lens-engine 0.1.20 → 0.1.21

package/cli.js

@@ -3838,6 +3838,79 @@ async function dashboardCommand() {
   openBrowser(url);
 }
 
+// packages/cli/src/commands/eval.ts
+function pct(n) {
+  return `${Math.round(n * 100)}%`;
+}
+function printTable(summary) {
+  const w = process.stdout.write.bind(process.stdout);
+  w("\n  LENS Eval \u2014 Baseline Results\n");
+  w("  \u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\u2550\n\n");
+  w("  Overall\n");
+  w("  \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n");
+  w(`  Queries:        ${summary.total}
+`);
+  w(`  Hit@1:          ${pct(summary.hit_at_1)}
+`);
+  w(`  Hit@3:          ${pct(summary.hit_at_3)}
+`);
+  w(`  Entry@1:        ${pct(summary.entry_hit_at_1)}
+`);
+  w(`  Entry@3:        ${pct(summary.entry_hit_at_3)}
+`);
+  w(`  Recall@5:       ${pct(summary.avg_recall_at_5)}
+`);
+  w(`  Avg Duration:   ${summary.avg_duration_ms}ms
+
+`);
+  if (summary.by_kind.length > 1) {
+    w("  By Kind\n");
+    w("  \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n");
+    for (const k of summary.by_kind) {
+      w(`  ${k.kind} (n=${k.count})
+`);
+      w(
+        `    Hit@1: ${pct(k.hit_at_1)}  Hit@3: ${pct(k.hit_at_3)}  Recall@5: ${pct(k.avg_recall_at_5)}  Avg: ${k.avg_duration_ms}ms
+`
+      );
+    }
+    w("\n");
+  }
+  w("  Per Query\n");
+  w("  \u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\u2500\n");
+  for (const r of summary.results) {
+    const mark = r.hit_at_3 ? "+" : "-";
+    const entryMark = r.entry_hit_at_3 ? "+" : "-";
+    w(`  [${mark}] ${r.id}  ${r.query.slice(0, 50).padEnd(50)}  `);
+    w(
+      `H@1:${r.hit_at_1 ? "Y" : "N"} H@3:${r.hit_at_3 ? "Y" : "N"} E@3:${entryMark} R@5:${pct(r.recall_at_5).padStart(4)} ${r.duration_ms}ms
+`
+    );
+    if (!r.hit_at_3) {
+      w(
+        `        expected: ${r.expected_files.slice(0, 2).map((f) => f.split("/").pop()).join(", ")}
+`
+      );
+      w(
+        `        got top3: ${r.returned_files.slice(0, 3).map((f) => f.split("/").pop()).join(", ")}
+`
+      );
+    }
+  }
+  w("\n");
+}
+async function evalCommand(opts) {
+  const { repo_id } = await ensureRepo();
+  const body = { repo_id };
+  if (opts.kind) body.filter_kind = opts.kind;
+  const summary = await post("/eval/run", body);
+  if (opts.json) {
+    output(summary, true);
+  } else {
+    printTable(summary);
+  }
+}
+
 // packages/cli/src/util/progress.ts
 var dim = (s) => `\x1B[2m${s}\x1B[0m`;
 var green = (s) => `\x1B[32m${s}\x1B[0m`;
@@ -4018,8 +4091,8 @@ async function showProgress(repoId, name, timeoutMs = 18e5) {
     await sleep2(RENDER_INTERVAL);
   }
 }
-function createBar(pct, width) {
-  const clamped = Math.min(pct, 100);
+function createBar(pct2, width) {
+  const clamped = Math.min(pct2, 100);
   const filled = Math.round(clamped / 100 * width);
   const bar = "\u2588".repeat(filled) + "\u2591".repeat(width - filled);
   return `\x1B[32m${bar}\x1B[0m`;
@@ -4813,6 +4886,7 @@ repo.command("watch-status").description("Show file watcher status for current r
 repo.command("mcp").description("Write .mcp.json for MCP agent integration").action(() => run(() => mcpCommand()));
 program2.command("context <goal>").description("Build an intelligent context pack for a goal").option("--json", "Output as JSON", false).action((goal, opts) => run(() => contextCommand(goal, opts), "context"));
 program2.command("index").description("Index the current repo").option("--json", "Output as JSON", false).option("--force", "Full re-scan (default: diff scan, changed files only)", false).option("--status", "Show index status", false).action((opts) => run(() => indexCommand(opts), "index"));
+program2.command("eval").description("Run evaluation harness against current repo").option("--json", "Output as JSON", false).option("--kind <kind>", "Filter by query kind (natural, symbol, error_message)").action((opts) => run(() => evalCommand(opts), "eval"));
 program2.command("status").description("Show repo index/embedding status").option("--json", "Output as JSON", false).action((opts) => run(() => statusCommand(opts), "status"));
 var daemon = program2.command("daemon").description("Daemon management");
 daemon.command("stats").description("Show global daemon statistics").option("--json", "Output as JSON", false).action((opts) => run(() => daemonStatsCommand(opts)));