang refs carry the call-site arity (`f/1`, `mod::f/2` — #1610); the
    // name index stores bare names, so existence is checked arity-less.
    if (ref.language === 'erlang') existenceName = existenceName.replace(/\/\d{1,3}$/, '');
    const tPre = this.profileStages ? process.hrtime.bigint() : 0n;
    const preFilterPass =
      isNixPathImportRef(ref) ||
      this.hasAnyPossibleMatch(existenceName) ||
      this.matchesAnyImport(ref) ||
      this.frameworks.some((f) => f.claimsReference?.(ref.referenceName));
    if (this.profileStages) this.stageAdd('preFilter', ref, preFilterPass, tPre);
    if (!preFilterPass) {
      return null;
    }

    // Function-as-value refs (#756) get a dedicated, strictly-gated path:
    // import-based resolution first (an imported callback resolves through its
    // import, the most precise cross-file signal), then matchFunctionRef
    // (same-file first, unique-only cross-file, function/method targets only).
    // They never reach the framework or fuzzy strategies below.
    if (ref.referenceKind === 'function_ref') {
      // `this.<member>` values (TS/JS) resolve ONLY against the enclosing
      // class's own members — never a same-named symbol elsewhere.
      if (ref.referenceName.startsWith('this.')) {
        return this.gateLanguage(this.resolveThisMemberFnRef(ref), ref);
      }
      const viaImport = this.gateLanguage(resolveViaImport(ref, this.context), ref);
      if (viaImport) {
        const target = this.queries.getNodeById(viaImport.targetNodeId);
        if (
          target &&
          (target.kind === 'function' ||
            target.kind === 'method' ||
            // Python (#1478): an imported class used as a value (`return
            // OrgSerializerFull`) resolves through its import like any
            // callback — mirrors matchFunctionRef's bareClassOk.
            (ref.language === 'python' && target.kind === 'class'))
        ) {
          return viaImport;
        }
      }
      return this.gateLanguage(matchFunctionRef(ref, this.context), ref);
    }

    // JVM FQN imports skip framework/name-matcher: `import com.example.Bar`
    // resolves directly through the qualifiedName index, which is unambiguous
    // even when several `Bar` classes exist in different packages.
    const tJvm = this.profileStages ? process.hrtime.bigint() : 0n;
    const jvmImport = resolveJvmImport(ref, this.context);
    if (this.profileStages) this.stageAdd('jvmImport', ref, !!jvmImport, tJvm);
    if (jvmImport) return jvmImport;

    // Razor/Blazor: a markup or `@code` type ref resolves through the file's
    // `@using` namespaces (incl. folder `_Imports.razor`). This precisely
    // disambiguates a simple name that exists in several namespaces — e.g.
    // `CatalogBrand` resolving to `BlazorShared.Models::CatalogBrand` (the DTO,
    // which the `.razor` `@using`s) rather than the same-named domain entity.
    if (ref.language === 'razor') {
      const razorResult = this.resolveRazorUsing(ref);
      if (razorResult) return razorResult;
    }

    // An explicit PHP class import owns its static calls, including an
    // unavailable method. Do not let same-name fallbacks change the receiver
    // to an unrelated Service/Repository type (#1545).
    const phpStaticImport = resolvePhpImportedStaticCall(ref, this.context);
    if (phpStaticImport !== undefined) return this.gateLanguage(phpStaticImport, ref);

    const candidates: ResolvedRef[] = [];

    // Strategy 1: Try framework-specific resolution. Cross-language bridges
    // are deliberately preserved (Drupal `routing.yml` → PHP controller, RN
    // JS → native `calls`) — `gateFrameworkLanguage` only drops a type/import
    // edge between two KNOWN families (see its doc), never a `calls` bridge or
    // a config↔code edge.
    const tFw = this.profileStages ? process.hrtime.bigint() : 0n;
    let fwEarly: ResolvedRef | null = null;
    for (const framework of this.frameworks) {
      const result = this.gateFrameworkLanguage(framework.resolve(ref, this.context), ref);
      if (result) {
        if (result.confidence >= 0.9) {
          fwEarly = result; // High confidence, return immediately (below)
          break;
        }
        candidates.push(result);
      }
    }
    if (this.profileStages) this.stageAdd('frameworks', ref, fwEarly !== null, tFw);
    if (fwEarly) return fwEarly;

    // Strategy 2: Try import-based resolution
    // A TS/JS/Python call-receiver chain (`useStore.getState().reset`, #1683)
    // names the ROOT's import, not the method's: letting resolveViaImport see
    // it binds the call to the imported store constant and the method is
    // never looked up. The name-matcher owns the chain shape for these
    // languages — the Java/Kotlin/C++ chains keep their existing path.
    if (
      ref.referenceKind === 'calls' &&
      CHAIN_SHAPE.test(ref.referenceName) &&
      (ref.language === 'typescript' || ref.language === 'javascript' || ref.language === 'tsx' || ref.language === 'jsx' || ref.language === 'python')
    ) {
      return this.gateLanguage(matchReference(ref, this.context), ref);
    }

    const tImp = this.profileStages ? process.hrtime.bigint() : 0n;
    const importResult = this.gateLanguage(resolveViaImport(ref, this.context), ref);
    if (this.profileStages) this.stageAdd('viaImport', ref, !!importResult, tImp);
    if (importResult) {
      if (importResult.confidence >= 0.9) return importResult;
      candidates.push(importResult);
    }

    // PHP include/require paths resolve to files via import resolution only.
    // If that didn't find the file, do NOT fall back to the symbol
    // name-matcher — it would mis-connect e.g. "inc/db.php" to an unrelated
    // db.php elsewhere in the tree (a wrong edge is worse than none, #660).
    // Terraform refs are directory-scoped by language semantics — the
    // framework resolver IS the whole rulebook (`var.X` can never legally
    // bind outside its module directory), so the name-matcher's
    // qualified-name fallback would only ever add wrong cross-module edges.
    // Nix static path imports are file references for the same reason —
    // falling through would let "./x.nix" name-match an unrelated node.
    if (isPhpIncludePathRef(ref) || isCobolCopybookRef(ref) || isNixPathImportRef(ref) || ref.language === 'terraform') {
      return candidates.length > 0
        ? candidates.reduce((best, curr) =>
            curr.confidence > best.confidence ? curr : best
          )
        : null;
    }

    // Strategy 3: Try name matching
    const tName = this.profileStages ? process.hrtime.bigint() : 0n;
    let nameResult = this.gateLanguage(matchReference(ref, this.context), ref);
    if (this.profileStages) this.stageAdd('nameMatch', ref, !!nameResult, tName);
    // Nix has no ambient cross-file namespace — a callee binds lexically
    // (same file) or through explicit import/callPackage wiring (the import
    // path above). A cross-file name match is wrong by construction: every
    // module `inherit (lib) mkOption`s the same nixpkgs helpers, so the
    // matcher would link each `mkOption` call to whichever file's inherit
    // binding it happened to pick. Same-file matches only.
    if (nameResult) {
      const target = this.queries.getNodeById(nameResult.targetNodeId);
      // A definition its language makes file-local — a C `static`, a Kotlin
      // `private fun`, a Go unexported name in another package, a Rust
      // non-`pub` item outside its module subtree — cannot be what a name in
      // another file means, whichever strategy chose it (#1730).
      if (target && !isVisibleAcrossFiles(target, ref, this.context)) {
        nameResult = null;
      } else if (ref.language === 'nix') {
        if (!target || target.filePath !== ref.filePath) {
          nameResult = null;
        }
      } else if (target && target.language === 'nix') {
        // The reverse direction is just as impossible: no other language can
        // symbolically call into a .nix binding (interop is eval/CLI, never a
        // linkable symbol) — without this, a Python script's `split()` lands
        // on some module's `split = ...` binding as a low-confidence match.
        nameResult = null;
      }
    }
    if (nameResult) {
      candidates.push(nameResult);
    }

    if (candidates.length === 0) {
      // Defer a chained static-factory/fluent call the first pass couldn't
      // resolve — its method may live on a supertype the receiver conforms to,
      // resolvable once implements/extends edges exist (the conformance pass).
      if (
        ref.referenceKind === 'calls' &&
        CHAIN_LANGUAGES.has(ref.language) &&
        CHAIN_SHAPE.test(ref.referenceName)
      ) {
        this.deferReference(ref, this.deferredChainRefs);
      } else if (
        // PHP `$this->prop->method()` (encoded `this->prop.method`): its method
        // may live on the property's declared supertype, resolvable only once
        // implements/extends edges exist — defer to the same conformance pass.
        ref.referenceKind === 'calls' &&
        ref.language === 'php' &&
        PHP_PROP_SHAPE.test(ref.referenceName)
      ) {
        this.deferReference(ref, this.deferredChainRefs);
      }
      return null;
    }

    // Return highest confidence candidate
    return candidates.reduce((best, curr) =>
      curr.confidence > best.confidence ? curr : best
    );
  }

  /**
   * Create edges from resolved references
   */
  createEdges(resolved: ResolvedRef[]): Edge[] {
    return resolved.flatMap((ref) => {
      // `function_ref` (#756) is internal-only: it persists as a `references`
      // edge (the registration site depends on the callback), distinguishable
      // by metadata.resolvedBy === 'function-ref'. callers/impact already
      // traverse `references`, so registration sites surface with no
      // graph-layer changes.
      let kind: Edge['kind'] =
        ref.edgeKind ??
        (ref.original.referenceKind === 'function_ref' ? 'references' : ref.original.referenceKind);

      // Promote "extends" to "implements" when a class/struct targets an interface
      if (kind === 'extends') {
        const targetNode = this.queries.getNodeById(ref.targetNodeId);
        if (targetNode && (targetNode.kind === 'interface' || targetNode.kind === 'protocol')) {
          const sourceNode = this.queries.getNodeById(ref.original.fromNodeId);
          if (sourceNode && sourceNode.kind !== 'interface' && sourceNode.kind !== 'protocol') {
            kind = 'implements';
          }
        }
      }

      // Promote "calls" to "instantiates" when the resolved target is a
      // class/struct/union. Languages without a `new` keyword (Python, Ruby)
      // express instantiation as `Foo()` — extraction can't tell that
      // apart from a function call without symbol info, but resolution
      // can: if `Foo` resolves to a class, the call IS an instantiation.
      if (kind === 'calls') {
        const targetNode = this.queries.getNodeById(ref.targetNodeId);
        if (
          targetNode &&
          (targetNode.kind === 'class' || targetNode.kind === 'struct' || targetNode.kind === 'union')
        ) {
          kind = 'instantiates';
        }
      }

      // One reference can name several targets — a navigation whose
      // destination is a conditional reaches every arm. Each becomes its own
      // edge, sharing this resolution's kind and confidence.
      const targets = [
        { targetNodeId: ref.targetNodeId, metadata: ref.metadata },
        ...(ref.alsoTargets ?? []),
      ];
      return targets.map((t) => ({
        source: ref.original.fromNodeId,
        target: t.targetNodeId,
        kind,
        line: ref.original.line,
        column: ref.original.column,
        metadata: {
          ...(t.metadata ?? {}),
          confidence: ref.confidence,
          resolvedBy: ref.resolvedBy,
          // The ORIGINAL reference text (and kind, when edge-kind promotion
          // rewrote it — calls→instantiates, extends→implements,
          // function_ref→references). If this edge's target is later removed
          // by a re-index, the edge is resurrected as exactly this ref and
          // re-resolved (#1240 removal case) — a faithful resurrection, so
          // re-resolution can never bind anywhere a full re-index wouldn't.
          // Reconstruction from the target node's name instead would strip
          // receiver/qualifier context (`h.greet` → `greet`) and risk a
          // wrong rebind; edges without refName (pre-#1240, synthesized) are
          // deliberately NOT resurrected for the same reason.
          refName: ref.original.referenceName,
          ...(ref.original.referenceKind !== kind ? { refKind: ref.original.referenceKind } : {}),
          // Uniform marker for function-as-value edges (#756), regardless of
          // which strategy resolved them (import vs matchFunctionRef) — lets
          // tooling label "callback registration" and lets validation diff
          // exactly the edges this feature added.
          ...(ref.original.referenceKind === 'function_ref' ? { fnRef: true } : {}),
        },
      }));
    });
  }

  /**
   * Split resolved refs into rows deletable by id and hand-built refs that
   * must fall back to the key-tuple delete. Rows loaded from the database
   * carry their row id and are deleted by exactly that id; the key tuple
   * omits line/col, so it also removes SIBLING rows — the same caller calling
   * the same callee at other lines — that a later batch hadn't attempted yet:
   * when a batch boundary split a caller's same-named call sites, the later
   * sites' edges were silently never created (#1269).
   */
  private static partitionResolvedCleanup(resolved: ResolvedRef[]): {
    rowIds: number[];
    legacyKeys: Array<{ fromNodeId: string; referenceName: string; referenceKind: string }>;
  } {
    const rowIds: number[] = [];
    const legacyKeys: Array<{ fromNodeId: string; referenceName: string; referenceKind: string }> = [];
    for (const r of resolved) {
      if (r.original.rowId != null) rowIds.push(r.original.rowId);
      else legacyKeys.push({
        fromNodeId: r.original.fromNodeId,
        referenceName: r.original.referenceName,
        referenceKind: r.original.referenceKind,
      });
    }
    return { rowIds, legacyKeys };
  }

  /**
   * Same row-id precision for parking unresolvable refs as status='failed'
   * (#1240): the key-tuple fallback would flip same-key sibling rows in later
   * batches to 'failed' before they were ever attempted, and resolution
   * outcome can differ per call site (receiver-type inference reads the
   * ref's line), so a sibling must not inherit this row's failure (#1269).
   */
  private static partitionFailedCleanup(unresolved: UnresolvedRef[]): {
    byRowId: Array<{ rowId: number; referenceName: string }>;
    legacyKeys: Array<{ fromNodeId: string; referenceName: string; referenceKind: string }>;
  } {
    const byRowId: Array<{ rowId: number; referenceName: string }> = [];
    const legacyKeys: Array<{ fromNodeId: string; referenceName: string; referenceKind: string }> = [];
    for (const r of unresolved) {
      if (r.rowId != null) byRowId.push({ rowId: r.rowId, referenceName: r.referenceName });
      else legacyKeys.push({
        fromNodeId: r.fromNodeId,
        referenceName: r.referenceName,
        referenceKind: r.referenceKind,
      });
    }
    return { byRowId, legacyKeys };
  }

  /** A deferred attempt is unfinished work, not a final failure (#1577). */
  private nonDeferredFailures(unresolved: UnresolvedRef[]): UnresolvedRef[] {
    return unresolved.filter((ref) => ref.rowId == null || !this.deferredRowIds.has(ref.rowId));
  }

  private deferReference(ref: UnresolvedRef, queue: UnresolvedRef[]): void {
    queue.push(ref);
    if (ref.rowId != null) this.deferredRowIds.add(ref.rowId);
  }

  /**
   * Resolve and persist edges to database
   */
  resolveAndPersist(
    unresolvedRefs: UnresolvedReference[],
    onProgress?: (current: number, total: number) => void
  ): ResolutionResult {
    const prerequisites = unresolvedRefs.filter(ReferenceResolver.isPrerequisite);
    if (prerequisites.length > 0 && prerequisites.length < unresolvedRefs.length) {
      const first = this.resolveAndPersist(prerequisites, (current) => onProgress?.(current, unresolvedRefs.length));
      const rest = this.resolveAndPersist(
        unresolvedRefs.filter((ref) => !ReferenceResolver.isPrerequisite(ref)),
        (current) => onProgress?.(prerequisites.length + current, unresolvedRefs.length)
      );
      return ReferenceResolver.mergeResults(first, rest);
    }
    const result = this.resolveAll(unresolvedRefs, onProgress);

    // Create edges from resolved references
    const edges = this.createEdges(result.resolved);

    // Insert edges into database
    if (edges.length > 0) {
      this.queries.insertEdges(edges);
    }

    // Clean up resolved refs from unresolved_refs table so metrics are accurate
    if (result.resolved.length > 0) {
      const { rowIds, legacyKeys } = ReferenceResolver.partitionResolvedCleanup(result.resolved);
      this.queries.deleteReferencesByRowIds(rowIds);
      this.queries.deleteSpecificResolvedReferences(legacyKeys);
    }

    // Park unresolvable refs as status='failed' — parity with
    // resolveAndPersistBatched. Deleting them was wrong (#1240): a ref whose
    // own file never changes is otherwise gone forever, so when a DIFFERENT
    // file later gains the export/symbol that would satisfy it, no sync can
    // recreate the edge — only a full re-index. Failed rows are excluded from
    // the pending readers, which preserves the #1187 orphan sweep's
    // invariant in status form: after a COMPLETED pass nothing it processed
    // is still 'pending', so any pending row at rest belongs to an
    // interrupted run and the sweep can key off the pending count.
    if (result.unresolved.length > 0) {
      const { byRowId, legacyKeys } = ReferenceResolver.partitionFailedCleanup(this.nonDeferredFailures(result.unresolved));
      this.queries.markReferencesFailedByRowIds(byRowId);
      this.queries.markReferencesFailed(legacyKeys);
    }

    return result;
  }

  /**
   * Yielding counterpart of {@link resolveAndPersist} for a caller-supplied
   * ref list — used by sync's failed-ref retry pass (#1240). Same persistence
   * semantics: resolved refs become edges and their rows are deleted;
   * still-unresolvable refs are (re-)marked failed (a no-op for rows already
   * in that status). Yields per-ref because sync can run on the daemon's
   * liveness-watchdog thread (#850/#1091) and a retry set is unbounded when
   * a large edit lands many popular symbol names at once.
   */
  async resolveAndPersistListYielding(refs: UnresolvedReference[]): Promise<ResolutionResult> {
    const prerequisites = refs.filter(ReferenceResolver.isPrerequisite);
    if (prerequisites.length > 0 && prerequisites.length < refs.length) {
      const first = await this.resolveAndPersistListYielding(prerequisites);
      const rest = await this.resolveAndPersistListYielding(refs.filter((ref) => !ReferenceResolver.isPrerequisite(ref)));
      return ReferenceResolver.mergeResults(first, rest);
    }
    const maybeYield = createYielder();
    const result = await this.resolveBatchYielding(refs, maybeYield);
    await this.persistResolutionResult(result, maybeYield);
    return result;
  }

  private async persistResolutionResult(result: ResolutionResult, maybeYield: MaybeYield): Promise<number> {
    const PERSIST_CHUNK = 1000;
    const edges = this.createEdges(result.resolved);
    for (let i = 0; i < edges.length; i += PERSIST_CHUNK) {
      this.queries.insertEdges(edges.slice(i, i + PERSIST_CHUNK));
      await maybeYield();
    }

    const resolvedCleanup = ReferenceResolver.partitionResolvedCleanup(result.resolved);
    for (let i = 0; i < resolvedCleanup.rowIds.length; i += PERSIST_CHUNK) {
      this.queries.deleteReferencesByRowIds(resolvedCleanup.rowIds.slice(i, i + PERSIST_CHUNK));
      await maybeYield();
    }
    for (let i = 0; i < resolvedCleanup.legacyKeys.length; i += PERSIST_CHUNK) {
      this.queries.deleteSpecificResolvedReferences(resolvedCleanup.legacyKeys.slice(i, i + PERSIST_CHUNK));
      await maybeYield();
    }

    const failedCleanup = ReferenceResolver.partitionFailedCleanup(this.nonDeferredFailures(result.unresolved));
    for (let i = 0; i < failedCleanup.byRowId.length; i += PERSIST_CHUNK) {
      this.queries.markReferencesFailedByRowIds(failedCleanup.byRowId.slice(i, i + PERSIST_CHUNK));
      await maybeYield();
    }
    for (let i = 0; i < failedCleanup.legacyKeys.length; i += PERSIST_CHUNK) {
      this.queries.markReferencesFailed(failedCleanup.legacyKeys.slice(i, i + PERSIST_CHUNK));
      await maybeYield();
    }

    return edges.length;
  }

  /** Finalize the durable queue only AFTER its edges have been inserted. */
  private async persistDeferredReferences(deferred: UnresolvedRef[], resolved: ResolvedRef[]): Promise<number> {
    for (const ref of deferred) if (ref.rowId != null) this.deferredRowIds.delete(ref.rowId);
    const matched = new Set(resolved.map((ref) => ref.original));
    const unresolved = deferred.filter((ref) => !matched.has(ref));
    const count = await this.persistResolutionResult({
      resolved,
      unresolved,
      stats: { total: deferred.length, resolved: resolved.length, unresolved: unresolved.length, byMethod: {} },
    }, createYielder());
    if (count > 0) this.clearCaches();
    return count;
  }

  /** Same two phases as the bounded DB reader: persist wiring before calls. */
  private static isPrerequisite(ref: UnresolvedReference): boolean {
    return ref.referenceKind === 'imports' || ref.referenceKind === 'extends' || ref.referenceKind === 'implements';
  }

  private static mergeResults(first: ResolutionResult, rest: ResolutionResult): ResolutionResult {
    const byMethod = { ...first.stats.byMethod };
    for (const [method, count] of Object.entries(rest.stats.byMethod)) {
      byMethod[method] = (byMethod[method] ?? 0) + count;
    }
    return {
      resolved: first.resolved.concat(rest.resolved),
      unresolved: first.unresolved.concat(rest.unresolved),
      stats: {
        total: first.stats.total + rest.stats.total,
        resolved: first.stats.resolved + rest.stats.resolved,
        unresolved: first.stats.unresolved + rest.stats.unresolved,
        byMethod,
      },
    };
  }

  /**
   * Second resolution pass for chained static-factory / fluent calls whose
   * chained method is defined on a SUPERTYPE the receiver's type conforms to —
   * a protocol-extension / inherited / default-interface method (#750). The
   * first pass can't resolve these because `implements`/`extends` edges aren't
   * built yet; this runs AFTER edges are persisted, so `context.getSupertypes`
   * (and the conformance fallback in resolveMethodOnType) can walk them.
   *
   * Operates only on the leftover unresolved refs that have the `inner().method`
   * chain shape, for the dotted-chain languages — a small set — and is idempotent
   * (re-resolving an already-resolved ref is a no-op since it's been deleted).
   * Returns the number of newly-created edges.
   */
  async resolveChainedCallsViaConformance(): Promise<number> {
    const deferred = this.deferredChainRefs;
    this.deferredChainRefs = [];
    if (deferred.length === 0) return 0;

    // Read fresh edges (the main pass built the implements/extends edges after
    // these refs were deferred). matchDottedCallChain now resolves a method on a
    // supertype via context.getSupertypes -> resolveMethodOnType's conformance walk.
    this.clearCaches();
    // This post-pass runs synchronously on the indexer's main thread; yield
    // periodically so the #850 liveness watchdog heartbeat can fire on a repo
    // with many deferred chained calls (#1091).
    const maybeYield = createYielder();
    const resolved: ResolvedRef[] = [];
    for (const ref of deferred) {
      // PHP `this->prop.method` resolves via matchMethodCall (declared-type
      // inference + resolveMethodOnType conformance walk); `::`-receiver
      // languages (Rust) split on `::` (matchScopedCallChain); other
      // dotted-receiver languages on `.` (matchDottedCallChain).
      const chainMatch = (ref.language === 'php' && PHP_PROP_SHAPE.test(ref.referenceName))
        ? matchMethodCall(ref, this.context)
        : SCOPED_CHAIN_LANGUAGES.has(ref.language)
        ? matchScopedCallChain(ref, this.context)
        : matchDottedCallChain(ref, this.context);
      const match = this.gateLanguage(chainMatch, ref);
      if (match) resolved.push(match);
      await maybeYield();
    }
    return this.persistDeferredReferences(deferred, resolved);
  }

  /**
   * Resolve one batch with a yield checkpoint between EVERY ref so the #850
   * liveness heartbeat can fire on a slow/dense batch (#1091). The checkpoint
   * granularity is per-ref — not per-N-refs — because per-ref cost is unbounded
   * in the worst case (a collision-heavy method name whose candidate set misses
   * the LRU re-fetches tens of thousands of rows): any fixed N multiplies that
   * worst case into the watchdog window, which is how v1.2.0 still got killed
   * at "Resolving refs" on large Java monorepos (#1122). `maybeYield()` is a
   * ~ns time check when under budget, so per-ref checkpoints cost nothing.
   * Behaviourally identical to `resolveAll(batch)`: `warmCaches()` is
   * idempotent (guarded) and `resolveOne` is independent per ref, so yielding
   * between refs changes only timing, never which edges get created.
   */
  private async resolveBatchYielding(
    batch: UnresolvedReference[],
    maybeYield: MaybeYield
  ): Promise<ResolutionResult> {
    this.warmCaches();
    this.advanceSupertypeGeneration();

    const resolved: ResolvedRef[] = [];
    const unresolved: UnresolvedRef[] = [];
    const byMethod: Record<string, number> = {};

    for (const raw of batch) {
      const ref: UnresolvedRef = {
        fromNodeId: raw.fromNodeId,
        referenceName: raw.referenceName,
        referenceKind: raw.referenceKind,
        line: raw.line,
        column: raw.column,
        filePath: raw.filePath || this.getFilePathFromNodeId(raw.fromNodeId),
        language: raw.language || this.getLanguageFromNodeId(raw.fromNodeId),
        rowId: raw.rowId,
      };
      const result = this.resolveOneTimed(ref);
      if (result) {
        resolved.push(result);
        byMethod[result.resolvedBy] = (byMethod[result.resolvedBy] || 0) + 1;
      } else {
        unresolved.push(ref);
      }
      // Fast-path the per-ref yield check: awaiting the async no-op costs a
      // microtask hop per ref, which dominates at ~10⁵ refs (see MaybeYield).
      const y = maybeYield();
      if (y) await y;
    }

    return {
      resolved,
      unresolved,
      stats: {
        total: batch.length,
        resolved: resolved.length,
        unresolved: unresolved.length,
        byMethod,
      },
    };
  }

  /**
   * Resolve a list of refs and return everything the ADMISSION side needs to
   * persist the outcome: resolutions, failures, the deferred post-pass refs
   * this run produced (drained, so the caller owns routing them), and stats.
   * This is the resolver-worker entry point — it runs the exact per-ref loop
   * of resolveBatchYielding, minus the main-thread yields (worker threads have
   * no watchdog heartbeat to starve). Results are in input order.
   */
  /**
   * CODEGRAPH_RESOLVE_PROFILE=1: per-outcome wall-clock histogram of
   * resolveOne, keyed by the winning strategy (`resolvedBy`) or
   * `fail:<referenceKind>` — the §7a.2 "profile the per-ref path" probe. The
   * kernel-scale batch loop is ~430s and CORE-INVARIANT (835.9s pooled-4-on-8
   * ≈ 812.5s sequential-on-2 for the whole superphase), so the next lever is
   * which CLASS of ref the time belongs to, not more parallelism. Off by
   * default: the hrtime pair costs ~100ns/ref only when the env is set.
   */
  private resolveProfile: Map<string, { n: number; ns: bigint }> | null =
    process.env.CODEGRAPH_RESOLVE_PROFILE ? new Map() : null;

  /**
   * CODEGRAPH_RESOLVE_PROFILE=2 additionally attributes time to the
   * STRATEGIES inside resolveOne (`stage:<name>|<refKind>|hit/miss` rows in
   * the same histogram) — i.e. WHICH machinery a failing class of refs pays
   * for, not just that it fails. =1 keeps the per-outcome rows only.
   */
  private profileStages: boolean = process.env.CODEGRAPH_RESOLVE_PROFILE === '2';

  private stageAdd(stage: string, ref: UnresolvedRef, hit: boolean, t0: bigint): void {
    if (!this.resolveProfile) return;
    const dt = process.hrtime.bigint() - t0;
    const key = `stage:${stage}|${ref.referenceKind}|${hit ? 'hit' : 'miss'}`;
    const slot = this.resolveProfile.get(key);
    if (slot) {
      slot.n++;
      slot.ns += dt;
    } else {
      this.resolveProfile.set(key, { n: 1, ns: dt });
    }
  }

  private resolveOneTimed(ref: UnresolvedRef): ResolvedRef | null {
    if (!this.resolveProfile) return this.resolveOne(ref);
    const t0 = process.hrtime.bigint();
    const result = this.resolveOne(ref);
    const dt = process.hrtime.bigint() - t0;
    const key = result ? result.resolvedBy : `fail:${ref.referenceKind}`;
    const slot = this.resolveProfile.get(key);
    if (slot) {
      slot.n++;
      slot.ns += dt;
    } else {
      this.resolveProfile.set(key, { n: 1, ns: dt });
    }
    return result;
  }

  /** Dump the CODEGRAPH_RESOLVE_PROFILE histogram to stderr (no-op when off). */
  dumpResolveProfile(label: string): void {
    if (!this.resolveProfile || this.resolveProfile.size === 0) return;
    const rows = [...this.resolveProfile.entries()]
      .map(([k, v]) => ({ k, n: v.n, ms: Number(v.ns / 1_000_000n) }))
      .sort((a, b) => b.ms - a.ms);
    for (const r of rows) {
      console.error(
        `[resolve-profile] ${label} ${r.k}: n=${r.n} total=${(r.ms / 1000).toFixed(1)}s avg=${((r.ms * 1000) / Math.max(1, r.n)).toFixed(0)}µs`
      );
    }
    // =2 only: this thread's matchReference sub-stage table rides along.
    dumpNameMatcherProfile(label);
  }

  resolveListForAdmission(refs: UnresolvedReference[]): {
    resolved: ResolvedRef[];
    unresolved: UnresolvedRef[];
    deferredChain: UnresolvedRef[];
    deferredThisMember: UnresolvedRef[];
    byMethod: Record<string, number>;
  } {
    this.warmCaches();
    this.advanceSupertypeGeneration();
    const resolved: ResolvedRef[] = [];
    const unresolved: UnresolvedRef[] = [];
    const byMethod: Record<string, number> = {};
    for (const raw of refs) {
      const ref: UnresolvedRef = {
        fromNodeId: raw.fromNodeId,
        referenceName: raw.referenceName,
        referenceKind: raw.referenceKind,
        line: raw.line,
        column: raw.column,
        filePath: raw.filePath || this.getFilePathFromNodeId(raw.fromNodeId),
        language: raw.language || this.getLanguageFromNodeId(raw.fromNodeId),
        rowId: raw.rowId,
      };
      const result = this.resolveOneTimed(ref);
      if (result) {
        resolved.push(result);
        byMethod[result.resolvedBy] = (byMethod[result.resolvedBy] || 0) + 1;
      } else {
        unresolved.push(ref);
      }
    }
    this.deferredRowIds.clear(); // the admission side now owns both queues
    return {
      resolved,
      unresolved,
      deferredChain: this.deferredChainRefs.splice(0),
      deferredThisMember: this.deferredThisMemberRefs.splice(0),
      byMethod,
    };
  }

  /**
   * The resolver's live ResolutionContext — resolver-pool workers use it to
   * run synthesis passes against their own read-only connection.
   */
  getResolutionContext(): ResolutionContext {
    return this.context;
  }

  /**
   * Re-queue deferred post-pass refs produced by resolver workers, preserving
   * their admission order so resolveChainedCallsViaConformance /
   * resolveDeferredThisMemberRefs process them exactly as the sequential path
   * would have.
   */
  appendDeferredFromWorkers(deferredChain: UnresolvedRef[], deferredThisMember: UnresolvedRef[]): void {
    for (const ref of deferredChain) this.deferReference(ref, this.deferredChainRefs);
    for (const ref of deferredThisMember) this.deferReference(ref, this.deferredThisMemberRefs);
  }

  /**
   * Resolve and persist in batches to keep memory bounded.
   * Processes unresolved references in chunks, persisting edges and cleaning
   * up resolved refs after each batch to avoid accumulating large arrays.
   */
  async resolveAndPersistBatched(
    onProgress?: (current: number, total: number) => void,
    batchSize: number = 5000,
    onSynthesisProgress?: (done: number, total: number) => void,
    // When provided, big batches fan out across a read-only resolver-worker
    // pool with results admitted in canonical order (see resolver-pool.ts).
    // Sequential fallback on any pool failure. CODEGRAPH_NO_PARALLEL_RESOLVE=1
    // disables entirely. bulkEdgeLoad hooks (when provided) bracket the batch
    // loop with drop/recreate of the non-unique edge indexes on big runs —
    // see DatabaseConnection.beginBulkEdgeLoad. backpressure (when provided)
    // is the WAL valve's writer-side backstop (WalCheckpointValve.backpressure):
    // called at pool-idle boundaries so a full backfill can actually complete —
    // the valve's timer-driven passive passes stay perpetually partial against
    // the pool's continuous reads, which is how a kernel-scale resolution grew
    // a 22GB WAL on a 4.6GB DB (migration plan §7a.1).
    parallel?: {
      dbPath: string;
      bulkEdgeLoad?: { begin: () => void; end: () => void | Promise<void> };
      /** unresolved_refs index window for the batched loop — the loop only
       *  reads the status index + PK; dropping the sync-path ref indexes cuts
       *  each per-batch DELETE's B-tree work (DatabaseConnection.beginBulkRefLoad). */
      refIndexLoad?: { begin: () => void; end: () => void | Promise<void> };
      backpressure?: () => Promise<void> | null;
    }
  ): Promise<ResolutionResult> {
    // Resolution runs on the indexer's MAIN thread, and the #850 liveness
    // watchdog SIGKILLs a process whose event loop stalls past its window (60s
    // by default). A single dense batch's resolveAll — or the synthesis pass
    // below — can exceed that on a large repo, killing a VALID in-progress index
    // (#1091). A shared yielder lets both give the watchdog heartbeat a regular
    // window to fire; see ./cooperative-yield.
    const maybeYield = createYielder();

    if (process.env.CODEGRAPH_SYNTH_TIMINGS) {
      console.error(`[pool-timing] backpressure hook: ${parallel?.backpressure ? 'present' : 'absent'}`);
    }

    // CODEGRAPH_RESOLVE_PROFILE loop-stage attribution: the §7a.2 kernel-scale
    // histogram showed resolveOne owns only ~93s of the ~436s batch loop —
    // these counters name where the other ~340s goes (reads, edge build+insert,
    // deletes/marks, the per-batch count guard).
    const loopProf: Record<string, number> | null = process.env.CODEGRAPH_RESOLVE_PROFILE
      ? { read: 0, settle: 0, backpressure: 0, recycle: 0, createEdges: 0, insertEdges: 0, deletes: 0, marks: 0, countGuard: 0 }
      : null;
    const lp = (k: string, t0: number): void => { if (loopProf) loopProf[k] = (loopProf[k] ?? 0) + (Date.now() - t0); };
    let tLp = 0;

    await this.warmCachesYielding(maybeYield);

    const total = this.queries.getUnresolvedReferencesCount();
    let processed = 0;
    const aggregateStats = {
      total: 0,
      resolved: 0,
      unresolved: 0,
      byMethod: {} as Record<string, number>,
    };

    // Parallel pool, started immediately but never awaited up front: early
    // batches run sequentially while the workers boot (module load + readonly
    // DB open + framework detect + cache warm ≈ hundreds of ms), and the loop
    // switches to fan-out the moment the pool reports ready — so pool boot
    // costs zero wall-clock. Any failure downgrades to sequential permanently.
    let pool: ResolverPool | null = null;
    let poolReady = false;
    // True once pool creation has been attempted by EITHER engage site (the
    // up-front ref-count gate or the adaptive projection below) — a pool that
    // failed or was destroyed must stay down (downgrade is permanent), and
    // tryCreate's sizing probes shouldn't re-run every batch on hosts that
    // declined.
    let poolEngageTried = false;
    const createPool = (t0: number, why: string): ResolverPool | null => {
      poolEngageTried = true;
      if (!parallel) return null;
      const p = ResolverPool.tryCreate(parallel.dbPath, this.projectRoot);
      p?.ready().then(
        () => {
          poolReady = true;
          if (process.env.CODEGRAPH_SYNTH_TIMINGS) console.error(`[pool-timing] pool ready after ${Date.now() - t0}ms (${why})`);
        },
        () => {
          void p.destroy().catch(() => undefined);
          if (pool === p) pool = null;
        }
      );
      return p;
    };
    if (parallel && total >= minRefsForPool()) {
      pool = createPool(Date.now(), 'ref-count');
    }
    // Adaptive engagement bar (see the batch-loop hook): projected remaining
    // sequential settle above this boots the pool mid-loop. Boot is async and
    // fan-out waits for ready, so a marginal engage costs background boot
    // only; the bar just needs to clear the fan-out's own overhead class.
    const ADAPTIVE_ENGAGE_SETTLE_MS = 400;
    let adaptiveSeqMs = 0;
    let adaptiveSeqRefs = 0;

    // Process in PIPELINED batches (double-buffer). The enumeration is the
    // head of the pending set in rowid order; every ref a persisted batch
    // processed leaves the pending set (resolved rows are deleted,
    // unresolvable ones flip to status='failed'), shifting the remaining
    // pending rows forward.
    let prevRemaining = Number.POSITIVE_INFINITY;

    // Cadence for the worker connection recycling below — ~8 batches
    // ≈ 40k refs between recycles keeps the WAL shallow at kernel scale
    // while a small sync never recycles at all. (25 recovered only half
    // the write tax — the WAL re-deepened between recycles; reopens are
    // sub-millisecond so the shorter cadence is ~free.)
    const RECYCLE_EVERY_BATCHES = 8;
    let batchesSinceRecycle = 0;

    // Fan-out result of ResolverPool.resolveBatch, settled (never rejecting)
    // so a fan-out begun before the previous batch's persist can't produce an
    // unhandled rejection while it waits to be awaited.
    type PoolSettled =
      | { ok: true; out: Awaited<ReturnType<ResolverPool['resolveBatch']>> }
      | { ok: false; err: unknown };
    type InFlight = { mode: 'pool'; settled: Promise<PoolSettled> } | { mode: 'seq' };

    // Begin one batch: fan out to the pool when it's ready and the batch is
    // big enough — workers then resolve batch k+1 WHILE the main thread
    // persists batch k (persist measured at ~58% of resolution wall on a
    // 255k-ref repo, all of it previously spent with the pool idle).
    // Sequential batches stay lazy: they run on the main thread at settle
    // time, where an early start would only contend with the persist.
    const beginBatch = (batch: UnresolvedReference[]): InFlight => {
      if (pool && poolReady && ResolverPool.worthParallel(batch.length)) {
        return {
          mode: 'pool',
          settled: pool.resolveBatch(batch).then(
            (out) => ({ ok: true as const, out }),
            (err: unknown) => ({ ok: false as const, err })
          ),
        };
      }
      return { mode: 'seq' };
    };

    // Settle an in-flight batch to a ResolutionResult. Deferred post-pass refs
    // are appended HERE, in loop order — never inside the fan-out promise — so
    // admission order stays exactly the sequential order even while a later
    // batch resolves concurrently. A pool failure downgrades to sequential
    // permanently and re-resolves this batch on the main thread.
    const settleBatch = async (
      inFlight: InFlight,
      batch: UnresolvedReference[]
    ): Promise<ResolutionResult> => {
      if (inFlight.mode === 'pool') {
        const settled = await inFlight.settled;
        if (settled.ok) {
          this.appendDeferredFromWorkers(settled.out.deferredChain, settled.out.deferredThisMember);
          return {
            resolved: settled.out.resolved,
            unresolved: settled.out.unresolved,
            stats: {
              total: batch.length,
              resolved: settled.out.resolved.length,
              unresolved: settled.out.unresolved.length,
              byMethod: settled.out.byMethod,
            },
          };
        }
        logDebug('Parallel resolution failed; falling back to sequential', {
          error: settled.err instanceof Error ? settled.err.message : String(settled.err),
        });
        if (pool) await pool.destroy().catch(() => undefined);
        pool = null;
      }
      return this.resolveBatchYielding(batch, maybeYield);
    };

    // Bulk edge load: on big runs, drop the non-unique edge indexes for the
    // duration of the batch loop (the identity index stays — OR IGNORE dedup
    // and the source-keyed supertype-walk reads both live on it). Recreated in
    // the inner finally BEFORE synthesis, whose passes read kind-keyed.
    // Measured on a 224k-edge resolution set: insert 2.8s → 1.1s + 0.3s
    // recreate. Same ref-count gate as the pool so small syncs never pay the
    // recreate cost.
    let bulkEdgesActive = false;
    if (parallel?.bulkEdgeLoad && total >= minRefsForPool()) {
      try {
        parallel.bulkEdgeLoad.begin();
        bulkEdgesActive = true;
      } catch { /* keep the indexes; inserts just pay the per-row maintenance */ }
    }
    // Same gate for the ref-index window: the loop's deletes stop maintaining
    // the five sync-path unresolved_refs indexes, and the end-of-loop rebuild
    // is near-free (only failed refs survive the loop).
    let bulkRefsActive = false;
    if (parallel?.refIndexLoad && total >= minRefsForPool()) {
      try {
        parallel.refIndexLoad.begin();
        bulkRefsActive = true;
      } catch { /* keep the indexes; deletes just pay the per-row maintenance */ }
    }

    try {
    try {
    // Orphans retain interruption/re-extraction order, not clean-index order.
    // A caller can precede its imports or supertypes by many batches (#1577).
    // Drain those prerequisites first, then start a fresh keyset cursor over
    // the remaining kinds. The disjoint filters let us prefetch across the
    // phase boundary before cleanup without re-reading the current batch.
    let prerequisites = true;
    let afterRowId = 0;
    const readNextBatch = (): UnresolvedReference[] => {
      let next = this.queries.getUnresolvedReferencesBatchAfter(afterRowId, batchSize, prerequisites);
      if (next.length === 0 && prerequisites) {
        prerequisites = false;
        afterRowId = 0;
        next = this.queries.getUnresolvedReferencesBatchAfter(afterRowId, batchSize, prerequisites);
      }
      if (next.length > 0) afterRowId = next[next.length - 1]!.rowId!;
      return next;
    };
    tLp = Date.now();
    let batch = readNextBatch();
    lp('read', tLp);
    let inFlight: InFlight | null = batch.length > 0 ? beginBatch(batch) : null;
    while (batch.length > 0 && inFlight) {
      // Prefetch the NEXT batch before this one persists: this batch's rows
      // are still pending (nothing has mutated the table since they were
      // read), so seeking past this batch's last row id in the same rowid
      // enumeration yields the following batch (keyset — OFFSET re-walked the
      // accumulated failed prefix every read, 54.6s at kernel scale, §7a.2).
      tLp = Date.now();
      const nextBatch = readNextBatch();
      lp('read', tLp);

      const tBatch = Date.now();
      const result = await settleBatch(inFlight, batch);
      if (process.env.CODEGRAPH_SYNTH_TIMINGS) console.error(`[pool-timing] batch ${inFlight.mode}: ${batch.length} refs in ${Date.now() - tBatch}ms`);
      lp('settle', tBatch);

      // Adaptive pool engagement: the fixed ref-count gate can't see PER-REF
      // cost, and settle rates differ ~9× by language (56k Rust refs cost
      // more sequential settle than 154k Go refs — 36µs vs 4µs measured on
      // tokio/prometheus). After each sequential batch, project the remaining
      // settle from the observed rate and boot the pool mid-loop when it
      // clears the bar. The loop already switches to fan-out only when the
      // async boot reports ready, admission order is mode-independent, and
      // 2-core/low-memory hosts still decline inside tryCreate's sizing —
      // so the switch changes wall-clock, never the graph.
      if (inFlight.mode === 'seq' && parallel && pool === null && !poolEngageTried) {
        adaptiveSeqMs += Date.now() - tBatch;
        adaptiveSeqRefs += batch.length;
        const remaining = total - processed - batch.length;
        const projectedMs = (adaptiveSeqMs / Math.max(1, adaptiveSeqRefs)) * Math.max(0, remaining);
        if (projectedMs >= ADAPTIVE_ENGAGE_SETTLE_MS) {
          if (process.env.CODEGRAPH_SYNTH_TIMINGS) {
            console.error(`[pool-timing] adaptive engage: projected ${Math.round(projectedMs)}ms sequential settle over ${remaining} remaining refs`);
          }
          pool = createPool(Date.now(), 'adaptive');
        }
      }

      // WAL-valve backstop at the ONE pool-idle boundary of the double-buffer
      // (this batch settled, the next not yet fanned out): past the hard cap
      // the writer parks for a full backfill here, where the pool's readers
      // are all between statements — so the backfill completes, readers
      // re-enter at SQLite's backfilled mark, and the next persist commit
      // WRAPS the WAL instead of growing it. No-op (one fstat) under the cap.
      tLp = Date.now();
      const bp = parallel?.backpressure?.();
      if (bp) await bp;
      lp('backpressure', tLp);

      // Recycle the workers' read connections periodically at this same
      // worker-idle boundary (batch k settled, batch k+1 not yet fanned
      // out): a long-lived reader pins WAL checkpoint progress, and the
      // deep WAL that accumulates behind it taxes the writer's OWN page
      // operations — the §7a.6 writes-under-readers finding (deletes
      // 42.6s → 118.8s from 0 to 4 attached readers; an aggressive valve
      // recovered the writes but paid +129s in full-park folds). Releasing
      // the read marks every ~25 batches lets the existing checkpoints
      // advance instead, at ~milliseconds of reopen cost. A failed recycle
      // downgrades to sequential permanently, same as a failed fan-out.
      if (pool && poolReady && ++batchesSinceRecycle >= RECYCLE_EVERY_BATCHES) {
        batchesSinceRecycle = 0;
        tLp = Date.now();
        try {
          await pool.recycleWorkers();
        } catch (err) {
          logDebug('Worker connection recycle failed; falling back to sequential', {
            error: err instanceof Error ? err.message : String(err),
          });
          await pool.destroy().catch(() => undefined);
          pool = null;
        }
        lp('recycle', tLp);
      }

      // Persist in bounded sub-transactions with yields between: a whole
      // batch's edge insert / keyed deletes are otherwise one solid
      // synchronous span each on a multi-GB index, sitting BETWEEN the
      // per-ref yields — the last unyielded stretch of the resolution loop.
      // Crash semantics are unchanged (already several transactions): edges
      // land before their refs are deleted, so a kill mid-way re-resolves
      // the remainder idempotently on the next run/sweep (#1187).
      const PERSIST_CHUNK = 1000;
      const tPersist = Date.now();

      // Persist edges BEFORE fanning out the next batch: later batches read
      // this batch's edges — resolveMethodOnType walks supertype chains over
      // `extends`/`implements` edges that earlier batches resolved, so a
      // receiver typed as a subclass only reaches a method declared on its
      // base class if those edges are visible. (Validated on dubbo: fanning
      // out first downgraded exactly those supertype-method resolutions from
      // the 0.9 typed-receiver path to the 0.65 word-overlap fallback.)
      tLp = Date.now();
      const edges = this.createEdges(result.resolved);
      lp('createEdges', tLp);
      tLp = Date.now();
      for (let i = 0; i < edges.length; i += PERSIST_CHUNK) {
        this.queries.insertEdges(edges.slice(i, i + PERSIST_CHUNK));
        await maybeYield();
      }
      lp('insertEdges', tLp);

      // NOW fan the next batch out — workers see exactly the edge state the
      // sequential baseline would (every batch ≤ this one committed), while
      // the main thread spends the REST of the persist (ref deletes + failed
      // parking below) overlapped with their resolution — the double-buffer.
      const nextInFlight = nextBatch.length > 0 ? beginBatch(nextBatch) : null;

      // Clean up resolved refs so they don't appear in the next batch —
      // by row id, so a same-key sibling ref in a LATER batch (same caller
      // calling the same callee at another line) is left pending for its own
      // attempt instead of being swept out with this batch's rows (#1269).
      tLp = Date.now();
      let removedThisBatch = 0;
      const resolvedCleanup = ReferenceResolver.partitionResolvedCleanup(result.resolved);
      for (let i = 0; i < resolvedCleanup.rowIds.length; i += PERSIST_CHUNK) {
        removedThisBatch += this.queries.deleteReferencesByRowIds(resolvedCleanup.rowIds.slice(i, i + PERSIST_CHUNK));
        await maybeYield();
      }
      for (let i = 0; i < resolvedCleanup.legacyKeys.length; i += PERSIST_CHUNK) {
        removedThisBatch += this.queries.deleteSpecificResolvedReferences(resolvedCleanup.legacyKeys.slice(i, i + PERSIST_CHUNK));
        await maybeYield();
      }
      lp('deletes', tLp);

      // Park unresolvable refs from this batch as status='failed' so they
      // leave the pending set (the batch reader and non-progress guard below
      // only see pending rows) but stay retryable when a later sync adds a
      // symbol that could satisfy them (#1240).
      tLp = Date.now();
      const failures = this.nonDeferredFailures(result.unresolved);
      const deferredCount = result.unresolved.length - failures.length;
      const failedCleanup = ReferenceResolver.partitionFailedCleanup(failures);
      for (let i = 0; i < failedCleanup.byRowId.length; i += PERSIST_CHUNK) {
        removedThisBatch += this.queries.markReferencesFailedByRowIds(failedCleanup.byRowId.slice(i, i + PERSIST_CHUNK));
        await maybeYield();
      }
      for (let i = 0; i < failedCleanup.legacyKeys.length; i += PERSIST_CHUNK) {
        removedThisBatch += this.queries.markReferencesFailed(failedCleanup.legacyKeys.slice(i, i + PERSIST_CHUNK));
        await maybeYield();
      }
      lp('marks', tLp);

      if (process.env.CODEGRAPH_SYNTH_TIMINGS) console.error(`[pool-timing] batch persist: ${Date.now() - tPersist}ms`);

      // Aggregate stats
      aggregateStats.total += result.stats.total;
      aggregateStats.resolved += result.stats.resolved;
      aggregateStats.unresolved += result.stats.unresolved;
      for (const [method, count] of Object.entries(result.stats.byMethod)) {
        aggregateStats.byMethod[method] = (aggregateStats.byMethod[method] || 0) + count;
      }

      processed += batch.length;
      onProgress?.(processed, total);

      // Yield so progress UI can render between batches
      await new Promise(resolve => setImmediate(resolve));

      // NOTE: there used to be an extra early break here when a batch resolved
      // nothing (`result.unresolved.length === batch.length`). That was wrong:
      // an all-unresolvable batch still DELETES its rows (progress), yet the
      // break abandoned every batch after it in the same run — on a repo whose
      // first 5000 refs are all external/stdlib calls, resolution stopped at
      // batch one and left the rest of the table as permanent orphans (#1187).
      // The count-based guard below catches the true no-progress case.

      // Non-progress guard (defense-in-depth). Ordinary attempts must leave
      // the pending set; a mismatched original reference can make legacy-key
      // cleanup a no-op. Keep the guard against that broken persistence even
      // though keyset pagination now advances independently of row cleanup.
      // An abandoned prefetched batch has no side effects until settleBatch.
      // Non-progress signal, now O(1): `changes` summed across this batch's
      // deletes + failed-parks is the DIRECT evidence the guard's old count
      // diff inferred — a resolver returning a mismatched name makes the keyed
      // cleanup no-op, which shows up here as zero removals. The per-batch
      // COUNT(*) it replaces walked every remaining pending row — O(N²/batch)
      // over a run, 93.9s of the kernel-scale batch loop (§7a.2). A REAL count
      // runs only on the suspicious path (claimed-work batch removed nothing —
      // e.g. every row was a sibling a legacy-key sweep already consumed),
      // where it arbitrates stop-vs-continue exactly as before.
      // Deferred refs legitimately remain pending for the post-pass. The
      // keyset cursor advances past them; they must not trigger this guard.
      if (removedThisBatch + deferredCount <= 0 && batch.length > 0) {
        tLp = Date.now();
        const remaining = this.queries.getUnresolvedReferencesCount();
        lp('countGuard', tLp);
        if (remaining >= prevRemaining) break;
        prevRemaining = remaining;
      }

      // Advance the pipeline: the prefetched batch (already fanned out when
      // the pool is on) becomes the current one.
      batch = nextBatch;
      inFlight = nextInFlight;
    }
    } finally {
      // Recreate the edge indexes BEFORE synthesis (kind-keyed reads) and on
      // any error path. A crash before this line is healed by the next
      // DatabaseConnection open (schema.sql re-applies IF NOT EXISTS).
      if (bulkRefsActive) {
        const tRef = Date.now();
        await parallel!.refIndexLoad!.end();
        if (process.env.CODEGRAPH_SYNTH_TIMINGS) console.error(`[phase-timing] ref-index-recreate: ${Date.now() - tRef}ms`);
      }
      if (bulkEdgesActive) {
        const tIdx = Date.now();
        await parallel!.bulkEdgeLoad!.end();
        if (process.env.CODEGRAPH_SYNTH_TIMINGS) console.error(`[phase-timing] edge-index-recreate: ${Date.now() - tIdx}ms`);
        // The recreate just wrote every non-unique edge index into the WAL
        // (multi-GB at kernel scale) with the pool idle — fold before the
        // synthesis passes pin readers against it for minutes.
        const bp = parallel?.backpressure?.();
        if (bp) await bp;
      }
    }

    // Dynamic-edge synthesis: now that all base `calls` edges are persisted,
    // synthesize observer/callback dispatch edges (dispatcher → registered
    // callbacks) that static parsing leaves out. Best-effort — never fail the
    // index on it. The pool (when it survived resolution) is REUSED to fan the
    // independent passes across its read-only workers — that's why its destroy
    // lives in the finally below, after synthesis, not at the end of the batch
    // loop. See docs/design/callback-edge-synthesis.md.
    const tSynth = Date.now();
    try {
      aggregateStats.byMethod['callback-synthesis'] = await synthesizeCallbackEdges(
        this.queries,
        this.context,
        onSynthesisProgress,
        pool,
        parallel?.backpressure
      );
    } catch {
      // synthesis is additive and optional; ignore failures
    }
    if (process.env.CODEGRAPH_SYNTH_TIMINGS) console.error(`[phase-timing] callback-synthesis: ${Date.now() - tSynth}ms`);
    } finally {
      if (pool) await pool.destroy().catch(() => undefined);
    }

    if (loopProf) {
      const parts = Object.entries(loopProf).map(([k, v]) => `${k}=${(v / 1000).toFixed(1)}s`).join(' ');
      console.error(`[resolve-profile] loop-stages ${parts}`);
    }
    this.dumpResolveProfile('main');

    return {
      resolved: [],
      unresolved: [],
      stats: aggregateStats,
    };
  }

  /**
   * Get detected frameworks
   */
  getDetectedFrameworks(): string[] {
    return this.frameworks.map((f) => f.name);
  }

  /**
   * True when `receiver` is a local name bound by an import that resolves to a
   * file IN THIS PROJECT — the only case where letting a python
   * built-in-method name through the filter is safe (#1681).
   *
   * Asking only whether SOME import bound the local name is not enough: every
   * import produces a mapping, stdlib and PyPI included, so that would also be
   * true for `os`, `requests`, `np`. Opening the filter for them lets
   * resolveViaImport find no project file, fall through to bare-name matching,
   * and bind `os.remove(p)` to whatever project method happens to be named
   * `remove` — reintroducing, through its own escape hatch, the fabricated-edge
   * class this filter exists to prevent.
   *
   * Resolving the specifier is the same question resolveViaImport will ask
   * next, so a receiver that passes here is one the qualified path can actually
   * serve; anything else stays a silent miss rather than a wrong edge.
   */
  private isPythonProjectModule(ref: UnresolvedRef, receiver: string): boolean {
    for (const imp of this.context.getImportMappings(ref.filePath, ref.language)) {
      if (imp.localName !== receiver) continue;
      // `import pkg.mod` / `import pkg.mod as m` binds the module `source`
      // names. `from pkg import mod` binds `pkg.mod`, and `from . import mod`
      // binds `.mod` — join without doubling the dot that makes `.` mean the
      // current package.
      const specifier = imp.isNamespace
        ? imp.source
        : imp.source.endsWith('.')
          ? `${imp.source}${imp.exportedName}`
          : `${imp.source}.${imp.exportedName}`;
      if (resolveImportPath(specifier, ref.filePath, ref.language!, this.context)) {
        return true;
      }
    }
    return false;
  }

  /**
   * Check if reference is to a built-in or external symbol
   */
  private isBuiltInOrExternal(ref: UnresolvedRef): boolean {
    const name = ref.referenceName;
    const isJsTs = ref.language === 'typescript' || ref.language === 'javascript'
      || ref.language === 'tsx' || ref.language === 'jsx' || ref.language === 'arkts';

    // JavaScript/TypeScript built-ins
    if (isJsTs && JS_BUILT_INS.has(name)) {
      return true;
    }

    // ArkTS resource-reference intrinsics — `$r('app.string.x')` /
    // `$rawfile('x.png')` are framework-provided and appear dozens of times
    // per UI file; without this they can resolve to a stray same-named
    // symbol (e.g. a checked-in hvigor wrapper's `$r`).
    if (ref.language === 'arkts' && (name === '$r' || name === '$rawfile')) {
      return true;
    }

    // Common JS/TS library calls (console.log, Math.floor, JSON.parse)
    if (isJsTs && (name.startsWith('console.') || name.startsWith('Math.') || name.startsWith('JSON.'))) {
      return true;
    }

    // React hooks from React itself
    if (isJsTs && REACT_HOOKS.has(name)) {
      return true;
    }

    // Python built-ins (bare calls only — dotted calls like console.print are method calls)
    if (ref.language === 'python' && PYTHON_BUILT_INS.has(name)) {
      return true;
    }

    // Python built-in method calls (e.g., list.extend, dict.update)
    if (ref.language === 'python') {
      const dotIdx = name.indexOf('.');
      if (dotIdx > 0) {
        const receiver = name.substring(0, dotIdx);
        const method = name.substring(dotIdx + 1);
        // Filter calls on built-in types (list.append, dict.update, etc.)
        if (PYTHON_BUILT_IN_TYPES.has(receiver)) {
          return true;
        }
        // Filter built-in methods on non-class receivers
        // (e.g., items.append where items is a local list variable)
        // But allow if the capitalized receiver matches a known codebase class,
        // OR the receiver is itself an imported project module — a module can
        // export a top-level function sharing a common collection-method name
        // (`ledger.append`, `from . import ledger`), and that call is a real
        // project dependency, not `list.append` (#1681). Without this, the
        // qualified ref never reaches resolveViaImport / resolvePythonModuleMember.
        if (PYTHON_BUILT_IN_METHODS.has(method)) {
          // A module-scope collection binding is stronger evidence than a
          // coincidentally matching class name (#1652). Only use this file's
          // binding: an unrelated module may reuse the receiver for a collection.
          const isCollection = this.context.getNodesByName(receiver).some((node) =>
            node.language === 'python' && node.filePath === ref.filePath &&
            (node.kind === 'variable' || node.kind === 'constant') &&
            node.qualifiedName === receiver &&
            /^=\s*(?:[\[{]|(?:dict|list|set|tuple|frozenset)\s*\(|\(\s*\)|\([^()]*,)/.test(node.signature ?? '')
          );
          if (isCollection) return true;

          const capitalized = receiver.charAt(0).toUpperCase() + receiver.slice(1);
          const isKnownClass = this.context.getNodesByName(capitalized).some((node) =>
            node.language === 'python' &&
            (node.kind === 'class' || node.kind === 'struct' || node.kind === 'interface')
          );
          const isProjectModule =
            !isKnownClass && this.isPythonProjectModule(ref, receiver);
          if (!isKnownClass && !isProjectModule) {
            return true;
          }
        }
      }
      // A bare name colliding with a builtin method (index, get, update, count…)
      // is only a builtin when NOTHING in the codebase declares it. A declared
      // symbol with that exact name — e.g. a Flask/FastAPI view `def index()` or
      // `def get()` — is a real reference target. Without this guard, every
      // handler named after a builtin method silently loses its route→handler edge.
      if (PYTHON_BUILT_IN_METHODS.has(name) && !this.knownNames?.has(name)) {
        return true;
      }
    }

    // Go standard library packages — refs like "fmt.Println", "http.ListenAndServe", etc.
    if (ref.language === 'go') {
      const dotIdx = name.indexOf('.');
      if (dotIdx > 0) {
        const pkg = name.substring(0, dotIdx);
        if (GO_STDLIB_PACKAGES.has(pkg)) {
          return true;
        }
      }
      if (GO_BUILT_INS.has(name)) {
        return true;
      }
    }

    // Pascal/Delphi built-ins and standard library units
    if (ref.language === 'pascal') {
      if (PASCAL_UNIT_PREFIXES.some((p) => name.startsWith(p))) {
        return true;
      }
      if (PASCAL_BUILT_INS.has(name)) {
        return true;
      }
    }

    // C/C++ standard library symbols (printf, malloc, std::vector, etc.).
    // Names that collide with user-defined symbols are NOT filtered —
    // C and C++ projects routinely shadow stdlib names (custom allocators
    // define `malloc`/`free`, stream wrappers define `read`/`write`/`open`,
    // containers define `move`/`swap`, logging libs wrap `printf`). Killing
    // those resolutions makes the graph wrong, not cleaner. We only filter
    // when there's no user node with this name — then name-matching would
    // produce zero edges anyway and the filter just short-circuits work.
    if (ref.language === 'c' || ref.language === 'cpp') {
      // C++ std:: namespace prefix — safe to filter unconditionally,
      // since `std::foo` is never a user-defined qualified name in
      // tree-sitter output.
      if (name.startsWith('std::')) return true;
      if (C_BUILT_INS.has(name) || CPP_BUILT_INS.has(name)) {
        return !this.hasAnyPossibleMatch(name);
      }
    }

    return false;
  }

  /**
   * Get file path from node ID
   */
  private getFilePathFromNodeId(nodeId: string): string {
    const node = this.queries.getNodeById(nodeId);
    return node?.filePath || '';
  }

  /**
   * Get language from node ID
   */
  private getLanguageFromNodeId(nodeId: string): UnresolvedRef['language'] {
    const node = this.queries.getNodeById(nodeId);
    return node?.language || 'unknown';
  }

  /**
   * Drop an import/name-strategy resolution that crosses a language family.
   * Two regimes (mirrors `applyLanguageGate`'s candidate filter):
   *  - `references` (type usage): STRICT — a `Type.member` static read names a
   *    same-family type, never a coincidentally same-named symbol in another
   *    language. Drops any non-same-family target.
   *  - `imports` (import binding / `#include`): both-known — a C++ `#include
   *    "X.h"` must not resolve to a same-named ObjC header on another platform
   *    (basename collision), but a singleton-family / SFC language (`vue` →
   *    `.ts`) importing across is left alone.
   * Applies to the import (strategy 2) + name-match (strategy 3) results.
   */
  /**
   * Collect the `@using` namespaces in scope for a `.razor`/`.cshtml` file: its
   * own `@using` directives plus every `_Imports.razor` from the file's folder up
   * to the project root (Razor `_Imports` cascade). Cached per file.
   */
  private getRazorUsings(filePath: string): string[] {
    const cached = this.razorUsingsCache.get(filePath);
    if (cached) return cached;
    const usings = new Set<string>();
    const addFrom = (src: string | null): void => {
      if (!src) return;
      for (const m of src.matchAll(/^\s*@using\s+(?:static\s+)?([A-Za-z_][\w.]*)/gm)) usings.add(m[1]!);
    };
    addFrom(this.context.readFile(filePath));
    let dir = filePath.includes('/') ? filePath.slice(0, filePath.lastIndexOf('/')) : '';
    // Walk up to the project root, reading each level's _Imports.razor.
    for (;;) {
      addFrom(this.context.readFile(dir ? `${dir}/_Imports.razor` : '_Imports.razor'));
      if (!dir) break;
      const slash = dir.lastIndexOf('/');
      dir = slash >= 0 ? dir.slice(0, slash) : '';
    }
    const arr = [...usings];
    this.razorUsingsCache.set(filePath, arr);
    return arr;
  }

  /**
   * Resolve a Razor/Blazor simple type ref through the file's `@using`
   * namespaces: `CatalogBrand` + `@using BlazorShared.Models` → the node whose
   * qualified name is `BlazorShared.Models::CatalogBrand`. Only resolves when the
   * `@using` set yields exactly ONE type (otherwise it stays ambiguous and falls
   * through to name-matching).
   */
  private resolveRazorUsing(ref: UnresolvedRef): ResolvedRef | null {
    if (ref.referenceName.includes('.') || ref.referenceName.includes('::')) return null;
    const usings = this.getRazorUsings(ref.filePath);
    if (usings.length === 0) return null;
    const found = new Map<string, Node>();
    for (const ns of usings) {
      for (const cand of this.context.getNodesByQualifiedName(`${ns}::${ref.referenceName}`)) {
        found.set(cand.id, cand);
      }
    }
    if (found.size !== 1) return null;
    const target = found.values().next().value!;
    return { original: ref, targetNodeId: target.id, confidence: 0.9, resolvedBy: 'import' };
  }

  /**
   * Resolve a CFML inheritance reference written as a component path (#1152).
   * Two forms exist in real code:
   *
   * - Dotted: `extends="coldbox.system.web.Controller"` — dots are directory
   *   separators from the webroot or a CFML mapping. Mappings live in server
   *   config / Application.cfc, so the leading segments may not exist in the
   *   repo at all (in the coldbox repo itself the path is `system/web/
   *   Controller.cfc` — the `coldbox.` root IS the repo). Matched by final
   *   segment (the class), corroborated right-to-left against the candidate's
   *   parent directories.
   * - Relative: `extends="../base"` / `extends="./base"` (the FW/1 style) —
   *   resolved against the referencing file's own directory.
   *
   * Conservative by design: a candidate needs at least one corroborating
   * directory segment (a dotted path whose only same-named class sits in an
   * unrelated directory is almost always an out-of-repo library supertype —
   * mxunit/testbox/coldbox-as-dependency), and a corroboration tie yields no
   * edge. Directory comparison is case-insensitive (CFML path resolution is);
   * the class segment itself is matched exactly, which real code satisfies —
   * dotted paths are written to match the on-disk file name.
   */
  private resolveCfmlComponentPath(ref: UnresolvedRef): ResolvedRef | null {
    const cfmlCandidates = (name: string): Node[] =>
      this.context
        .getNodesByName(name)
        .filter(
          (n) =>
            (n.kind === 'class' || n.kind === 'interface') &&
            (n.language === 'cfml' || n.language === 'cfscript')
        );
    const norm = (p: string): string => p.replace(/\\/g, '/').toLowerCase();

    // Relative-path form: `../base`, `./base`, `sub/thing` — resolve against
    // the referencing file's directory and require an exact (case-insensitive)
    // file match.
    if (ref.referenceName.includes('/')) {
      const rel = ref.referenceName.replace(/\.cfc$/i, '');
      const fromDir = ref.filePath.replace(/\\/g, '/').split('/').slice(0, -1);
      const parts = [...fromDir];
      for (const seg of rel.split('/')) {
        if (seg === '' || seg === '.') continue;
        if (seg === '..') {
          if (parts.length === 0) return null; // escapes the project root
          parts.pop();
        } else {
          parts.push(seg);
        }
      }
      const wantPath = norm(parts.join('/') + '.cfc');
      const className = parts[parts.length - 1];
      if (!className) return null;
      const target = cfmlCandidates(className).find((c) => norm(c.filePath) === wantPath);
      return target
        ? { original: ref, targetNodeId: target.id, confidence: 0.95, resolvedBy: 'file-path' }
        : null;
    }

    // Dotted form.
    const segments = ref.referenceName.split('.').map((s) => s.trim()).filter(Boolean);
    if (segments.length < 2) return null;
    const className = segments[segments.length - 1]!;
    const dirSegments = segments.slice(0, -1);

    let best: Node | null = null;
    let bestScore = 0;
    let tie = false;
    for (const cand of cfmlCandidates(className)) {
      const dirs = cand.filePath.replace(/\\/g, '/').split('/').slice(0, -1);
      // Count matching directory segments right-to-left: for
      // `coldbox.system.web.Controller` vs `system/web/Controller.cfc`,
      // `web` and `system` match, then the repo root ends the run → score 2.
      let score = 0;
      while (
        score < dirSegments.length &&
        score < dirs.length &&
        dirSegments[dirSegments.length - 1 - score]!.toLowerCase() ===
          dirs[dirs.length - 1 - score]!.toLowerCase()
      ) {
        score++;
      }
      if (score > bestScore) {
        best = cand;
        bestScore = score;
        tie = false;
      } else if (score === bestScore && score > 0) {
        tie = true;
      }
    }
    if (!best || bestScore === 0 || tie) return null;
    return { original: ref, targetNodeId: best.id, confidence: 0.9, resolvedBy: 'qualified-name' };
  }

  /**
   * Resolve a `this.<member>` function-as-value reference (#756/#808) to the
   * ENCLOSING CLASS's own member — never a same-named symbol elsewhere. The
   * registration idiom (`btn.on('click', this.handleClick)`) names a member
   * of the class being defined, so the only valid target shares the
   * from-symbol's qualified-name scope. Function/method targets only — a
   * property (a data field, post-#808 classification) yields no edge — same
   * file required, no fallback of any kind.
   */
  private resolveThisMemberFnRef(ref: UnresolvedRef): ResolvedRef | null {
    const member = ref.referenceName.slice('this.'.length);
    if (!member) return null;
    const fromNode = this.queries.getNodeById(ref.fromNodeId);
    if (!fromNode) return null;
    // A hook declared at class-body level (Ruby `before_action :authenticate`)
    // attributes to the CLASS node itself — its qualified name IS the scope.
    // For members, strip the member segment.
    let classPrefix: string;
    if (SUPERTYPE_BEARING_KINDS.has(fromNode.kind) || fromNode.kind === 'module') {
      classPrefix = fromNode.qualifiedName;
    } else {
      const sep = fromNode.qualifiedName.lastIndexOf('::');
      if (sep <= 0) return null; // not inside a class scope
      classPrefix = fromNode.qualifiedName.slice(0, sep);
    }
    const candidates = this.context
      .getNodesByQualifiedName(`${classPrefix}::${member}`)
      .filter(
        (n) =>
          (n.kind === 'function' || n.kind === 'method') &&
          n.filePath === ref.filePath &&
          n.id !== ref.fromNodeId
      );
    if (candidates.length === 0) {
      // Not on the class itself — possibly INHERITED. implements/extends
      // edges don't exist yet in this pass, so retry in the supertype pass
      // (resolveDeferredThisMemberRefs) instead of giving up.
      this.deferReference(ref, this.deferredThisMemberRefs);
      return null;
    }
    const target = candidates.reduce((a, b) => (a.startLine <= b.startLine ? a : b));
    return {
      original: ref,
      targetNodeId: target.id,
      confidence: 0.95,
      resolvedBy: 'function-ref',
    };
  }

  /**
   * Second pass for `this.<member>` refs whose member wasn't on the enclosing
   * class itself (#808): once implements/extends edges exist, walk the
   * class's supertypes (transitively, depth-capped) and resolve the member on
   * the nearest one that declares it — `this.handleSubmit` registered in a
   * subclass resolves to `FormBase::handleSubmit`. Validated targets only
   * (function/method kind, same language family); no match → no edge.
   * Mirrors resolveChainedCallsViaConformance's lifecycle. Returns the number
   * of newly-created edges.
   */
  async resolveDeferredThisMemberRefs(): Promise<number> {
    const deferred = this.deferredThisMemberRefs;
    this.deferredThisMemberRefs = [];
    if (deferred.length === 0) return 0;

    this.clearCaches();
    // Synchronous main-thread post-pass with a per-ref supertype BFS — yield
    // periodically so the #850 liveness watchdog heartbeat can fire (#1091).
    const maybeYield = createYielder();
    const resolved: ResolvedRef[] = [];
    for (const ref of deferred) {
      await maybeYield();
      const member = ref.referenceName.slice('this.'.length);
      const fromNode = this.queries.getNodeById(ref.fromNodeId);
      if (!fromNode || !member) continue;
      // Class-body-level hooks (Ruby) attribute to the CLASS node itself.
      let className: string;
      if (SUPERTYPE_BEARING_KINDS.has(fromNode.kind) || fromNode.kind === 'module') {
        className = fromNode.name;
      } else {
        const sep = fromNode.qualifiedName.lastIndexOf('::');
        if (sep <= 0) continue;
        const classPrefix = fromNode.qualifiedName.slice(0, sep);
        className = classPrefix.includes('::')
          ? classPrefix.slice(classPrefix.lastIndexOf('::') + 2)
          : classPrefix;
      }

      // NODE-anchored BFS up the supertype graph: start from the class node
      // in the ref's own file (never a same-named class elsewhere — rails has
      // a dozen `Engine`s), follow implements/extends EDGES to supertype
      // NODES, and look members up through `contains` edges. No name-based
      // unions anywhere — a name-keyed getSupertypes('Engine') merged every
      // Engine's parents and produced a cross-class wrong edge on rails.
      let frontierNodes = this.context
        .getNodesByName(className)
        .filter(
          (n) =>
            SUPERTYPE_BEARING_KINDS.has(n.kind) &&
            n.filePath === ref.filePath
        );
      if (frontierNodes.length === 0) {
        // The class itself may be declared in another file (partial/reopened
        // classes); fall back to same-family nodes of that name.
        frontierNodes = this.context
          .getNodesByName(className)
          .filter(
            (n) =>
              SUPERTYPE_BEARING_KINDS.has(n.kind) &&
              sameLanguageFamily(n.language, ref.language)
          );
      }
      const seenNodes = new Set<string>(frontierNodes.map((n) => n.id));
      let target: Node | null = null;
      for (let depth = 0; depth < 5 && frontierNodes.length > 0 && !target; depth++) {
        const next: Node[] = [];
        for (const typeNode of frontierNodes) {
          for (const edge of this.queries.getOutgoingEdges(typeNode.id, ['implements', 'extends'])) {
            const superNode = this.queries.getNodeById(edge.target);
            if (!superNode || seenNodes.has(superNode.id)) continue;
            seenNodes.add(superNode.id);
            if (!SUPERTYPE_BEARING_KINDS.has(superNode.kind)) continue;
            // Member lookup anchored on the supertype's contains edges.
            for (const c of this.queries.getOutgoingEdges(superNode.id, ['contains'])) {
              const m = this.queries.getNodeById(c.target);
              if (
                m &&
                m.name === member &&
                (m.kind === 'function' || m.kind === 'method') &&
                sameLanguageFamily(m.language, ref.language)
              ) {
                target = m;
                break;
              }
            }
            if (target) break;
            next.push(superNode);
          }
          if (target) break;
        }
        frontierNodes = next;
      }

      if (target) {
        resolved.push({
          original: ref,
          targetNodeId: target.id,
          confidence: 0.85,
          resolvedBy: 'function-ref',
        });
      }
    }
    return this.persistDeferredReferences(deferred, resolved);
  }

  private gateLanguage(result: ResolvedRef | null, ref: UnresolvedRef): ResolvedRef | null {
    if (!result) return result;
    const tgt = this.getLanguageFromNodeId(result.targetNodeId);
    if (!tgt || !ref.language) return result;
    if ((ref.referenceKind === 'references' || ref.referenceKind === 'function_ref') && !sameLanguageFamily(tgt, ref.language)) return null;
    if (ref.referenceKind === 'imports' && crossesKnownFamily(tgt, ref.language)) return null;
    return result;
  }

  /**
   * Drop a FRAMEWORK-strategy resolution that crosses two *known* language
   * families for a type-usage (`references`) or import-binding (`imports`)
   * edge. The framework strategy is intentionally ungated for cross-language
   * bridges, but those legitimate bridges are either `calls` edges (RN/Expo
   * JS → native) or config↔code edges whose config side (`yaml`/`blade`/…) is
   * not a known programming-language family. A `references`/`imports` edge
   * between two *known* families is always a coincidental name collision — the
   * React/Svelte/Vue PascalCase component resolvers name-match `getNodesByName`
   * without a language check, so a TS `<TestRunner>` ref happily matched a
   * Kotlin `class TestRunner`. Gating only the both-known-cross-family case
   * lets config bridges and `calls` bridges through untouched.
   */
  private gateFrameworkLanguage(result: ResolvedRef | null, ref: UnresolvedRef): ResolvedRef | null {
    if (!result) return result;
    if (ref.referenceKind !== 'references' && ref.referenceKind !== 'imports') return result;
    const tgt = this.getLanguageFromNodeId(result.targetNodeId);
    // Package imports cannot target prose found by a framework's name lookup.
    if (ref.referenceKind === 'imports' && (tgt as string) === 'markdown' && (ref.language as string) !== 'markdown') return null;
    if (tgt && ref.language && crossesKnownFamily(tgt, ref.language)) return null;
    return result;
  }
}

/**
 * Create a reference resolver instance
 */
export function createResolver(projectRoot: string, queries: QueryBuilder): ReferenceResolver {
  const resolver = new ReferenceResolver(projectRoot, queries);
  resolver.initialize();
  return resolver;
}