sigmap 4.3.0 → 5.1.0

package/AGENTS.md

@@ -12,20 +12,23 @@ Use this marker block for all appendable context files:
 ## Auto-generated signatures
 <!-- Updated by gen-context.js -->
 You are a coding assistant with full knowledge of this codebase.
-Below are the code signatures extracted by SigMap v4.1.2 on 2026-04-16T17:45:07.132Z.
+Below are the code signatures extracted by SigMap v5.1.0 on 2026-04-16T21:33:38.411Z.
 
 Use these signatures to answer questions about the code accurately.
 
 ## Code Signatures
 
-<!-- Generated by SigMap gen-context.js v4.1.2 -->
+<!-- Generated by SigMap gen-context.js v5.1.0 -->
 <!-- DO NOT EDIT below the marker line — run gen-context.js to regenerate -->
 
 # Code signatures
 
-## changes (last 5 commits — 53 minutes ago)
+## changes (last 5 commits — 16 minutes ago)
 ```
-src/retrieval/ranker.js                       +_parseContextFile  +buildSigIndex  ~buildSigIndex  ~rank
+src/config/loader.js                          +loadBaseConfig  ~loadConfig  ~deepClone
+src/format/dashboard.js                       ~computeExtractorCoverage  ~readBenchmarkTrend
+src/judge/judge-engine.js                     +tokenize  +groundedness  +judge
+src/retrieval/ranker.js                       +detectIntent  ~formatRankJSON
 ```
 
 ## packages
@@ -146,9 +149,41 @@ function adapt(context, adapterName, opts = {}) → string
 
 ## src
 
-### src/config/defaults.js
+### src/config/loader.js
 ```
-module.exports = { DEFAULTS }
+module.exports = { loadConfig, loadBaseConfig }
+function loadBaseConfig(extendsVal, cwd)
+function detectAutoSrcDirs(cwd, excludeList) → string[]
+function loadConfig(cwd) → object
+function deepClone(obj)
+```
+
+### src/format/dashboard.js
+```
+module.exports = { generateDashboardHtml, renderHistoryCharts, computeExtractorCoverage, percentile, overBudgetStreak }
+function toNumber(v)
+function percentile(values, p)
+function overBudgetStreak(entries)
+function loadConfig(cwd)
+function shouldExclude(rel, excludeSet)
+function detectLanguage(filePath)
+function walkFiles(dir, maxDepth, depth, out, excludeSet)
+function computeExtractorCoverage(cwd)
+function readBenchmarkTrend(cwd)
+function lineChartSvg(values, title, ySuffix)
+function barChartSvg(perLanguage)
+function sparkline(values)
+function buildDashboardData(cwd, health)
+function generateDashboardHtml(cwd, health)
+function renderHistoryCharts(cwd, health)
+```
+
+### src/judge/judge-engine.js
+```
+module.exports = { groundedness, judge }
+function tokenize(text)
+function groundedness(response, context)
+function judge(response, context, opts = {})
 ```
 
 ### src/mcp/server.js
@@ -162,13 +197,14 @@ function start(cwd)
 
 ### src/retrieval/ranker.js
 ```
-module.exports = { rank, buildSigIndex, scoreFile, formatRankTable, formatRankJSON, DEFAULT_WEIGHTS }
+module.exports = { rank, buildSigIndex, scoreFile, formatRankTable, formatRankJSON, DEFAULT_WEIGHTS, detectIntent }
 function scoreFile(filePath, sigs, queryTokens, weights) → number
 function rank(query, sigIndex, opts) → { file: string, score: nu
 function _parseContextFile(contextPath) → Map<string, string[]>
 function buildSigIndex(cwd, opts) → Map<string, string[]>
 function formatRankTable(results, query) → string
 function formatRankJSON(results, query) → object
+function detectIntent(query)
 ```
 
 ### src/analysis/coverage-score.js
@@ -178,12 +214,9 @@ function coverageScore(cwd, fileEntries, config) → { * score: number, * grad
 function _walk(dir, excludeSet, out)
 ```
 
-### src/config/loader.js
+### src/config/defaults.js
 ```
-module.exports = { loadConfig }
-function detectAutoSrcDirs(cwd, excludeList) → string[]
-function loadConfig(cwd) → object
-function deepClone(obj)
+module.exports = { DEFAULTS }
 ```
 
 ### src/eval/analyzer.js
@@ -527,26 +560,6 @@ function formatCache(content) → string
 function formatCachePayload(content, model) → string
 ```
 
-### src/format/dashboard.js
-```
-module.exports = { generateDashboardHtml, renderHistoryCharts, computeExtractorCoverage, percentile, overBudgetStreak }
-function toNumber(v)
-function percentile(values, p)
-function overBudgetStreak(entries)
-function loadConfig(cwd)
-function shouldExclude(rel, excludeSet)
-function detectLanguage(filePath)
-function walkFiles(dir, maxDepth, depth, out, excludeSet)
-function computeExtractorCoverage(cwd)
-function readBenchmarkTrend(cwd)
-function lineChartSvg(values, title, ySuffix)
-function barChartSvg(perLanguage)
-function sparkline(values)
-function buildDashboardData(cwd, health)
-function generateDashboardHtml(cwd, health)
-function renderHistoryCharts(cwd, health)
-```
-
 ### src/format/llm-txt.js
 ```
 module.exports = { format, outputPath }

package/CHANGELOG.md

@@ -10,6 +10,26 @@ Format: [Semantic Versioning](https://semver.org/)
 
 ---
 
+## [5.1.0] — 2026-04-16
+
+### Added
+
+- **Benchmark history tracking** — all three benchmark scripts (`run-retrieval-benchmark.mjs`, `run-benchmark.mjs`, `run-task-benchmark.mjs`) now append a structured NDJSON entry to `.context/benchmark-history.ndjson` after each run (`type: "retrieval" | "token-reduction" | "task"`).
+- **`sigmap history` benchmark trend rows** — when `.context/benchmark-history.ndjson` exists, `sigmap history` prints a retrieval `hit@5` sparkline row and a token-reduction sparkline row below the usage table. The command no longer exits early when the usage log is empty.
+- **Dashboard `readBenchmarkTrend` uses local history** — `src/format/dashboard.js` now prefers `.context/benchmark-history.ndjson` over the CI-only `benchmarks/results/` directory, so the dashboard hit@5 trend chart populates for all users after running any benchmark locally.
+
+---
+
+## [5.0.0] — 2026-04-16
+
+### Added
+
+- **`sigmap judge --response <file> --context <file>`** — rule-based groundedness scoring engine (`src/judge/judge-engine.js`). Computes a 0–1 score from token overlap between an LLM response and its source context. Exits 0 when verdict is `pass`, exits 1 on `fail`. Supports `--json` (emits `{ score, verdict, reasons }`) and `--threshold` override.
+- **Config `extends`** — `gen-context.config.json` now accepts an `"extends"` key pointing to a local JSON file path or HTTPS URL. The base config is deep-merged (DEFAULTS → base → local), with HTTPS responses cached for 1 hour in `.context/config-cache/`.
+- **`sigmap history [--last N] [--json]`** — displays last N usage log entries as a table with a Unicode sparkline (▁▂▃▄▅▆▇█) for the token trend. Reads from `.context/usage.ndjson` (requires `tracking: true` in config).
+
+---
+
 ## [4.3.0] — 2026-04-16
 
 ### Added

package/gen-context.js

@@ -221,6 +221,47 @@ __factories["./src/config/loader"] = function(module, exports) {
     });
   }
 
+  const BASE_CONFIG_TTL_MS = 60 * 60 * 1000;
+
+  function loadBaseConfig(extendsVal, cwd) {
+    if (!extendsVal || typeof extendsVal !== 'string') return {};
+    if (extendsVal.startsWith('https://') || extendsVal.startsWith('http://')) {
+      const cacheDir  = path.join(cwd, '.context', 'config-cache');
+      const cacheKey  = Buffer.from(extendsVal).toString('base64').replace(/[^a-zA-Z0-9_-]/g, '_');
+      const cachePath = path.join(cacheDir, `${cacheKey}.json`);
+      if (fs.existsSync(cachePath)) {
+        const age = Date.now() - fs.statSync(cachePath).mtimeMs;
+        if (age < BASE_CONFIG_TTL_MS) {
+          try { return JSON.parse(fs.readFileSync(cachePath, 'utf8')); } catch (_) {}
+        }
+      }
+      try {
+        const { execSync } = require('child_process');
+        const proto = extendsVal.startsWith('https') ? 'https' : 'http';
+        const out = execSync(
+          `node -e "const h=require('${proto}');let d='';h.get(${JSON.stringify(extendsVal)},r=>{r.on('data',c=>d+=c);r.on('end',()=>process.stdout.write(d))}).on('error',()=>process.exit(1))"`,
+          { timeout: 10000, encoding: 'utf8' }
+        );
+        const parsed = JSON.parse(out);
+        if (!fs.existsSync(cacheDir)) fs.mkdirSync(cacheDir, { recursive: true });
+        fs.writeFileSync(cachePath, JSON.stringify(parsed), 'utf8');
+        return parsed;
+      } catch (err) {
+        process.stderr.write(`[sigmap] config extends: could not fetch ${extendsVal}: ${err.message}\n`);
+        if (fs.existsSync(cachePath)) {
+          try { return JSON.parse(fs.readFileSync(cachePath, 'utf8')); } catch (_) {}
+        }
+        return {};
+      }
+    }
+    const absPath = path.resolve(cwd, extendsVal);
+    try { return JSON.parse(fs.readFileSync(absPath, 'utf8')); }
+    catch (err) {
+      process.stderr.write(`[sigmap] config extends: could not load ${absPath}: ${err.message}\n`);
+      return {};
+    }
+  }
+
   /**
    * Load and merge configuration for a given working directory.
    *
@@ -250,18 +291,31 @@ __factories["./src/config/loader"] = function(module, exports) {
 
     // Warn on unknown keys (helps catch typos)
     for (const key of Object.keys(userConfig)) {
-      if (key.startsWith('_')) continue; // allow _comment etc.
+      if (key.startsWith('_') || key === 'extends') continue;
       if (!KNOWN_KEYS.has(key)) {
         console.warn(`[sigmap] unknown config key: "${key}" (ignored)`);
       }
     }
 
-    // Deep merge: top-level known keys from user override defaults
-    // For object values (e.g. mcp), merge one level deep
+    // Deep merge: DEFAULTS → base (extends) → user config
+    const baseConfig = loadBaseConfig(userConfig.extends, cwd);
     const merged = deepClone(DEFAULTS);
+
+    for (const key of Object.keys(baseConfig)) {
+      if (key.startsWith('_') || key === 'extends') continue;
+      if (!KNOWN_KEYS.has(key)) continue;
+      const val = baseConfig[key];
+      if (val !== null && typeof val === 'object' && !Array.isArray(val) &&
+          typeof merged[key] === 'object' && !Array.isArray(merged[key])) {
+        merged[key] = Object.assign({}, merged[key], val);
+      } else {
+        merged[key] = val;
+      }
+    }
+
     for (const key of Object.keys(userConfig)) {
-      if (key.startsWith('_')) continue;
-      if (!KNOWN_KEYS.has(key)) continue; // skip unknown keys
+      if (key.startsWith('_') || key === 'extends') continue;
+      if (!KNOWN_KEYS.has(key)) continue;
       const val = userConfig[key];
       if (val !== null && typeof val === 'object' && !Array.isArray(val) &&
           typeof merged[key] === 'object' && !Array.isArray(merged[key])) {
@@ -292,7 +346,7 @@ __factories["./src/config/loader"] = function(module, exports) {
     return JSON.parse(JSON.stringify(obj));
   }
 
-  module.exports = { loadConfig, detectAutoSrcDirs };
+  module.exports = { loadConfig, loadBaseConfig, detectAutoSrcDirs };
 
 };
 
@@ -3098,6 +3152,25 @@ __factories["./src/format/cache"] = function(module, exports) {
     }
 
     function readBenchmarkTrend(cwd) {
+      // Prefer per-user history file written by benchmark scripts
+      const histPath = path.join(cwd, '.context', 'benchmark-history.ndjson');
+      if (fs.existsSync(histPath)) {
+        const histValues = [];
+        try {
+          for (const line of fs.readFileSync(histPath, 'utf8').trim().split('\n').filter(Boolean)) {
+            try {
+              const obj = JSON.parse(line);
+              if (obj.type === 'retrieval') {
+                const v = toNumber(obj.hitAt5Pct);
+                if (v !== null) histValues.push(v);
+              }
+            } catch (_) {}
+          }
+        } catch (_) {}
+        if (histValues.length > 0) return histValues.slice(-30);
+      }
+
+      // Fallback: legacy benchmarks/results directory (CI artifacts)
       const resultDir = path.join(cwd, 'benchmarks', 'results');
       if (!fs.existsSync(resultDir)) return [];
       const files = [];
@@ -4654,7 +4727,7 @@ __factories["./src/mcp/server"] = function(module, exports) {
   
   const SERVER_INFO = {
     name: 'sigmap',
-    version: '4.3.0',
+    version: '5.1.0',
     description: 'SigMap MCP server — code signatures on demand',
   };
   
@@ -5252,6 +5325,61 @@ __factories["./src/security/scanner"] = function(module, exports) {
   
 };
 
+// ── ./src/judge/judge-engine ──
+__factories["./src/judge/judge-engine"] = function(module, exports) {
+  'use strict';
+
+  const STOP = new Set([
+    'the','a','an','in','on','at','to','of','for','and','or','but',
+    'is','are','was','were','be','been','being','have','has','had',
+    'do','does','did','will','would','could','should','may','might',
+    'shall','can','not','with','from','by','as','this','that','it',
+  ]);
+
+  function tokenize(text) {
+    return (text || '').toLowerCase().match(/\b[a-z][a-z0-9_]{2,}\b/g) || [];
+  }
+
+  function groundedness(response, context) {
+    if (!response || !context) return 0;
+    const ctxTokens = new Set(tokenize(context).filter((t) => !STOP.has(t)));
+    if (ctxTokens.size === 0) return 0;
+    const respTokens = tokenize(response).filter((t) => !STOP.has(t));
+    if (respTokens.length === 0) return 0;
+    const matched = respTokens.filter((t) => ctxTokens.has(t));
+    return parseFloat((matched.length / respTokens.length).toFixed(3));
+  }
+
+  const GENERIC_MARKERS = [
+    'however, based on my knowledge',
+    'generally speaking',
+    'in general',
+    'typically,',
+    'usually,',
+    'as a general rule',
+  ];
+
+  function judge(response, context, opts) {
+    opts = opts || {};
+    const score = groundedness(response, context);
+    const threshold = opts.threshold !== undefined ? opts.threshold : 0.25;
+    const reasons = [];
+    if (score < threshold) {
+      reasons.push(`score ${score} is below threshold ${threshold} — response may not be grounded in context`);
+    }
+    if (response) {
+      const lower = response.toLowerCase();
+      for (const m of GENERIC_MARKERS) {
+        if (lower.includes(m)) reasons.push(`response contains generic phrase: "${m}"`);
+      }
+    }
+    const verdict = score >= threshold && reasons.length === 0 ? 'pass' : 'fail';
+    return { score, verdict, reasons };
+  }
+
+  module.exports = { groundedness, judge };
+};
+
 // ── ./src/tracking/logger ──
 __factories["./src/tracking/logger"] = function(module, exports) {
   
@@ -6262,7 +6390,7 @@ const path = require('path');
 const os = require('os');
 const { execSync } = require('child_process');
 
-const VERSION = '4.3.0';
+const VERSION = '5.1.0';
 const MARKER = '\n\n## Auto-generated signatures\n<!-- Updated by gen-context.js -->\n';
 
 function requireSourceOrBundled(key) {
@@ -8313,6 +8441,124 @@ function main() {
     process.exit(0);
   }
 
+  // v5.0: `sigmap judge --response <file> --context <file>` — groundedness scoring
+  if (args[0] === 'judge') {
+    const respIdx = args.indexOf('--response');
+    const ctxIdx  = args.indexOf('--context');
+
+    if (respIdx < 0 || ctxIdx < 0) {
+      console.error('[sigmap] Usage: sigmap judge --response <file> --context <file> [--json] [--threshold 0.25]');
+      process.exit(1);
+    }
+
+    const respFile = (args[respIdx + 1] || '').trim();
+    const ctxFile  = (args[ctxIdx + 1]  || '').trim();
+
+    if (!respFile || respFile.startsWith('--') || !ctxFile || ctxFile.startsWith('--')) {
+      console.error('[sigmap] --response and --context require file paths');
+      process.exit(1);
+    }
+
+    let responseText = '', contextText = '';
+    try { responseText = fs.readFileSync(path.resolve(cwd, respFile), 'utf8'); }
+    catch (e) { console.error(`[sigmap] cannot read --response file: ${e.message}`); process.exit(1); }
+    try { contextText = fs.readFileSync(path.resolve(cwd, ctxFile), 'utf8'); }
+    catch (e) { console.error(`[sigmap] cannot read --context file: ${e.message}`); process.exit(1); }
+
+    const thrIdx = args.indexOf('--threshold');
+    const judgeOpts = thrIdx >= 0 ? { threshold: parseFloat(args[thrIdx + 1]) || 0.25 } : {};
+
+    const { judge: runJudge } = requireSourceOrBundled('./src/judge/judge-engine');
+    const result = runJudge(responseText, contextText, judgeOpts);
+
+    if (args.includes('--json')) {
+      process.stdout.write(JSON.stringify(result) + '\n');
+    } else {
+      const bar = '─'.repeat(44);
+      console.log([
+        bar,
+        ` sigmap judge`,
+        ` Score     : ${result.score}`,
+        ` Verdict   : ${result.verdict}`,
+        result.reasons.length ? ` Reasons   :\n   ${result.reasons.join('\n   ')}` : ` Reasons   : none`,
+        bar,
+      ].join('\n'));
+    }
+    process.exit(result.verdict === 'pass' ? 0 : 1);
+  }
+
+  // v5.0: `sigmap history` — show last N usage log entries with sparkline
+  if (args[0] === 'history') {
+    const { readLog } = requireSourceOrBundled('./src/tracking/logger');
+    const entries = readLog(cwd);
+
+    const nIdx = args.indexOf('--last');
+    const n    = nIdx >= 0 ? (parseInt(args[nIdx + 1], 10) || 10) : 10;
+    const last = entries.slice(-n);
+
+    if (args.includes('--json')) {
+      process.stdout.write(JSON.stringify(last) + '\n');
+      process.exit(0);
+    }
+
+    const SPARK_CHARS = '▁▂▃▄▅▆▇█';
+    function sparkline(values) {
+      if (values.length === 0) return '';
+      const min = Math.min(...values);
+      const max = Math.max(...values);
+      const range = max - min || 1;
+      return values.map((v) => {
+        const idx = Math.round(((v - min) / range) * (SPARK_CHARS.length - 1));
+        return SPARK_CHARS[idx];
+      }).join('');
+    }
+
+    const bar = '─'.repeat(62);
+    console.log(bar);
+    console.log(` sigmap history  (last ${Math.max(last.length, 1)} runs)`);
+    console.log(bar);
+
+    if (last.length === 0) {
+      console.log(' No usage log entries. Enable tracking: true in config to start recording runs.');
+    } else {
+      console.log(` ${'Date'.padEnd(24)} ${'Files'.padStart(5)} ${'Tokens'.padStart(7)} ${'Reduction'.padStart(9)} ${'Budget?'.padStart(7)}`);
+      console.log(` ${'─'.repeat(24)} ${'─'.repeat(5)} ${'─'.repeat(7)} ${'─'.repeat(9)} ${'─'.repeat(7)}`);
+      for (const e of last) {
+        const date = (e.ts || '').slice(0, 19).replace('T', ' ');
+        const files = String(e.fileCount || 0).padStart(5);
+        const tok   = String(e.finalTokens || 0).padStart(7);
+        const red   = `${e.reductionPct || 0}%`.padStart(9);
+        const over  = (e.overBudget ? '  ⚠ yes' : '     no').padStart(7);
+        console.log(` ${date.padEnd(24)} ${files} ${tok} ${red} ${over}`);
+      }
+      console.log(bar);
+      const tokens = last.map((e) => e.finalTokens || 0);
+      console.log(` Token trend: ${sparkline(tokens)}`);
+    }
+
+    // Show benchmark trend row if .context/benchmark-history.ndjson exists
+    const benchHistPath = path.join(cwd, '.context', 'benchmark-history.ndjson');
+    if (fs.existsSync(benchHistPath)) {
+      try {
+        const benchEntries = fs.readFileSync(benchHistPath, 'utf8').trim().split('\n')
+          .map((l) => { try { return JSON.parse(l); } catch (_) { return null; } }).filter(Boolean);
+        const retrieval = benchEntries.filter((e) => e.type === 'retrieval').slice(-n);
+        if (retrieval.length > 0) {
+          const hits = retrieval.map((e) => e.hitAt5Pct || 0);
+          console.log(` hit@5 trend: ${sparkline(hits)}  ${hits.at(-1)}% (latest)`);
+        }
+        const tokenBench = benchEntries.filter((e) => e.type === 'token-reduction').slice(-n);
+        if (tokenBench.length > 0) {
+          const reds = tokenBench.map((e) => e.reduction || e.avgReductionPct || 0);
+          console.log(` tok reduce : ${sparkline(reds)}  ${reds.at(-1)}% (latest)`);
+        }
+      } catch (_) {}
+    }
+
+    console.log(bar);
+    process.exit(0);
+  }
+
   // Feature 6: `sigmap sync` — write all outputs + llms.txt + print compact diff
   if (args[0] === 'sync') {
     try {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sigmap",
-  "version": "4.3.0",
+  "version": "5.1.0",
   "description": "Zero-dependency AI context engine — 97% token reduction. No npm install. Runs on Node 18+.",
   "main": "gen-context.js",
   "exports": {

package/packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sigmap-cli",
-  "version": "4.3.0",
+  "version": "5.1.0",
   "description": "SigMap CLI wrapper — thin adapter for programmatic CLI invocation",
   "main": "index.js",
   "keywords": [

package/packages/core/package.json

@@ -1,6 +1,6 @@
 {
   "name": "sigmap-core",
-  "version": "4.3.0",
+  "version": "5.1.0",
   "description": "SigMap core library — zero-dependency code signature extraction, retrieval, and security scanning",
   "main": "index.js",
   "keywords": [

package/src/config/loader.js

@@ -4,6 +4,65 @@ const fs = require('fs');
 const path = require('path');
 const { DEFAULTS } = require('./defaults');
 
+const BASE_CONFIG_TTL_MS = 60 * 60 * 1000; // 1 hour
+
+function loadBaseConfig(extendsVal, cwd) {
+  if (!extendsVal || typeof extendsVal !== 'string') return {};
+
+  if (extendsVal.startsWith('https://') || extendsVal.startsWith('http://')) {
+    const cacheDir  = path.join(cwd, '.context', 'config-cache');
+    const cacheKey  = Buffer.from(extendsVal).toString('base64url').replace(/[^a-zA-Z0-9_-]/g, '_');
+    const cachePath = path.join(cacheDir, `${cacheKey}.json`);
+
+    if (fs.existsSync(cachePath)) {
+      const age = Date.now() - fs.statSync(cachePath).mtimeMs;
+      if (age < BASE_CONFIG_TTL_MS) {
+        try { return JSON.parse(fs.readFileSync(cachePath, 'utf8')); } catch (_) {}
+      }
+    }
+
+    try {
+      const https = require('https');
+      const http  = require('http');
+      const mod   = extendsVal.startsWith('https://') ? https : http;
+      const raw   = (() => {
+        let data = '';
+        return new Promise((resolve, reject) => {
+          mod.get(extendsVal, (res) => {
+            res.on('data', (c) => { data += c; });
+            res.on('end', () => resolve(data));
+          }).on('error', reject);
+        });
+      })();
+      // sync fallback: use execSync with node -e
+      const { execSync } = require('child_process');
+      const out = execSync(
+        `node -e "const h=require('${extendsVal.startsWith('https') ? 'https' : 'http'}');let d='';h.get(${JSON.stringify(extendsVal)},r=>{r.on('data',c=>d+=c);r.on('end',()=>process.stdout.write(d))}).on('error',()=>process.exit(1))"`,
+        { timeout: 10000, encoding: 'utf8' }
+      );
+      const parsed = JSON.parse(out);
+      if (!fs.existsSync(cacheDir)) fs.mkdirSync(cacheDir, { recursive: true });
+      fs.writeFileSync(cachePath, JSON.stringify(parsed), 'utf8');
+      return parsed;
+    } catch (err) {
+      process.stderr.write(`[sigmap] config extends: could not fetch ${extendsVal}: ${err.message}\n`);
+      if (fs.existsSync(cachePath)) {
+        try { return JSON.parse(fs.readFileSync(cachePath, 'utf8')); } catch (_) {}
+      }
+      return {};
+    }
+  }
+
+  // Local file path
+  const absPath = path.resolve(cwd, extendsVal);
+  try {
+    return JSON.parse(fs.readFileSync(absPath, 'utf8'));
+  } catch (err) {
+    process.stderr.write(`[sigmap] config extends: could not load ${absPath}: ${err.message}\n`);
+    return {};
+  }
+}
+
 // Keys that are valid in gen-context.config.json
 const KNOWN_KEYS = new Set(Object.keys(DEFAULTS));
 
@@ -173,17 +232,30 @@ function loadConfig(cwd) {
 
   // Warn on unknown keys (helps catch typos)
   for (const key of Object.keys(userConfig)) {
-    if (key.startsWith('_')) continue; // allow _comment etc.
+    if (key.startsWith('_') || key === 'extends') continue;
     if (!KNOWN_KEYS.has(key)) {
       console.warn(`[sigmap] unknown config key: "${key}" (ignored)`);
     }
   }
 
-  // Deep merge: top-level known keys from user override defaults
-  // For object values (e.g. mcp), merge one level deep
+  // Deep merge: DEFAULTS → base (extends) → user config
+  const baseConfig = loadBaseConfig(userConfig.extends, cwd);
   const merged = deepClone(DEFAULTS);
+
+  for (const key of Object.keys(baseConfig)) {
+    if (key.startsWith('_') || key === 'extends') continue;
+    if (!KNOWN_KEYS.has(key)) continue;
+    const val = baseConfig[key];
+    if (val !== null && typeof val === 'object' && !Array.isArray(val) &&
+        typeof merged[key] === 'object' && !Array.isArray(merged[key])) {
+      merged[key] = Object.assign({}, merged[key], val);
+    } else {
+      merged[key] = val;
+    }
+  }
+
   for (const key of Object.keys(userConfig)) {
-    if (key.startsWith('_')) continue;
+    if (key.startsWith('_') || key === 'extends') continue;
     if (!KNOWN_KEYS.has(key)) continue; // skip unknown keys
     const val = userConfig[key];
     if (val !== null && typeof val === 'object' && !Array.isArray(val) &&
@@ -214,4 +286,4 @@ function deepClone(obj) {
   return JSON.parse(JSON.stringify(obj));
 }
 
-module.exports = { loadConfig };
+module.exports = { loadConfig, loadBaseConfig };

package/src/format/dashboard.js

@@ -140,6 +140,26 @@ function computeExtractorCoverage(cwd) {
 }
 
 function readBenchmarkTrend(cwd) {
+  // Prefer per-user history file written by benchmark scripts
+  const histPath = path.join(cwd, '.context', 'benchmark-history.ndjson');
+  if (fs.existsSync(histPath)) {
+    const values = [];
+    try {
+      const lines = fs.readFileSync(histPath, 'utf8').trim().split('\n').filter(Boolean);
+      for (const line of lines) {
+        try {
+          const obj = JSON.parse(line);
+          if (obj.type === 'retrieval') {
+            const v = toNumber(obj.hitAt5Pct);
+            if (v !== null) values.push(v);
+          }
+        } catch (_) {}
+      }
+    } catch (_) {}
+    if (values.length > 0) return values.slice(-30);
+  }
+
+  // Fallback: legacy benchmarks/results directory (CI artifacts)
   const resultDir = path.join(cwd, 'benchmarks', 'results');
   if (!fs.existsSync(resultDir)) return [];
 

package/src/judge/judge-engine.js

@@ -0,0 +1,55 @@
+'use strict';
+
+const STOP = new Set([
+  'the','a','an','in','on','at','to','of','for','and','or','but',
+  'is','are','was','were','be','been','being','have','has','had',
+  'do','does','did','will','would','could','should','may','might',
+  'shall','can','not','with','from','by','as','this','that','it',
+]);
+
+function tokenize(text) {
+  return (text || '').toLowerCase().match(/\b[a-z][a-z0-9_]{2,}\b/g) || [];
+}
+
+function groundedness(response, context) {
+  if (!response || !context) return 0;
+  const ctxTokens = new Set(tokenize(context).filter((t) => !STOP.has(t)));
+  if (ctxTokens.size === 0) return 0;
+  const respTokens = tokenize(response).filter((t) => !STOP.has(t));
+  if (respTokens.length === 0) return 0;
+  const matched = respTokens.filter((t) => ctxTokens.has(t));
+  return parseFloat((matched.length / respTokens.length).toFixed(3));
+}
+
+const GENERIC_MARKERS = [
+  'however, based on my knowledge',
+  'generally speaking',
+  'in general',
+  'typically,',
+  'usually,',
+  'as a general rule',
+];
+
+function judge(response, context, opts = {}) {
+  const score = groundedness(response, context);
+  const threshold = opts.threshold !== undefined ? opts.threshold : 0.25;
+  const reasons = [];
+
+  if (score < threshold) {
+    reasons.push(`score ${score} is below threshold ${threshold} — response may not be grounded in context`);
+  }
+
+  if (response) {
+    const lower = response.toLowerCase();
+    for (const m of GENERIC_MARKERS) {
+      if (lower.includes(m)) {
+        reasons.push(`response contains generic phrase: "${m}"`);
+      }
+    }
+  }
+
+  const verdict = score >= threshold && reasons.length === 0 ? 'pass' : 'fail';
+  return { score, verdict, reasons };
+}
+
+module.exports = { groundedness, judge };

package/src/mcp/server.js

@@ -18,7 +18,7 @@ const { readContext, searchSignatures, getMap, createCheckpoint, getRouting, exp
 
 const SERVER_INFO = {
   name: 'sigmap',
-  version: '4.3.0',
+  version: '5.1.0',
   description: 'SigMap MCP server — code signatures on demand',
 };