Fix production retrieval correctness

CHANGELOG.md

@@ -4,6 +4,13 @@ All notable changes to the TypeScript package will be documented in this file.
 
 ## [Unreleased]
 
+### Changed
+
+- **Default graph discovery is stricter on duplicate/generated paths**: legacy detect and `--spi` now share hard ignores for nested worktrees, VCS metadata, `graphify-out`, dependency stores, and common build/cache outputs, while keeping tests, benchmarks, fixtures, and mocks indexable unless the user excludes them.
+- **Retrieval exclusions are intent-aware**: prompts like "exclude tests" or "do not include benchmarks" no longer classify as `test` intent, and the parsed excluded domains/terms now suppress matching retrieval candidates instead of only affecting wording.
+- **Slice-v1 anchors and traversal are more truthful for production prompts**: literal file-path mentions are distinguished from lexical source-path overlap, explicit `Class.method` prompts anchor the method instead of the class, and pipeline-shaped NestJS prompts now walk backward/forward through controller, service, orchestrator, and persistence paths without exploding into sibling controller methods.
+- **Context-pack diagnostics catch semantically wrong packs**: diagnostics now flag excluded-domain selections, polluted source paths, controller-only pipeline packs, missing method anchors/runtime pipeline evidence, and test-dominated production packs.
+
 ## [0.22.0] - 2026-05-11
 
 ### Added

README.md

@@ -153,6 +153,16 @@ graphify-ts --help                              # full surface
 
 ---
 
+## Default discovery rules
+
+`graphify-ts generate` now hard-ignores nested VCS/worktree copies and generated/build output by default: `.worktrees/`, `worktrees/`, `.git/`, `graphify-out/`, `node_modules/`, `dist/`, `build/`, `coverage/`, cache folders, source maps, lock/build artifacts, and temp/log files.
+
+Tests, benchmarks, fixtures, mocks, and config files are **not** hard-ignored anymore. They still get indexed so retrieval can use them when you ask for them, but production/runtime prompts now soft-penalize them and honor prompt exclusions like "exclude tests, benchmarks, fixtures".
+
+`.graphifyignore` still adds extra ignore rules, and negated entries such as `!vendor/**` or `!lib/**` can re-include a default hard-ignore when you intentionally want it indexed.
+
+---
+
 ## Trust + limitations
 
 Everything stays local by default. No telemetry, no cloud upload, no API key required.

src/contracts/context-pack-diagnostics.ts

@@ -10,6 +10,8 @@
 // rather than from a model judgement. That keeps the surface fully
 // deterministic and CI-asseratable.
 
+import type { SourceDomain } from '../shared/source-discovery.js'
+
 /** The categories of structural problems a context-pack can exhibit. Each
  *  enum value maps to one rule in computeContextPackDiagnostics. */
 export type ContextPackDiagnosticKind =
@@ -22,6 +24,13 @@ export type ContextPackDiagnosticKind =
   | 'low_avg_match_score'
   | 'orphan_nodes'
   | 'no_graph_signals'
+  | 'excluded_domain_selected'
+  | 'test_dominated_pack'
+  | 'controller_only_pipeline_pack'
+  | 'missing_method_anchor'
+  | 'missing_runtime_pipeline'
+  | 'polluted_source_path_selected'
+  | 'missing_structural_evidence'
 
 export type ContextPackDiagnosticSeverity = 'info' | 'warn' | 'error'
 
@@ -48,6 +57,12 @@ export interface ContextPackQualitySignals {
   /** token_count from the pack as a fraction of task_contract.budget.
    *  Capped at 1.0 for over-budget packs. */
   budget_utilization: number
+  /** Source-domain distribution across selected nodes. */
+  domain_distribution: Partial<Record<SourceDomain, number>>
+  /** Domains the prompt explicitly excluded. */
+  excluded_domains: string[]
+  /** Number of selected nodes from polluted/generated paths. */
+  polluted_source_path_count: number
 }
 
 export interface ContextPackDiagnostics {

src/contracts/context-pack.ts

@@ -1,5 +1,6 @@
 import type { RetrievalGateDecision } from './retrieval-gate.js'
 import type { TaskIntentKind } from './task-intent.js'
+import type { SourceDomain } from '../shared/source-discovery.js'
 
 export type ContextPackTaskKind = 'explain' | 'review' | 'impact'
 
@@ -93,6 +94,7 @@ export interface ContextPackNode {
   framework?: string | undefined
   framework_role?: string | undefined
   framework_boost?: number | undefined
+  source_domain?: SourceDomain | undefined
   evidence_class?: ContextPackEvidenceClass | undefined
   representation_type?: ContextRepresentationType | undefined
   representation_reason?: string | undefined

src/contracts/retrieval-gate.ts

@@ -19,11 +19,23 @@ export type RetrievalIntent =
   | 'chitchat'
   | 'unknown'
 
+export type RetrievalExcludedDomain =
+  | 'test'
+  | 'benchmark'
+  | 'fixture'
+  | 'generated'
+  | 'docs'
+  | 'config'
+  | 'build_artifact'
+
 export interface RetrievalGateSignals {
   has_pr_diff: boolean
   has_stack_trace: boolean
   mentioned_paths: ReadonlyArray<string>
   mentioned_symbols: ReadonlyArray<string>
+  excluded_domains?: ReadonlyArray<RetrievalExcludedDomain>
+  excluded_terms?: ReadonlyArray<string>
+  excluded_path_hints?: ReadonlyArray<string>
 }
 
 export interface RetrievalGateDecision {

src/pipeline/detect.ts

@@ -1,7 +1,12 @@
-import { Dirent, existsSync, lstatSync, mkdirSync, readFileSync, readdirSync, realpathSync, statSync, writeFileSync } from 'node:fs'
+import { Dirent, lstatSync, mkdirSync, readFileSync, readdirSync, realpathSync, statSync, writeFileSync } from 'node:fs'
 import { basename, dirname, extname, relative, resolve, sep } from 'node:path'
 
 import { sidecarAwareFileFingerprint } from '../shared/binary-ingest-sidecar.js'
+import {
+  isDiscoveryPathIgnored,
+  isIgnoredByPatterns,
+  loadGraphifyignorePatterns,
+} from '../shared/source-discovery.js'
 
 export const FileType = {
   CODE: 'code',
@@ -109,44 +114,20 @@ const SKIP_DIRS = new Set([
   '.venv',
   'env',
   '.env',
-  'graphify-out',
-  'node_modules',
   '__pycache__',
-  '.git',
-  'dist',
-  'build',
   'target',
-  'out',
   'site-packages',
   'lib64',
   '.pytest_cache',
   '.mypy_cache',
   '.ruff_cache',
   '.tox',
   '.eggs',
-  'test',
-  'tests',
-  '__tests__',
-  'spec',
-  'specs',
-  'e2e',
-  'cypress',
-  'playwright',
-  'coverage',
   'storybook-static',
-  'fixtures',
-  '__fixtures__',
-  '__mocks__',
-  'mocks',
 ])
 
 const NOISE_FILE_PATTERNS: RegExp[] = [
-  /\.(test|spec)\.(ts|tsx|js|jsx|mjs|cjs)$/i,
   /\.stories\.(ts|tsx|js|jsx)$/i,
-  /\.mock\.(ts|tsx|js|jsx)$/i,
-  /^(vitest|jest|webpack|rollup|vite|babel)\.config\./i,
-  /^setupTests\.(ts|tsx|js|jsx)$/i,
-  /^jest\.setup\.(ts|tsx|js|jsx)$/i,
 ]
 
 function isNoiseFile(name: string): boolean {
@@ -157,21 +138,6 @@ function toPosixPath(path: string): string {
   return path.split(sep).join('/')
 }
 
-function globToRegExp(pattern: string): RegExp {
-  const wildcardCount = [...pattern].filter((character) => character === '*').length
-  if (pattern.length > 512 || wildcardCount > 32) {
-    return /^$/
-  }
-
-  const escaped = pattern.replace(/[.+^${}()|[\]\\]/g, '\\$&')
-  const wildcarded = escaped.replace(/\*/g, '.*').replace(/\?/g, '.')
-  return new RegExp(`^${wildcarded}$`)
-}
-
-function matchesPattern(value: string, pattern: string): boolean {
-  return globToRegExp(pattern).test(value)
-}
-
 function isNoiseDir(part: string): boolean {
   return SKIP_DIRS.has(part) || part.endsWith('_venv') || part.endsWith('_env') || part.endsWith('.egg-info')
 }
@@ -234,53 +200,11 @@ export function countWords(path: string): number {
 }
 
 export function _loadGraphifyignore(root: string): string[] {
-  try {
-    const content = readFileSync(resolve(root, '.graphifyignore'), 'utf8')
-    return content
-      .split(/\r?\n/)
-      .map((line) => line.trim())
-      .filter((line) => line.length > 0 && !line.startsWith('#'))
-  } catch {
-    return []
-  }
+  return loadGraphifyignorePatterns(root)
 }
 
 export function _isIgnored(path: string, root: string, patterns: string[]): boolean {
-  if (patterns.length === 0) {
-    return false
-  }
-
-  const relativePath = toPosixPath(relative(resolve(root), resolve(path)))
-  if (relativePath.startsWith('..')) {
-    return false
-  }
-
-  const pathParts = relativePath.split('/')
-  const fileName = basename(path)
-
-  for (const rawPattern of patterns) {
-    const pattern = rawPattern.replace(/^\/+|\/+$/g, '')
-    if (!pattern) {
-      continue
-    }
-
-    if (matchesPattern(relativePath, pattern) || matchesPattern(fileName, pattern)) {
-      return true
-    }
-
-    for (let index = 0; index < pathParts.length; index += 1) {
-      const part = pathParts[index]
-      if (!part) {
-        continue
-      }
-      const prefix = pathParts.slice(0, index + 1).join('/')
-      if (matchesPattern(part, pattern) || matchesPattern(prefix, pattern)) {
-        return true
-      }
-    }
-  }
-
-  return false
+  return isIgnoredByPatterns(path, root, patterns)
 }
 
 function isWithinRoot(rootRealPath: string, candidateRealPath: string): boolean {
@@ -312,7 +236,7 @@ function visitDirectory(
       continue
     }
 
-    if (_isIgnored(entryPath, root, ignorePatterns)) {
+    if (isDiscoveryPathIgnored(entryPath, root, ignorePatterns)) {
       continue
     }
 
@@ -390,11 +314,6 @@ function collectFiles(root: string, followSymlinks: boolean, ignorePatterns: str
 
   visitDirectory(resolvedRoot, resolvedRoot, followSymlinks, ignorePatterns, [rootRealPath], rootRealPath, files)
 
-  const memoryDir = resolve(resolvedRoot, 'graphify-out', 'memory')
-  if (existsSync(memoryDir)) {
-    visitDirectory(memoryDir, resolvedRoot, followSymlinks, ignorePatterns, [rootRealPath], rootRealPath, files)
-  }
-
   return [...new Set(files)].sort()
 }
 

src/pipeline/spi/build.ts

@@ -50,6 +50,10 @@ import {
 import { dirname, extname, join, relative, resolve } from 'node:path'
 import ts from 'typescript'
 
+import {
+  isDiscoveryPathIgnored,
+  loadGraphifyignorePatterns,
+} from '../../shared/source-discovery.js'
 import type {
   SemanticProgramIndex,
   SpiDiagnostic,
@@ -85,19 +89,6 @@ export type BuildSpiOptions = {
 // BuildSpiOptions / buildSpi directly.
 export type BuildSpiFileLayerOptions = BuildSpiOptions
 
-const SKIP_DIRS = new Set<string>([
-  'node_modules',
-  'dist',
-  'build',
-  '.next',
-  'coverage',
-  '.git',
-  'graphify-out',
-  '.test-artifacts',
-  '.turbo',
-  '.vercel',
-])
-
 const EXT_TO_LANG: Record<string, SpiLanguage> = {
   '.ts': 'typescript',
   '.tsx': 'tsx',
@@ -127,7 +118,8 @@ export function buildSpi(opts: BuildSpiOptions): SemanticProgramIndex {
   const diagnostics: SpiDiagnostic[] = []
 
   const absPaths: string[] = []
-  collectFiles(root, absPaths)
+  const ignorePatterns = loadGraphifyignorePatterns(root)
+  collectFiles(root, root, ignorePatterns, absPaths)
 
   const pathToFileId = new Map<string, string>()
   for (const abs of absPaths) {
@@ -199,19 +191,19 @@ export function buildSpi(opts: BuildSpiOptions): SemanticProgramIndex {
 // buildSpi directly.
 export const buildSpiFileLayer = buildSpi
 
-function collectFiles(dir: string, out: string[]): void {
+function collectFiles(root: string, dir: string, ignorePatterns: readonly string[], out: string[]): void {
   let entries: import('node:fs').Dirent<string>[]
   try {
     entries = readdirSync(dir, { withFileTypes: true, encoding: 'utf8' })
   } catch {
     return
   }
   for (const entry of entries) {
-    if (SKIP_DIRS.has(entry.name)) continue
     if (entry.isSymbolicLink()) continue
     const full = join(dir, entry.name)
+    if (isDiscoveryPathIgnored(full, root, ignorePatterns)) continue
     if (entry.isDirectory()) {
-      collectFiles(full, out)
+      collectFiles(root, full, ignorePatterns, out)
     } else if (entry.isFile()) {
       out.push(full)
     }