Upload 2 files

benchmark.py

@@ -0,0 +1,573 @@
+#!/usr/bin/env python3
+"""
+Vortexa Indexer Benchmark — VTX Codebase
+=========================================
+Benchmarks TWO embedding models from the vortexa library on the VTX codebase:
+
+  Model A: VTXAI/vtx-embed-1M   (LF4Embedder — 4-bit static nano, ~1 MB, fastest)
+  Model B: VTXAI/vtx-embed-7M   (LF4Embedder — 4-bit static mini, ~7 MB, default)
+
+Metrics reported per model
+  - Indexing time (ms)
+  - Peak RAM usage during index (MB)
+  - Number of files indexed
+  - Number of chunks produced
+  - Chunk config (size / overlap)
+  - Model size / embedding dim
+  - Per-query latency (mean / p50 / p95 / p99)
+  - R@1  — query-level: was the ground-truth file ranked #1?
+  - R@5  — query-level: was the ground-truth file in top-5?
+  - R@10 — query-level: was the ground-truth file in top-10?
+
+30-query test set is hand-crafted from real VTX source symbols and concepts.
+
+Usage:
+    uv run python benchmark.py
+    uv run python benchmark.py --root /path/to/other/codebase
+    uv run python benchmark.py --alpha 0.7   # override hybrid weight
+    uv run python benchmark.py --no-m2v      # skip Model2Vec (faster)
+"""
+
+from __future__ import annotations
+
+import argparse
+import gc
+import json
+import shutil
+import sys
+import tempfile
+import time
+from dataclasses import asdict, dataclass, field
+from pathlib import Path
+from typing import Any
+
+# ── stdlib resource tracking ─────────────────────────────────────────────────
+try:
+    import resource as _resource
+
+    def _peak_rss_mb() -> float:
+        """Peak resident-set-size in MB (Linux/macOS)."""
+        return _resource.getrusage(_resource.RUSAGE_SELF).ru_maxrss / 1024
+
+except ImportError:  # Windows
+
+    def _peak_rss_mb() -> float:
+        try:
+            import psutil
+
+            return psutil.Process().memory_info().rss / 1024 / 1024
+        except Exception:
+            return 0.0
+
+
+# ── vortexa imports ───────────────────────────────────────────────────────────
+try:
+    from vortexa.core.embedding import LF4Embedder, Model2VecEmbedder
+    from vortexa.core.indexer import CodebaseIndexer
+    from vortexa.core.types import ChunkConfig, IndexStats
+except ModuleNotFoundError as exc:
+    sys.exit(
+        f"\n[ERROR] Could not import vortexa: {exc}\n\n"
+        "vortexa must be available in the SAME Python environment used to run this\n"
+        "script. The simplest ways:\n\n"
+        "  # Recommended — uv injects vortexa into a fresh env at runtime:\n"
+        "  uv run --with vortexa python benchmark.py\n\n"
+        "  # Or add it to the project venv permanently:\n"
+        "  uv add --dev vortexa\n"
+        "  uv run python benchmark.py\n\n"
+        "Do NOT run with .venv/bin/python directly; that venv lacks vortexa.\n"
+    )
+
+# ═════════════════════════════════════════════════════════════════════════════
+# 30-Query Test Suite
+# Each entry: (query_text, expected_file_substring_or_None, description)
+# expected_file = None  →  query is "open", we still measure latency but skip R@k
+# expected_file = str   →  ground truth: result.chunk.file_path must CONTAIN this
+# ═════════════════════════════════════════════════════════════════════════════
+QUERIES: list[tuple[str, str | None, str]] = [
+    # ── Core agent runner ──────────────────────────────────────────────────
+    ("agent runner main loop execution", "agent_runner", "Agent runner entry point"),
+    ("how does the agent run tasks asynchronously", "agent_runner", "Async task execution"),
+    # ── CLI ───────────────────────────────────────────────────────────────
+    ("command line interface argument parsing", "cli", "CLI argument parsing"),
+    ("vtx CLI entry point argparse", "cli", "CLI argparse setup"),
+    # ── Config ────────────────────────────────────────────────────────────
+    ("configuration loading and defaults", "config", "Config loading"),
+    ("user config file path resolution", "config", "Config path resolution"),
+    # ── LLM / models ──────────────────────────────────────────────────────
+    ("LLM API call streaming response", "llm", "LLM streaming"),
+    ("model provider selection gemini openai", "llm", "Model provider"),
+    ("token counting and context window management", "llm", "Token counting"),
+    # ── Context / governance ──────────────────────────────────────────────
+    ("context window governance truncation", "context_governance", "Context governance"),
+    ("context item priority ranking", "context", "Context priority"),
+    # ── Dispatcher ────────────────────────────────────────────────────────
+    ("tool dispatcher routing function calls", "dispatcher", "Tool dispatcher"),
+    ("dispatch tool call with arguments", "dispatcher", "Tool call dispatch"),
+    # ── Events ────────────────────────────────────────────────────────────
+    ("event system publish subscribe hooks", "events", "Event system"),
+    ("streaming event output to terminal", "events", "Event streaming"),
+    # ── Git integration ───────────────────────────────────────────────────
+    ("git branch creation and switching", "git_branch", "Git branch"),
+    ("github CLI integration shell command", "gh_cli", "GitHub CLI"),
+    # ── Diff display ──────────────────────────────────────────────────────
+    ("diff display unified patch format", "diff_display", "Diff display"),
+    ("show file changes colored diff output", "diff_display", "Colored diff"),
+    # ── Extensions ────────────────────────────────────────────────────────
+    ("plugin extension registration loading", "extensions", "Extension loading"),
+    # ── Skills / builtin ──────────────────────────────────────────────────
+    ("builtin skill definition and registration", "builtin_skill", "Builtin skills"),
+    ("skill slash command frontmatter parsing", "builtin_skill", "Skill frontmatter"),
+    # ── Agents ────────────────────────────────────────────────────────────
+    ("subagent invocation and communication", "agents", "Subagent invocation"),
+    ("agent message passing protocol", "agents", "Agent messaging"),
+    # ── Hooks ─────────────────────────────────────────────────────────────
+    ("pre-commit hook integration", "hooks", "Pre-commit hooks"),
+    # ── Async utilities ───────────────────────────────────────────────────
+    ("async utility helpers gather timeout", "async_utils", "Async utilities"),
+    # ── Headless mode ─────────────────────────────────────────────────────
+    ("headless non-interactive execution mode", "headless", "Headless mode"),
+    # ── General codebase understanding ────────────────────────────────────
+    ("error handling exception retry logic", None, "Generic: error handling (open)"),
+    ("logging setup structured log output", None, "Generic: logging (open)"),
+    ("test fixtures mock patching pytest", "tests", "Test utilities"),
+]
+
+assert len(QUERIES) == 30, f"Expected 30 queries, got {len(QUERIES)}"
+
+
+# ═════════════════════════════════════════════════════════════════════════════
+# Data classes for results
+# ═════════════════════════════════════════════════════════════════════════════
+@dataclass
+class QueryResult:
+    query: str
+    description: str
+    expected_file: str | None
+    latency_ms: float
+    top1_file: str | None
+    top5_files: list[str]
+    top10_files: list[str]
+    hit_at_1: bool | None  # None = open query (no ground truth)
+    hit_at_5: bool | None
+    hit_at_10: bool | None
+    top1_score: float
+
+
+@dataclass
+class ModelBenchResult:
+    model_name: str
+    model_label: str
+    embedding_dim: int
+    model_size_mb: float
+    index_dir: str
+
+    # Index stats
+    indexed_files: int = 0
+    total_chunks: int = 0
+    chunk_size: int = 0
+    chunk_overlap: int = 0
+    languages: dict[str, int] = field(default_factory=dict)
+    index_time_ms: float = 0.0
+    peak_ram_mb: float = 0.0
+    memo_hits: int = 0
+    memo_misses: int = 0
+
+    # Search results
+    query_results: list[QueryResult] = field(default_factory=list)
+
+    # Aggregate metrics
+    recall_at_1: float = 0.0
+    recall_at_5: float = 0.0
+    recall_at_10: float = 0.0
+    mean_latency_ms: float = 0.0
+    p50_latency_ms: float = 0.0
+    p95_latency_ms: float = 0.0
+    p99_latency_ms: float = 0.0
+    evaluated_queries: int = 0
+    open_queries: int = 0
+
+
+# ═════════════════════════════════════════════════════════════════════════════
+# Benchmark runner
+# ═════════════════════════════════════════════════════════════════════════════
+def _check_hit(result_files: list[str], expected: str) -> bool:
+    return any(expected.lower() in f.lower() for f in result_files)
+
+
+def run_benchmark(
+    root: Path,
+    model_label: str,
+    embedder: Any,
+    alpha: float | None,
+    top_k: int = 10,
+    verbose: bool = True,
+) -> ModelBenchResult:
+    model_name = getattr(embedder, "_model_id", None) or getattr(
+        embedder, "_model_name", "unknown"
+    )
+    if verbose:
+        print(f"\n{'═' * 60}")
+        print(f"  Benchmarking: {model_label}")
+        print(f"  Model ID   : {model_name}")
+        print(f"{'═' * 60}")
+
+    # Temporary isolated index dir to avoid cache pollution between runs
+    tmp_index = tempfile.mkdtemp(prefix=f"vtx_bench_{model_label.replace(' ', '_')}_")
+
+    chunk_cfg = ChunkConfig(chunk_size=1500, chunk_overlap=200)
+    indexer = CodebaseIndexer(
+        root=root, model=embedder, index_dir=tmp_index, chunk_config=chunk_cfg
+    )
+
+    # ── Index ────────────────────────────────────────────────────────────
+    gc.collect()
+    ram_before = _peak_rss_mb()
+
+    if verbose:
+        print("  [1/3] Indexing codebase …", end="", flush=True)
+
+    t0 = time.perf_counter()
+    stats: IndexStats = indexer.index(force=True, include_text_files=False)
+    index_elapsed = (time.perf_counter() - t0) * 1000
+
+    ram_after = _peak_rss_mb()
+    peak_ram = max(ram_after - ram_before, 0.0)
+
+    if verbose:
+        print(f" done in {index_elapsed:.0f} ms  ({stats.total_chunks} chunks)")
+
+    # Embedding dim and model size
+    emb_dim = embedder.dim
+    model_size_mb = 0.0
+    # Try to get the size from the underlying model
+    if hasattr(embedder, "_model") and embedder._model is not None:
+        raw = embedder._model
+        if hasattr(raw, "model_size_mb"):
+            model_size_mb = raw.model_size_mb
+        elif hasattr(raw, "dim"):
+            model_size_mb = 0.0  # model2vec doesn't expose size easily
+
+    result = ModelBenchResult(
+        model_name=model_name,
+        model_label=model_label,
+        embedding_dim=emb_dim,
+        model_size_mb=model_size_mb,
+        index_dir=tmp_index,
+        indexed_files=stats.indexed_files,
+        total_chunks=stats.total_chunks,
+        chunk_size=chunk_cfg.chunk_size,
+        chunk_overlap=chunk_cfg.chunk_overlap,
+        languages=dict(stats.languages),
+        index_time_ms=round(index_elapsed, 1),
+        peak_ram_mb=round(peak_ram, 1),
+        memo_hits=stats.memo_hits,
+        memo_misses=stats.memo_misses,
+    )
+
+    # ── Search — 30 queries ───────────────────────────────────────────────
+    if verbose:
+        print(f"  [2/3] Running {len(QUERIES)} queries …")
+
+    latencies: list[float] = []
+    hits1: list[bool] = []
+    hits5: list[bool] = []
+    hits10: list[bool] = []
+    open_count = 0
+
+    for i, (query, expected_file, description) in enumerate(QUERIES):
+        t_q = time.perf_counter()
+        search_results = indexer.search(query, top_k=top_k, alpha=alpha)
+        q_elapsed = (time.perf_counter() - t_q) * 1000
+        latencies.append(q_elapsed)
+
+        top_files_all = [r.chunk.file_path for r in search_results]
+        top1_file = top_files_all[0] if top_files_all else None
+        top5_files = top_files_all[:5]
+        top10_files = top_files_all[:10]
+        top1_score = search_results[0].score if search_results else 0.0
+
+        if expected_file is None:
+            hit1 = hit5 = hit10 = None
+            open_count += 1
+        else:
+            hit1 = _check_hit(top_files_all[:1], expected_file)
+            hit5 = _check_hit(top5_files, expected_file)
+            hit10 = _check_hit(top10_files, expected_file)
+            hits1.append(hit1)
+            hits5.append(hit5)
+            hits10.append(hit10)
+
+        qr = QueryResult(
+            query=query,
+            description=description,
+            expected_file=expected_file,
+            latency_ms=round(q_elapsed, 2),
+            top1_file=top1_file,
+            top5_files=top5_files,
+            top10_files=top10_files,
+            hit_at_1=hit1,
+            hit_at_5=hit5,
+            hit_at_10=hit10,
+            top1_score=round(top1_score, 4),
+        )
+        result.query_results.append(qr)
+
+        if verbose:
+            status = ("✓" if hit1 else "~" if hit5 else "✗") if expected_file else "○"
+            print(
+                f"    [{i + 1:02d}/{len(QUERIES)}] {status} {description[:42]:<42} "
+                f"{q_elapsed:6.1f}ms  score={top1_score:.3f}"
+            )
+
+    # ── Aggregate ─────────────────────────────────────────────────────────
+    import statistics
+
+    n_eval = len(hits1)
+    result.evaluated_queries = n_eval
+    result.open_queries = open_count
+    result.recall_at_1 = round(sum(hits1) / n_eval, 4) if n_eval else 0.0
+    result.recall_at_5 = round(sum(hits5) / n_eval, 4) if n_eval else 0.0
+    result.recall_at_10 = round(sum(hits10) / n_eval, 4) if n_eval else 0.0
+    result.mean_latency_ms = round(statistics.mean(latencies), 2)
+    result.p50_latency_ms = round(statistics.median(latencies), 2)
+    sorted_lat = sorted(latencies)
+    result.p95_latency_ms = round(sorted_lat[int(0.95 * len(sorted_lat)) - 1], 2)
+    result.p99_latency_ms = round(sorted_lat[int(0.99 * len(sorted_lat)) - 1], 2)
+
+    if verbose:
+        print("  [3/3] Cleanup temporary index …")
+
+    shutil.rmtree(tmp_index, ignore_errors=True)
+
+    return result
+
+
+# ═════════════════════════════════════════════════════════════════════════════
+# Rich terminal report
+# ═════════════════════════════════════════════════════════════════════════════
+_RESET = "\033[0m"
+_BOLD = "\033[1m"
+_GREEN = "\033[92m"
+_RED = "\033[91m"
+_YELLOW = "\033[93m"
+_CYAN = "\033[96m"
+_MAGENTA = "\033[95m"
+_DIM = "\033[2m"
+
+
+def _pct(v: float) -> str:
+    colour = _GREEN if v >= 0.8 else (_YELLOW if v >= 0.5 else _RED)
+    return f"{colour}{v * 100:.1f}%{_RESET}"
+
+
+def _ms(v: float) -> str:
+    colour = _GREEN if v < 50 else (_YELLOW if v < 200 else _RED)
+    return f"{colour}{v:.1f}ms{_RESET}"
+
+
+def print_report(results: list[ModelBenchResult], alpha: float | None) -> None:
+    print(f"\n{_BOLD}{'═' * 70}{_RESET}")
+    print(f"{_BOLD}{'  VORTEXA BENCHMARK REPORT':^70}{_RESET}")
+    print(f"{_BOLD}{'═' * 70}{_RESET}")
+    print("  Codebase : VTX (vtx-coding-agent)")
+    n_eval_q = sum(1 for q in QUERIES if q[1] is not None)
+    n_open_q = sum(1 for q in QUERIES if q[1] is None)
+    print(f"  Queries  : {len(QUERIES)} total  ({n_eval_q} evaluated, {n_open_q} open)")
+    print(f"  alpha    : {'adaptive (auto)' if alpha is None else alpha}")
+    print()
+
+    # ── Summary table ─────────────────────────────────────────────────────
+    col = 26
+    hdr = f"{'Metric':<{col}}"
+    for r in results:
+        hdr += f"  {_BOLD}{r.model_label[:22]:<22}{_RESET}"
+    print(hdr)
+    print(f"{'─' * (col + len(results) * 24)}")
+
+    def row(label: str, vals: list[str]) -> None:
+        line = f"{label:<{col}}"
+        for v in vals:
+            line += f"  {v:<22}"
+        print(line)
+
+    # Index metrics
+    row("Model ID", [f"{_DIM}{r.model_name[:20]}{_RESET}" for r in results])
+    row("Embedding dim", [str(r.embedding_dim) for r in results])
+    row(
+        "Model size (MB)",
+        [f"{r.model_size_mb:.1f}" if r.model_size_mb else "n/a" for r in results],
+    )
+    row("Chunk size / overlap", [f"{r.chunk_size} / {r.chunk_overlap}" for r in results])
+    row("Files indexed", [str(r.indexed_files) for r in results])
+    row("Total chunks", [str(r.total_chunks) for r in results])
+    row("Memo hits / misses", [f"{r.memo_hits} / {r.memo_misses}" for r in results])
+    row("Index time", [_ms(r.index_time_ms) for r in results])
+    row("Peak RAM delta (MB)", [f"{r.peak_ram_mb:.1f}" for r in results])
+    print(f"{'─' * (col + len(results) * 24)}")
+
+    # Retrieval metrics
+    row("R@1  (evaluated)", [_pct(r.recall_at_1) for r in results])
+    row("R@5  (evaluated)", [_pct(r.recall_at_5) for r in results])
+    row("R@10 (evaluated)", [_pct(r.recall_at_10) for r in results])
+    print(f"{'─' * (col + len(results) * 24)}")
+
+    # Latency
+    row("Latency mean", [_ms(r.mean_latency_ms) for r in results])
+    row("Latency p50", [_ms(r.p50_latency_ms) for r in results])
+    row("Latency p95", [_ms(r.p95_latency_ms) for r in results])
+    row("Latency p99", [_ms(r.p99_latency_ms) for r in results])
+    print(f"{'─' * (col + len(results) * 24)}")
+
+    # Languages
+    for r in results:
+        langs_str = ", ".join(
+            f"{k}:{v}" for k, v in sorted(r.languages.items(), key=lambda x: -x[1])[:5]
+        )
+        row("Top languages", [langs_str if rr is r else "" for rr in results])
+        break  # Only one row for language overview
+
+    print()
+
+    # ── Per-query breakdown ───────────────────────────────────────────────
+    print(f"{_BOLD}Per-Query Breakdown{_RESET}")
+    print(f"{'─' * 100}")
+    header = f"{'#':<3} {'Description':<44} {'Expected':<18}"
+    for _r in results:
+        header += f"  {'@1 @5 @10':^12}  {'ms':>6}"
+    print(header)
+    print(f"{'─' * 100}")
+
+    for i, (_query, expected, desc) in enumerate(QUERIES):
+        line = f"{i + 1:<3} {desc[:43]:<44} {(expected or '(open)')[:17]:<18}"
+        for r in results:
+            qr = r.query_results[i]
+            if qr.hit_at_1 is None:
+                hits = f"{'○':^4} {'○':^4} {'○':^4}"
+            else:
+                h1 = f"{_GREEN}✓{_RESET}" if qr.hit_at_1 else f"{_RED}✗{_RESET}"
+                h5 = f"{_GREEN}✓{_RESET}" if qr.hit_at_5 else f"{_RED}✗{_RESET}"
+                h10 = f"{_GREEN}✓{_RESET}" if qr.hit_at_10 else f"{_RED}✗{_RESET}"
+                hits = f"{h1:<4} {h5:<4} {h10:<4}"
+            line += f"  {hits}  {qr.latency_ms:6.1f}"
+        print(line)
+
+    print(f"{'─' * 100}")
+    print()
+
+
+# ═════════════════════════════════════════════════════════════════════════════
+# Save JSON report
+# ═════════════════════════════════════════════════════════════════════════════
+def save_json_report(results: list[ModelBenchResult], output_path: Path) -> None:
+    report = []
+    for r in results:
+        d = asdict(r)
+        report.append(d)
+    output_path.write_text(json.dumps(report, indent=2))
+    print(f"  JSON report saved → {output_path}")
+
+
+# ═════════════════════════════════════════════════════════════════════════════
+# Main
+# ═════════════════════════════════════════════════════════════════════════════
+def main() -> None:
+    parser = argparse.ArgumentParser(
+        description="Benchmark vortexa indexers on the VTX codebase.",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+    )
+    parser.add_argument(
+        "--root",
+        type=Path,
+        default=Path(__file__).parent,
+        help="Codebase root to index (default: this file's directory)",
+    )
+    parser.add_argument(
+        "--alpha",
+        type=float,
+        default=None,
+        help="Hybrid search alpha (0=BM25, 1=semantic, None=adaptive)",
+    )
+    parser.add_argument(
+        "--top-k", type=int, default=10, help="Max results per query (default: 10)"
+    )
+    parser.add_argument(
+        "--no-m2v", action="store_true", help="Skip Model2Vec benchmark (faster, fewer deps)"
+    )
+    parser.add_argument("--no-lf4", action="store_true", help="Skip LF4 (vtx-embed-7M) benchmark")
+    parser.add_argument(
+        "--output",
+        type=Path,
+        default=Path("benchmark_results.json"),
+        help="Path for JSON report output",
+    )
+    parser.add_argument("--quiet", action="store_true", help="Suppress per-query progress output")
+    args = parser.parse_args()
+
+    root = args.root.resolve()
+    if not root.exists():
+        sys.exit(f"[ERROR] Root not found: {root}")
+
+    verbose = not args.quiet
+
+    print(f"\n{_BOLD}{_CYAN}Vortexa Indexer Benchmark{_RESET}")
+    print(f"{_DIM}Codebase root: {root}{_RESET}")
+    import importlib.util
+
+    _vsite = importlib.util.find_spec("vortexa")
+    _vloc = Path(_vsite.origin).parent if _vsite else "unknown"
+    print(f"{_DIM}vortexa site : {_vloc}{_RESET}\n")
+
+    models_to_run: list[tuple[str, Any]] = []
+
+    if not args.no_lf4:
+        models_to_run.append(("LF4 vtx-embed-1M (nano)", LF4Embedder("VTXAI/vtx-embed-1M")))
+        models_to_run.append(("LF4 vtx-embed-7M (mini)", LF4Embedder("VTXAI/vtx-embed-7M")))
+
+    if not args.no_m2v:
+        models_to_run.append(
+            ("Model2Vec JARVIS-tool-search", Model2VecEmbedder("AI4free/JARVIS-tool-search-v1"))
+        )
+
+    if not models_to_run:
+        sys.exit("[ERROR] All models disabled — enable at least one.")
+
+    # Pre-warm: touch both models to download weights before we start timing
+    print(f"{_BOLD}Pre-warming models (downloading weights if needed)…{_RESET}")
+    for label, emb in models_to_run:
+        print(f"  loading {label} …", end="", flush=True)
+        t_load = time.perf_counter()
+        _ = emb.dim  # triggers lazy load
+        print(f" ready in {(time.perf_counter() - t_load) * 1000:.0f}ms")
+
+    results: list[ModelBenchResult] = []
+
+    for label, emb in models_to_run:
+        r = run_benchmark(
+            root=root,
+            model_label=label,
+            embedder=emb,
+            alpha=args.alpha,
+            top_k=args.top_k,
+            verbose=verbose,
+        )
+        results.append(r)
+
+    print_report(results, args.alpha)
+    save_json_report(results, args.output)
+
+    # ── Winner callout ────────────────────────────────────────────────────
+    if len(results) > 1:
+        best_r1 = max(results, key=lambda r: r.recall_at_1)
+        best_speed = min(results, key=lambda r: r.mean_latency_ms)
+        r1_pct = f"{best_r1.recall_at_1 * 100:.1f}%"
+        spd_ms = f"{best_speed.mean_latency_ms:.1f}ms mean"
+        print(f"{_BOLD}Winner R@1    :{_RESET} {_GREEN}{best_r1.model_label}{_RESET}  ({r1_pct})")
+        print(
+            f"{_BOLD}Winner Speed  :{_RESET} {_GREEN}{best_speed.model_label}{_RESET}  ({spd_ms})"
+        )
+        print()
+
+
+if __name__ == "__main__":
+    main()

benchmark_results.json

@@ -0,0 +1,1872 @@
+[
+  {
+    "model_name": "VTXAI/vtx-embed-1M",
+    "model_label": "LF4 vtx-embed-1M (nano)",
+    "embedding_dim": 64,
+    "model_size_mb": 0.786432,
+    "index_dir": "/tmp/vtx_bench_LF4_vtx-embed-1M_(nano)_onbo80p2",
+    "indexed_files": 330,
+    "total_chunks": 2040,
+    "chunk_size": 1500,
+    "chunk_overlap": 200,
+    "languages": {
+      "python": 1839,
+      "typescript": 27,
+      "tsx": 129,
+      "javascript": 10,
+      "bash": 16,
+      "css": 13,
+      "powershell": 5
+    },
+    "index_time_ms": 16107.8,
+    "peak_ram_mb": 0.0,
+    "memo_hits": 0,
+    "memo_misses": 2040,
+    "query_results": [
+      {
+        "query": "agent runner main loop execution",
+        "description": "Agent runner entry point",
+        "expected_file": "agent_runner",
+        "latency_ms": 5.53,
+        "top1_file": "Site/src/content/blog/agentic-loop.ts",
+        "top5_files": [
+          "Site/src/content/blog/agentic-loop.ts",
+          "src/vtx/agent_runner.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/loop.py",
+          "benchmark.py"
+        ],
+        "top10_files": [
+          "Site/src/content/blog/agentic-loop.ts",
+          "src/vtx/agent_runner.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/loop.py",
+          "benchmark.py",
+          "src/vtx/runtime.py",
+          "src/vtx/ui/agent_runner.py",
+          "src/vtx/sdk/handoffs.py",
+          "src/vtx/agents/loader.py",
+          "src/vtx/agents/registry.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0216
+      },
+      {
+        "query": "how does the agent run tasks asynchronously",
+        "description": "Async task execution",
+        "expected_file": "agent_runner",
+        "latency_ms": 4.28,
+        "top1_file": "src/vtx/runtime.py",
+        "top5_files": [
+          "src/vtx/runtime.py",
+          "Site/src/content/blog/agentic-loop.ts",
+          "src/vtx/agents/loader.py",
+          "benchmark.py",
+          "src/vtx/tools/task.py"
+        ],
+        "top10_files": [
+          "src/vtx/runtime.py",
+          "Site/src/content/blog/agentic-loop.ts",
+          "src/vtx/agents/loader.py",
+          "benchmark.py",
+          "src/vtx/tools/task.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/prompts/identity.py",
+          "src/vtx/loop.py",
+          "src/vtx/sdk/run_config.py",
+          "src/vtx/sdk/results.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0233
+      },
+      {
+        "query": "command line interface argument parsing",
+        "description": "CLI argument parsing",
+        "expected_file": "cli",
+        "latency_ms": 5.21,
+        "top1_file": "src/vtx/cli.py",
+        "top5_files": [
+          "src/vtx/cli.py",
+          "src/vtx/tools/bash.py",
+          "benchmark.py",
+          "src/vtx/ui/agent_runner.py",
+          "src/vtx/ui/input.py"
+        ],
+        "top10_files": [
+          "src/vtx/cli.py",
+          "src/vtx/tools/bash.py",
+          "benchmark.py",
+          "src/vtx/ui/agent_runner.py",
+          "src/vtx/ui/input.py",
+          "src/vtx/ui/formatting.py",
+          "src/vtx/ui/commands/sessions.py",
+          "src/vtx/llm/tool_parser.py",
+          "src/vtx/llm/sdk/openai.py",
+          "src/vtx/loop.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0172
+      },
+      {
+        "query": "vtx CLI entry point argparse",
+        "description": "CLI argparse setup",
+        "expected_file": "cli",
+        "latency_ms": 4.7,
+        "top1_file": "src/vtx/cli.py",
+        "top5_files": [
+          "src/vtx/cli.py",
+          "benchmark.py",
+          "src/vtx/gh_cli.py",
+          "scripts/show_themes.py",
+          "src/vtx/extensions.py"
+        ],
+        "top10_files": [
+          "src/vtx/cli.py",
+          "benchmark.py",
+          "src/vtx/gh_cli.py",
+          "scripts/show_themes.py",
+          "src/vtx/extensions.py",
+          "src/vtx/session.py",
+          "src/vtx/llm/provider_catalog.py",
+          "src/vtx/turn.py",
+          "src/vtx/diff_display.py",
+          "src/vtx/runtime.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.025
+      },
+      {
+        "query": "configuration loading and defaults",
+        "description": "Config loading",
+        "expected_file": "config",
+        "latency_ms": 4.58,
+        "top1_file": "src/vtx/config.py",
+        "top5_files": [
+          "src/vtx/config.py",
+          "src/vtx/llm/oauth/dynamic.py",
+          "src/vtx/llm/provider_catalog.py",
+          "src/vtx/sdk/run_config.py",
+          "src/vtx/llm/sdk/supercode.py"
+        ],
+        "top10_files": [
+          "src/vtx/config.py",
+          "src/vtx/llm/oauth/dynamic.py",
+          "src/vtx/llm/provider_catalog.py",
+          "src/vtx/sdk/run_config.py",
+          "src/vtx/llm/sdk/supercode.py",
+          "benchmark.py",
+          "src/vtx/sdk/agent.py",
+          "src/vtx/llm/dynamic_models.py",
+          "Site/src/content/docs/index.ts",
+          "src/vtx/runtime.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0202
+      },
+      {
+        "query": "user config file path resolution",
+        "description": "Config path resolution",
+        "expected_file": "config",
+        "latency_ms": 4.56,
+        "top1_file": "src/vtx/config.py",
+        "top5_files": [
+          "src/vtx/config.py",
+          "src/vtx/extensions.py",
+          "scripts/show_themes.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/ui/path_complete.py"
+        ],
+        "top10_files": [
+          "src/vtx/config.py",
+          "src/vtx/extensions.py",
+          "scripts/show_themes.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/ui/path_complete.py",
+          "src/vtx/tools/edit.py",
+          "benchmark.py",
+          "src/vtx/tools/skill.py",
+          "src/vtx/llm/provider_catalog.py",
+          "src/vtx/context/agent_mds.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0198
+      },
+      {
+        "query": "LLM API call streaming response",
+        "description": "LLM streaming",
+        "expected_file": "llm",
+        "latency_ms": 4.74,
+        "top1_file": "src/vtx/ui/blocks.py",
+        "top5_files": [
+          "src/vtx/ui/blocks.py",
+          "src/vtx/llm/sdk/supercode.py",
+          "src/vtx/turn.py",
+          "src/vtx/llm/phase_parser.py",
+          "src/vtx/llm/base.py"
+        ],
+        "top10_files": [
+          "src/vtx/ui/blocks.py",
+          "src/vtx/llm/sdk/supercode.py",
+          "src/vtx/turn.py",
+          "src/vtx/llm/phase_parser.py",
+          "src/vtx/llm/base.py",
+          "src/vtx/loop.py",
+          "benchmark.py",
+          "src/vtx/core/compaction.py",
+          "src/vtx/llm/providers/mock.py",
+          "src/vtx/llm/dynamic_models.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.016
+      },
+      {
+        "query": "model provider selection gemini openai",
+        "description": "Model provider",
+        "expected_file": "llm",
+        "latency_ms": 4.37,
+        "top1_file": "src/vtx/llm/providers/openai_sdk.py",
+        "top5_files": [
+          "src/vtx/llm/providers/openai_sdk.py",
+          "src/vtx/runtime.py",
+          "src/vtx/ui/commands/providers.py",
+          "src/vtx/llm/models.py",
+          "src/vtx/ui/commands/auth.py"
+        ],
+        "top10_files": [
+          "src/vtx/llm/providers/openai_sdk.py",
+          "src/vtx/runtime.py",
+          "src/vtx/ui/commands/providers.py",
+          "src/vtx/llm/models.py",
+          "src/vtx/ui/commands/auth.py",
+          "src/vtx/ui/selection_mode.py",
+          "src/vtx/llm/provider_catalog.py",
+          "src/vtx/sdk/agent.py",
+          "src/vtx/ui/completion_ui.py",
+          "src/vtx/llm/model_fetcher.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0192
+      },
+      {
+        "query": "token counting and context window management",
+        "description": "Token counting",
+        "expected_file": "llm",
+        "latency_ms": 4.62,
+        "top1_file": "src/vtx/core/compaction.py",
+        "top5_files": [
+          "src/vtx/core/compaction.py",
+          "src/vtx/ui/widgets.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/session.py"
+        ],
+        "top10_files": [
+          "src/vtx/core/compaction.py",
+          "src/vtx/ui/widgets.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/session.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/sdk/tracing/tracing_impl.py",
+          "src/vtx/llm/context_length.py",
+          "benchmark.py",
+          "src/vtx/context_governance.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0152
+      },
+      {
+        "query": "context window governance truncation",
+        "description": "Context governance",
+        "expected_file": "context_governance",
+        "latency_ms": 5.18,
+        "top1_file": "src/vtx/context_governance.py",
+        "top5_files": [
+          "src/vtx/context_governance.py",
+          "src/vtx/core/compaction.py",
+          "benchmark.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/context/agent_mds.py"
+        ],
+        "top10_files": [
+          "src/vtx/context_governance.py",
+          "src/vtx/core/compaction.py",
+          "benchmark.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/context/agent_mds.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/runtime.py",
+          "src/vtx/loop.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0182
+      },
+      {
+        "query": "context item priority ranking",
+        "description": "Context priority",
+        "expected_file": "context",
+        "latency_ms": 4.5,
+        "top1_file": "src/vtx/ui/input.py",
+        "top5_files": [
+          "src/vtx/ui/input.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/ui/tree.py",
+          "src/vtx/sdk/items.py",
+          "src/vtx/ui/completion_ui.py"
+        ],
+        "top10_files": [
+          "src/vtx/ui/input.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/ui/tree.py",
+          "src/vtx/sdk/items.py",
+          "src/vtx/ui/completion_ui.py",
+          "src/vtx/context/skills.py",
+          "src/vtx/context/git.py",
+          "src/vtx/sdk/items_base.py",
+          "src/vtx/llm/context_length.py",
+          "Site/src/index.css"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": true,
+        "top1_score": 0.0182
+      },
+      {
+        "query": "tool dispatcher routing function calls",
+        "description": "Tool dispatcher",
+        "expected_file": "dispatcher",
+        "latency_ms": 5.9,
+        "top1_file": "src/vtx/dispatcher.py",
+        "top5_files": [
+          "src/vtx/dispatcher.py",
+          "benchmark.py",
+          "src/vtx/llm/tool_parser.py",
+          "src/vtx/context_governance.py",
+          "src/vtx/llm/sdk/supercode.py"
+        ],
+        "top10_files": [
+          "src/vtx/dispatcher.py",
+          "benchmark.py",
+          "src/vtx/llm/tool_parser.py",
+          "src/vtx/context_governance.py",
+          "src/vtx/llm/sdk/supercode.py",
+          "src/vtx/llm/providers/supercode.py",
+          "src/vtx/prompts/tooling.py",
+          "src/vtx/tools/read.py",
+          "src/vtx/sdk/tools.py",
+          "src/vtx/turn.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0187
+      },
+      {
+        "query": "dispatch tool call with arguments",
+        "description": "Tool call dispatch",
+        "expected_file": "dispatcher",
+        "latency_ms": 6.33,
+        "top1_file": "src/vtx/turn.py",
+        "top5_files": [
+          "src/vtx/turn.py",
+          "benchmark.py",
+          "src/vtx/ui/session_ui.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/llm/tool_parser.py"
+        ],
+        "top10_files": [
+          "src/vtx/turn.py",
+          "benchmark.py",
+          "src/vtx/ui/session_ui.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/llm/tool_parser.py",
+          "src/vtx/ui/tree.py",
+          "src/vtx/prompts/tooling.py",
+          "src/vtx/tools/base.py",
+          "src/vtx/tools/find.py",
+          "src/vtx/tools/task.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0188
+      },
+      {
+        "query": "event system publish subscribe hooks",
+        "description": "Event system",
+        "expected_file": "events",
+        "latency_ms": 6.25,
+        "top1_file": "src/vtx/hooks/loader.py",
+        "top5_files": [
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/bridge.py",
+          "benchmark.py",
+          "src/vtx/hooks/agent_hook.py",
+          "src/vtx/hooks/registry.py"
+        ],
+        "top10_files": [
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/bridge.py",
+          "benchmark.py",
+          "src/vtx/hooks/agent_hook.py",
+          "src/vtx/hooks/registry.py",
+          "src/vtx/loop.py",
+          "src/vtx/extensions.py",
+          "src/vtx/hooks/runtime.py",
+          "Site/src/hooks/useRouteMeta.ts",
+          "src/vtx/hooks/types.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0211
+      },
+      {
+        "query": "streaming event output to terminal",
+        "description": "Event streaming",
+        "expected_file": "events",
+        "latency_ms": 4.75,
+        "top1_file": "src/vtx/turn.py",
+        "top5_files": [
+          "src/vtx/turn.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/ui/blocks.py",
+          "benchmark.py",
+          "Site/src/components/FeaturesPage.tsx"
+        ],
+        "top10_files": [
+          "src/vtx/turn.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/ui/blocks.py",
+          "benchmark.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/sdk/guardrails/types.py",
+          "src/vtx/tools/task.py",
+          "src/vtx/events.py",
+          "Site/src/components/TerminalBlock.tsx",
+          "src/vtx/ui/agent_runner.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": true,
+        "top1_score": 0.0192
+      },
+      {
+        "query": "git branch creation and switching",
+        "description": "Git branch",
+        "expected_file": "git_branch",
+        "latency_ms": 4.5,
+        "top1_file": "src/vtx/git_branch.py",
+        "top5_files": [
+          "src/vtx/git_branch.py",
+          "src/vtx/context/git.py",
+          "src/vtx/ui/widgets.py",
+          "benchmark.py",
+          "Site/src/components/Capabilities.tsx"
+        ],
+        "top10_files": [
+          "src/vtx/git_branch.py",
+          "src/vtx/context/git.py",
+          "src/vtx/ui/widgets.py",
+          "benchmark.py",
+          "Site/src/components/Capabilities.tsx",
+          "src/vtx/ui/startup.py",
+          "scripts/install.sh",
+          "src/vtx/ui/commands/settings.py",
+          "Site/src/content/blog/agentskill.ts",
+          "src/vtx/config.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0268
+      },
+      {
+        "query": "github CLI integration shell command",
+        "description": "GitHub CLI",
+        "expected_file": "gh_cli",
+        "latency_ms": 4.53,
+        "top1_file": "src/vtx/ui/agent_runner.py",
+        "top5_files": [
+          "src/vtx/ui/agent_runner.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/tools/bash.py",
+          "scripts/install.ps1",
+          "scripts/install.sh"
+        ],
+        "top10_files": [
+          "src/vtx/ui/agent_runner.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/tools/bash.py",
+          "scripts/install.ps1",
+          "scripts/install.sh",
+          "Site/src/hooks/useRouteMeta.ts",
+          "benchmark.py",
+          "src/vtx/ui/commands/auth.py",
+          "src/vtx/cli.py",
+          "tests/ui/test_shell_command_detection.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0174
+      },
+      {
+        "query": "diff display unified patch format",
+        "description": "Diff display",
+        "expected_file": "diff_display",
+        "latency_ms": 5.29,
+        "top1_file": "src/vtx/tools/edit.py",
+        "top5_files": [
+          "src/vtx/tools/edit.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/ui/blocks.py",
+          "benchmark.py",
+          "src/vtx/diff_display.py"
+        ],
+        "top10_files": [
+          "src/vtx/tools/edit.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/ui/blocks.py",
+          "benchmark.py",
+          "src/vtx/diff_display.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/ui/formatting.py",
+          "Site/src/index.css",
+          "src/vtx/tools/task.py",
+          "Site/src/components/MarkdownRenderer.tsx"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0166
+      },
+      {
+        "query": "show file changes colored diff output",
+        "description": "Colored diff",
+        "expected_file": "diff_display",
+        "latency_ms": 5.76,
+        "top1_file": "src/vtx/ui/widgets.py",
+        "top5_files": [
+          "src/vtx/ui/widgets.py",
+          "src/vtx/ui/launch.py",
+          "src/vtx/ui/blocks.py",
+          "src/vtx/tools/edit.py",
+          "src/vtx/ui/chat.py"
+        ],
+        "top10_files": [
+          "src/vtx/ui/widgets.py",
+          "src/vtx/ui/launch.py",
+          "src/vtx/ui/blocks.py",
+          "src/vtx/tools/edit.py",
+          "src/vtx/ui/chat.py",
+          "benchmark.py",
+          "src/vtx/tools/write.py",
+          "Site/src/index.css",
+          "src/vtx/ui/styles.py",
+          "src/vtx/ui/commands/settings.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0191
+      },
+      {
+        "query": "plugin extension registration loading",
+        "description": "Extension loading",
+        "expected_file": "extensions",
+        "latency_ms": 5.5,
+        "top1_file": "src/vtx/extensions.py",
+        "top5_files": [
+          "src/vtx/extensions.py",
+          "Site/src/content/blog/mcp-extensions.ts",
+          "benchmark.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/extensions.py"
+        ],
+        "top10_files": [
+          "src/vtx/extensions.py",
+          "Site/src/content/blog/mcp-extensions.ts",
+          "benchmark.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/extensions.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "Site/scripts/prerender.mjs",
+          "Site/src/content/blog/mcp-extensions.ts",
+          "src/vtx/cli.py",
+          "src/vtx/llm/providers/supercode.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0229
+      },
+      {
+        "query": "builtin skill definition and registration",
+        "description": "Builtin skills",
+        "expected_file": "builtin_skill",
+        "latency_ms": 4.26,
+        "top1_file": "src/vtx/context/skills.py",
+        "top5_files": [
+          "src/vtx/context/skills.py",
+          "src/vtx/tools/skill.py",
+          "src/vtx/sdk/skills.py",
+          "benchmark.py",
+          "Site/src/content/blog/agentskill.ts"
+        ],
+        "top10_files": [
+          "src/vtx/context/skills.py",
+          "src/vtx/tools/skill.py",
+          "src/vtx/sdk/skills.py",
+          "benchmark.py",
+          "Site/src/content/blog/agentskill.ts",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/context/loader.py",
+          "src/vtx/context/skills.py",
+          "src/vtx/extensions.py",
+          "src/vtx/tools/skill.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0209
+      },
+      {
+        "query": "skill slash command frontmatter parsing",
+        "description": "Skill frontmatter",
+        "expected_file": "builtin_skill",
+        "latency_ms": 4.37,
+        "top1_file": "src/vtx/context/skills.py",
+        "top5_files": [
+          "src/vtx/context/skills.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/autocomplete.py",
+          "src/vtx/ui/app.py",
+          "Site/src/content/blog/agentskill.ts"
+        ],
+        "top10_files": [
+          "src/vtx/context/skills.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/autocomplete.py",
+          "src/vtx/ui/app.py",
+          "Site/src/content/blog/agentskill.ts",
+          "Site/src/components/Capabilities.tsx",
+          "src/vtx/tools/skill.py",
+          "src/vtx/context/skills.py",
+          "src/vtx/sdk/skills.py",
+          "src/vtx/ui/commands/agents.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0206
+      },
+      {
+        "query": "subagent invocation and communication",
+        "description": "Subagent invocation",
+        "expected_file": "agents",
+        "latency_ms": 4.08,
+        "top1_file": "src/vtx/tools/task.py",
+        "top5_files": [
+          "src/vtx/tools/task.py",
+          "src/vtx/loop.py",
+          "src/vtx/config.py",
+          "src/vtx/ui/app.py",
+          "benchmark.py"
+        ],
+        "top10_files": [
+          "src/vtx/tools/task.py",
+          "src/vtx/loop.py",
+          "src/vtx/config.py",
+          "src/vtx/ui/app.py",
+          "benchmark.py",
+          "src/vtx/llm/providers/openai_sdk.py",
+          "src/vtx/tools/background.py",
+          "src/vtx/tools/ask_user.py",
+          "src/vtx/ui/chat.py",
+          "src/vtx/agents/api.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": true,
+        "top1_score": 0.0169
+      },
+      {
+        "query": "agent message passing protocol",
+        "description": "Agent messaging",
+        "expected_file": "agents",
+        "latency_ms": 4.67,
+        "top1_file": "src/vtx/agents/loader.py",
+        "top5_files": [
+          "src/vtx/agents/loader.py",
+          "src/vtx/ui/app_protocol.py",
+          "src/vtx/ui/commands/sessions.py",
+          "src/vtx/ui/commands/agents.py",
+          "src/vtx/agents/api.py"
+        ],
+        "top10_files": [
+          "src/vtx/agents/loader.py",
+          "src/vtx/ui/app_protocol.py",
+          "src/vtx/ui/commands/sessions.py",
+          "src/vtx/ui/commands/agents.py",
+          "src/vtx/agents/api.py",
+          "src/vtx/agent_runner.py",
+          "src/vtx/runtime.py",
+          "src/vtx/hooks/agent_hook.py",
+          "src/vtx/ui/agent_runner.py",
+          "benchmark.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0127
+      },
+      {
+        "query": "pre-commit hook integration",
+        "description": "Pre-commit hooks",
+        "expected_file": "hooks",
+        "latency_ms": 4.65,
+        "top1_file": "src/vtx/hooks/loader.py",
+        "top5_files": [
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/bridge.py",
+          "src/vtx/hooks/types.py",
+          "benchmark.py",
+          "src/vtx/hooks/agent_hook.py"
+        ],
+        "top10_files": [
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/bridge.py",
+          "src/vtx/hooks/types.py",
+          "benchmark.py",
+          "src/vtx/hooks/agent_hook.py",
+          "src/vtx/hooks/registry.py",
+          "Site/scripts/prerender.mjs",
+          "src/vtx/hooks/__init__.py",
+          "src/vtx/hooks/runtime.py",
+          "src/vtx/hooks/bridge.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0214
+      },
+      {
+        "query": "async utility helpers gather timeout",
+        "description": "Async utilities",
+        "expected_file": "async_utils",
+        "latency_ms": 4.27,
+        "top1_file": "src/vtx/turn.py",
+        "top5_files": [
+          "src/vtx/turn.py",
+          "src/vtx/hooks/bridge.py",
+          "benchmark.py",
+          "src/vtx/tools/web.py",
+          "src/vtx/update_check.py"
+        ],
+        "top10_files": [
+          "src/vtx/turn.py",
+          "src/vtx/hooks/bridge.py",
+          "benchmark.py",
+          "src/vtx/tools/web.py",
+          "src/vtx/update_check.py",
+          "src/vtx/tools/background.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/llm/oauth/copilot.py",
+          "src/vtx/llm/sdk/anthropic.py",
+          "src/vtx/llm/sdk/openai.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.018
+      },
+      {
+        "query": "headless non-interactive execution mode",
+        "description": "Headless mode",
+        "expected_file": "headless",
+        "latency_ms": 4.91,
+        "top1_file": "Site/src/components/FeaturesPage.tsx",
+        "top5_files": [
+          "Site/src/components/FeaturesPage.tsx",
+          "Site/src/hooks/useRouteMeta.ts",
+          "scripts/install.sh",
+          "src/vtx/headless.py",
+          "benchmark.py"
+        ],
+        "top10_files": [
+          "Site/src/components/FeaturesPage.tsx",
+          "Site/src/hooks/useRouteMeta.ts",
+          "scripts/install.sh",
+          "src/vtx/headless.py",
+          "benchmark.py",
+          "Site/src/components/Capabilities.tsx",
+          ".agents/skills/vtx-tmux-test/run-e2e-tests.sh",
+          "src/vtx/agents/registry.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/ui/input.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0195
+      },
+      {
+        "query": "error handling exception retry logic",
+        "description": "Generic: error handling (open)",
+        "expected_file": null,
+        "latency_ms": 4.43,
+        "top1_file": "src/vtx/turn.py",
+        "top5_files": [
+          "src/vtx/turn.py",
+          "src/vtx/core/errors.py",
+          "src/vtx/llm/rate_limit.py",
+          "src/vtx/tools/background.py",
+          "benchmark.py"
+        ],
+        "top10_files": [
+          "src/vtx/turn.py",
+          "src/vtx/core/errors.py",
+          "src/vtx/llm/rate_limit.py",
+          "src/vtx/tools/background.py",
+          "benchmark.py",
+          "src/vtx/llm/sdk/openai.py",
+          "src/vtx/llm/sdk/anthropic.py",
+          "src/vtx/llm/base.py",
+          "src/vtx/sdk/permissions.py",
+          "src/vtx/extensions.py"
+        ],
+        "hit_at_1": null,
+        "hit_at_5": null,
+        "hit_at_10": null,
+        "top1_score": 0.018
+      },
+      {
+        "query": "logging setup structured log output",
+        "description": "Generic: logging (open)",
+        "expected_file": null,
+        "latency_ms": 4.74,
+        "top1_file": "src/vtx/loop.py",
+        "top5_files": [
+          "src/vtx/loop.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/session_ui.py",
+          ".agents/skills/vtx-tmux-test/setup-test-project.sh"
+        ],
+        "top10_files": [
+          "src/vtx/loop.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/session_ui.py",
+          ".agents/skills/vtx-tmux-test/setup-test-project.sh",
+          "src/vtx/llm/sdk/openai.py",
+          "benchmark.py",
+          "src/vtx/agents/api.py",
+          "src/vtx/ui/export.py",
+          "src/vtx/hooks/loader.py"
+        ],
+        "hit_at_1": null,
+        "hit_at_5": null,
+        "hit_at_10": null,
+        "top1_score": 0.0166
+      },
+      {
+        "query": "test fixtures mock patching pytest",
+        "description": "Test utilities",
+        "expected_file": "tests",
+        "latency_ms": 4.19,
+        "top1_file": "benchmark.py",
+        "top5_files": [
+          "benchmark.py",
+          ".agents/skills/vtx-tmux-test/run-e2e-tests.sh",
+          "tests/llm/test_mock_provider.py",
+          "tests/ui/test_shell_command_detection.py",
+          "src/vtx/llm/dynamic_models.py"
+        ],
+        "top10_files": [
+          "benchmark.py",
+          ".agents/skills/vtx-tmux-test/run-e2e-tests.sh",
+          "tests/llm/test_mock_provider.py",
+          "tests/ui/test_shell_command_detection.py",
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/llm/sdk/openai.py",
+          "Site/src/index.css",
+          "tests/sdk/test_provider_field.py",
+          "tests/test_agentic_loop.py",
+          "tests/sdk/test_integration.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0141
+      }
+    ],
+    "recall_at_1": 0.3929,
+    "recall_at_5": 0.6071,
+    "recall_at_10": 0.7143,
+    "mean_latency_ms": 4.85,
+    "p50_latency_ms": 4.66,
+    "p95_latency_ms": 5.9,
+    "p99_latency_ms": 6.25,
+    "evaluated_queries": 28,
+    "open_queries": 2
+  },
+  {
+    "model_name": "VTXAI/vtx-embed-7M",
+    "model_label": "LF4 vtx-embed-7M (mini)",
+    "embedding_dim": 256,
+    "model_size_mb": 4.72448,
+    "index_dir": "/tmp/vtx_bench_LF4_vtx-embed-7M_(mini)_dt5m3366",
+    "indexed_files": 330,
+    "total_chunks": 2040,
+    "chunk_size": 1500,
+    "chunk_overlap": 200,
+    "languages": {
+      "python": 1839,
+      "typescript": 27,
+      "tsx": 129,
+      "javascript": 10,
+      "bash": 16,
+      "css": 13,
+      "powershell": 5
+    },
+    "index_time_ms": 3056.6,
+    "peak_ram_mb": 0.0,
+    "memo_hits": 0,
+    "memo_misses": 2040,
+    "query_results": [
+      {
+        "query": "agent runner main loop execution",
+        "description": "Agent runner entry point",
+        "expected_file": "agent_runner",
+        "latency_ms": 6.35,
+        "top1_file": "Site/src/content/blog/agentic-loop.ts",
+        "top5_files": [
+          "Site/src/content/blog/agentic-loop.ts",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/loop.py",
+          "src/vtx/agent_runner.py",
+          "src/vtx/runtime.py"
+        ],
+        "top10_files": [
+          "Site/src/content/blog/agentic-loop.ts",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/loop.py",
+          "src/vtx/agent_runner.py",
+          "src/vtx/runtime.py",
+          "src/vtx/ui/agent_runner.py",
+          "src/vtx/sdk/handoffs.py",
+          "src/vtx/agents/loader.py",
+          "src/vtx/ui/commands/agents.py",
+          "src/vtx/sdk/run_config.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0228
+      },
+      {
+        "query": "how does the agent run tasks asynchronously",
+        "description": "Async task execution",
+        "expected_file": "agent_runner",
+        "latency_ms": 6.0,
+        "top1_file": "src/vtx/runtime.py",
+        "top5_files": [
+          "src/vtx/runtime.py",
+          "Site/src/content/blog/agentic-loop.ts",
+          "src/vtx/agents/loader.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/tools/task.py"
+        ],
+        "top10_files": [
+          "src/vtx/runtime.py",
+          "Site/src/content/blog/agentic-loop.ts",
+          "src/vtx/agents/loader.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/tools/task.py",
+          "src/vtx/tools/background.py",
+          "src/vtx/sdk/run_config.py",
+          "Site/src/content/blog/agentskill.ts",
+          "src/vtx/ui/commands/agents.py",
+          "src/vtx/sdk/results.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.024
+      },
+      {
+        "query": "command line interface argument parsing",
+        "description": "CLI argument parsing",
+        "expected_file": "cli",
+        "latency_ms": 5.96,
+        "top1_file": "src/vtx/cli.py",
+        "top5_files": [
+          "src/vtx/cli.py",
+          "benchmark.py",
+          "scripts/show_themes.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/extensions.py"
+        ],
+        "top10_files": [
+          "src/vtx/cli.py",
+          "benchmark.py",
+          "scripts/show_themes.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/extensions.py",
+          "src/vtx/ui/agent_runner.py",
+          "src/vtx/agents/api.py",
+          "src/vtx/tools/edit.py",
+          "src/vtx/ui/input.py",
+          "src/vtx/tools/find.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0195
+      },
+      {
+        "query": "vtx CLI entry point argparse",
+        "description": "CLI argparse setup",
+        "expected_file": "cli",
+        "latency_ms": 6.15,
+        "top1_file": "src/vtx/cli.py",
+        "top5_files": [
+          "src/vtx/cli.py",
+          "src/vtx/extensions.py",
+          "scripts/show_themes.py",
+          "benchmark.py",
+          "src/vtx/gh_cli.py"
+        ],
+        "top10_files": [
+          "src/vtx/cli.py",
+          "src/vtx/extensions.py",
+          "scripts/show_themes.py",
+          "benchmark.py",
+          "src/vtx/gh_cli.py",
+          "src/vtx/prompts/builder.py",
+          "src/vtx/ui/launch.py",
+          "src/vtx/session.py",
+          "src/vtx/cli.py",
+          "src/vtx/config.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0269
+      },
+      {
+        "query": "configuration loading and defaults",
+        "description": "Config loading",
+        "expected_file": "config",
+        "latency_ms": 6.13,
+        "top1_file": "src/vtx/config.py",
+        "top5_files": [
+          "src/vtx/config.py",
+          "src/vtx/hooks/loader.py",
+          "src/vtx/llm/oauth/dynamic.py",
+          "src/vtx/sdk/run_config.py",
+          "src/vtx/llm/provider_catalog.py"
+        ],
+        "top10_files": [
+          "src/vtx/config.py",
+          "src/vtx/hooks/loader.py",
+          "src/vtx/llm/oauth/dynamic.py",
+          "src/vtx/sdk/run_config.py",
+          "src/vtx/llm/provider_catalog.py",
+          "src/vtx/extensions.py",
+          "src/vtx/config.py",
+          "benchmark.py",
+          "Site/src/content/docs/index.ts",
+          "src/vtx/dispatcher.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0209
+      },
+      {
+        "query": "user config file path resolution",
+        "description": "Config path resolution",
+        "expected_file": "config",
+        "latency_ms": 6.22,
+        "top1_file": "src/vtx/config.py",
+        "top5_files": [
+          "src/vtx/config.py",
+          "src/vtx/extensions.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/llm/provider_catalog.py",
+          "scripts/show_themes.py"
+        ],
+        "top10_files": [
+          "src/vtx/config.py",
+          "src/vtx/extensions.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/llm/provider_catalog.py",
+          "scripts/show_themes.py",
+          "src/vtx/tools/edit.py",
+          "src/vtx/context/agent_mds.py",
+          "src/vtx/tools_manager.py",
+          "src/vtx/ui/path_complete.py",
+          "Site/vite.config.ts"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0201
+      },
+      {
+        "query": "LLM API call streaming response",
+        "description": "LLM streaming",
+        "expected_file": "llm",
+        "latency_ms": 5.86,
+        "top1_file": "src/vtx/llm/sdk/supercode.py",
+        "top5_files": [
+          "src/vtx/llm/sdk/supercode.py",
+          "src/vtx/turn.py",
+          "src/vtx/ui/blocks.py",
+          "src/vtx/llm/providers/mock.py",
+          "src/vtx/llm/dynamic_models.py"
+        ],
+        "top10_files": [
+          "src/vtx/llm/sdk/supercode.py",
+          "src/vtx/turn.py",
+          "src/vtx/ui/blocks.py",
+          "src/vtx/llm/providers/mock.py",
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/llm/phase_parser.py",
+          "src/vtx/llm/sdk/base.py",
+          "src/vtx/llm/base.py",
+          "src/vtx/llm/rate_limit.py",
+          "benchmark.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.018
+      },
+      {
+        "query": "model provider selection gemini openai",
+        "description": "Model provider",
+        "expected_file": "llm",
+        "latency_ms": 5.62,
+        "top1_file": "src/vtx/llm/providers/openai_sdk.py",
+        "top5_files": [
+          "src/vtx/llm/providers/openai_sdk.py",
+          "src/vtx/ui/commands/providers.py",
+          "src/vtx/runtime.py",
+          "src/vtx/llm/sdk/openai.py",
+          "src/vtx/ui/commands/models.py"
+        ],
+        "top10_files": [
+          "src/vtx/llm/providers/openai_sdk.py",
+          "src/vtx/ui/commands/providers.py",
+          "src/vtx/runtime.py",
+          "src/vtx/llm/sdk/openai.py",
+          "src/vtx/ui/commands/models.py",
+          "src/vtx/llm/models.py",
+          "src/vtx/ui/commands/auth.py",
+          "src/vtx/ui/selection_mode.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/app.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0182
+      },
+      {
+        "query": "token counting and context window management",
+        "description": "Token counting",
+        "expected_file": "llm",
+        "latency_ms": 5.65,
+        "top1_file": "src/vtx/llm/dynamic_models.py",
+        "top5_files": [
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/core/compaction.py",
+          "src/vtx/session.py",
+          "Site/src/content/blog/memory-compaction.ts"
+        ],
+        "top10_files": [
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/core/compaction.py",
+          "src/vtx/session.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "src/vtx/loop.py",
+          "src/vtx/tools/background.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/llm/model_fetcher.py",
+          "src/vtx/sdk/tracing/tracing_impl.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0175
+      },
+      {
+        "query": "context window governance truncation",
+        "description": "Context governance",
+        "expected_file": "context_governance",
+        "latency_ms": 6.55,
+        "top1_file": "src/vtx/llm/dynamic_models.py",
+        "top5_files": [
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/context_governance.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/llm/model_fetcher.py"
+        ],
+        "top10_files": [
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/context_governance.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/llm/model_fetcher.py",
+          "src/vtx/core/compaction.py",
+          "src/vtx/llm/context_length.py",
+          "src/vtx/context/skills.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "src/vtx/context/git.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0189
+      },
+      {
+        "query": "context item priority ranking",
+        "description": "Context priority",
+        "expected_file": "context",
+        "latency_ms": 7.03,
+        "top1_file": "src/vtx/ui/commands/settings.py",
+        "top5_files": [
+          "src/vtx/ui/commands/settings.py",
+          "src/vtx/ui/input.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/ui/tree.py",
+          "src/vtx/ui/completion_ui.py"
+        ],
+        "top10_files": [
+          "src/vtx/ui/commands/settings.py",
+          "src/vtx/ui/input.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/ui/tree.py",
+          "src/vtx/ui/completion_ui.py",
+          "src/vtx/llm/dynamic_models.py",
+          "src/vtx/sdk/items.py",
+          "src/vtx/llm/oauth/dynamic.py",
+          "src/vtx/context/skills.py",
+          "src/vtx/ui/floating_list.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": true,
+        "top1_score": 0.0186
+      },
+      {
+        "query": "tool dispatcher routing function calls",
+        "description": "Tool dispatcher",
+        "expected_file": "dispatcher",
+        "latency_ms": 6.54,
+        "top1_file": "src/vtx/dispatcher.py",
+        "top5_files": [
+          "src/vtx/dispatcher.py",
+          "src/vtx/runtime.py",
+          "scripts/supercode_proxy.py",
+          "src/vtx/llm/sdk/supercode.py",
+          "src/vtx/llm/providers/supercode.py"
+        ],
+        "top10_files": [
+          "src/vtx/dispatcher.py",
+          "src/vtx/runtime.py",
+          "scripts/supercode_proxy.py",
+          "src/vtx/llm/sdk/supercode.py",
+          "src/vtx/llm/providers/supercode.py",
+          "src/vtx/llm/tool_parser.py",
+          "src/vtx/turn.py",
+          "src/vtx/llm/base.py",
+          "src/vtx/context_governance.py",
+          "src/vtx/ui/export.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0222
+      },
+      {
+        "query": "dispatch tool call with arguments",
+        "description": "Tool call dispatch",
+        "expected_file": "dispatcher",
+        "latency_ms": 7.46,
+        "top1_file": "src/vtx/turn.py",
+        "top5_files": [
+          "src/vtx/turn.py",
+          "src/vtx/ui/tree.py",
+          "src/vtx/ui/session_ui.py",
+          "src/vtx/dispatcher.py",
+          "src/vtx/llm/base.py"
+        ],
+        "top10_files": [
+          "src/vtx/turn.py",
+          "src/vtx/ui/tree.py",
+          "src/vtx/ui/session_ui.py",
+          "src/vtx/dispatcher.py",
+          "src/vtx/llm/base.py",
+          "src/vtx/llm/tool_parser.py",
+          "src/vtx/tools/base.py",
+          "src/vtx/llm/sdk/supercode.py",
+          "src/vtx/tools/task.py",
+          "src/vtx/llm/providers/mock.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0184
+      },
+      {
+        "query": "event system publish subscribe hooks",
+        "description": "Event system",
+        "expected_file": "events",
+        "latency_ms": 6.33,
+        "top1_file": "src/vtx/hooks/loader.py",
+        "top5_files": [
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/bridge.py",
+          "src/vtx/hooks/registry.py",
+          "src/vtx/hooks/runtime.py",
+          "src/vtx/hooks/agent_hook.py"
+        ],
+        "top10_files": [
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/bridge.py",
+          "src/vtx/hooks/registry.py",
+          "src/vtx/hooks/runtime.py",
+          "src/vtx/hooks/agent_hook.py",
+          "src/vtx/agent_runner.py",
+          "src/vtx/turn.py",
+          "src/vtx/loop.py",
+          "Site/src/content/blog/mcp-extensions.ts",
+          "src/vtx/extensions.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0213
+      },
+      {
+        "query": "streaming event output to terminal",
+        "description": "Event streaming",
+        "expected_file": "events",
+        "latency_ms": 6.67,
+        "top1_file": "src/vtx/turn.py",
+        "top5_files": [
+          "src/vtx/turn.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/ui/blocks.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "Site/src/components/Hero.tsx"
+        ],
+        "top10_files": [
+          "src/vtx/turn.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/ui/blocks.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "Site/src/components/Hero.tsx",
+          "src/vtx/tools/task.py",
+          "src/vtx/events.py",
+          "Site/src/components/TerminalBlock.tsx",
+          "src/vtx/extensions.py",
+          "benchmark.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": true,
+        "top1_score": 0.0185
+      },
+      {
+        "query": "git branch creation and switching",
+        "description": "Git branch",
+        "expected_file": "git_branch",
+        "latency_ms": 6.88,
+        "top1_file": "src/vtx/git_branch.py",
+        "top5_files": [
+          "src/vtx/git_branch.py",
+          "src/vtx/context/git.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/ui/startup.py",
+          "scripts/install.sh"
+        ],
+        "top10_files": [
+          "src/vtx/git_branch.py",
+          "src/vtx/context/git.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/ui/startup.py",
+          "scripts/install.sh",
+          "scripts/install.ps1",
+          "src/vtx/prompts/builder.py",
+          "src/vtx/ui/commands/settings.py",
+          "src/vtx/git_branch.py",
+          "src/vtx/prompts/identity.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0286
+      },
+      {
+        "query": "github CLI integration shell command",
+        "description": "GitHub CLI",
+        "expected_file": "gh_cli",
+        "latency_ms": 6.06,
+        "top1_file": "src/vtx/tools/bash.py",
+        "top5_files": [
+          "src/vtx/tools/bash.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/agent_runner.py",
+          "scripts/install.sh",
+          "scripts/install.ps1"
+        ],
+        "top10_files": [
+          "src/vtx/tools/bash.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/agent_runner.py",
+          "scripts/install.sh",
+          "scripts/install.ps1",
+          "src/vtx/ui/input.py",
+          "src/vtx/llm/oauth/copilot.py",
+          "src/vtx/ui/formatting.py",
+          "src/vtx/context/git.py",
+          "src/vtx/ui/commands/agents.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0165
+      },
+      {
+        "query": "diff display unified patch format",
+        "description": "Diff display",
+        "expected_file": "diff_display",
+        "latency_ms": 6.21,
+        "top1_file": "src/vtx/tools/edit.py",
+        "top5_files": [
+          "src/vtx/tools/edit.py",
+          "src/vtx/ui/blocks.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/diff_display.py",
+          "src/vtx/tools/bash.py"
+        ],
+        "top10_files": [
+          "src/vtx/tools/edit.py",
+          "src/vtx/ui/blocks.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/diff_display.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/core/types.py",
+          "src/vtx/tools/skill.py",
+          "benchmark.py",
+          "src/vtx/ui/widgets.py",
+          "src/vtx/themes.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0191
+      },
+      {
+        "query": "show file changes colored diff output",
+        "description": "Colored diff",
+        "expected_file": "diff_display",
+        "latency_ms": 5.67,
+        "top1_file": "src/vtx/ui/widgets.py",
+        "top5_files": [
+          "src/vtx/ui/widgets.py",
+          "src/vtx/tools/edit.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/ui/launch.py",
+          "src/vtx/ui/blocks.py"
+        ],
+        "top10_files": [
+          "src/vtx/ui/widgets.py",
+          "src/vtx/tools/edit.py",
+          "src/vtx/tools/write.py",
+          "src/vtx/ui/launch.py",
+          "src/vtx/ui/blocks.py",
+          "src/vtx/ui/commands/settings.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/ui/selection_mode.py",
+          "src/vtx/config.py",
+          "src/vtx/ui/chat.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0192
+      },
+      {
+        "query": "plugin extension registration loading",
+        "description": "Extension loading",
+        "expected_file": "extensions",
+        "latency_ms": 6.21,
+        "top1_file": "src/vtx/extensions.py",
+        "top5_files": [
+          "src/vtx/extensions.py",
+          "Site/src/content/blog/mcp-extensions.ts",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/hooks/registry.py",
+          "src/vtx/ui/app.py"
+        ],
+        "top10_files": [
+          "src/vtx/extensions.py",
+          "Site/src/content/blog/mcp-extensions.ts",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/hooks/registry.py",
+          "src/vtx/ui/app.py",
+          "src/vtx/hooks/loader.py",
+          "src/vtx/extensions.py",
+          "benchmark.py",
+          "Site/scripts/prerender.mjs",
+          "src/vtx/hooks/types.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0231
+      },
+      {
+        "query": "builtin skill definition and registration",
+        "description": "Builtin skills",
+        "expected_file": "builtin_skill",
+        "latency_ms": 7.28,
+        "top1_file": "src/vtx/context/skills.py",
+        "top5_files": [
+          "src/vtx/context/skills.py",
+          "src/vtx/tools/skill.py",
+          "src/vtx/sdk/skills.py",
+          "src/vtx/context/loader.py",
+          "Site/src/content/blog/agentskill.ts"
+        ],
+        "top10_files": [
+          "src/vtx/context/skills.py",
+          "src/vtx/tools/skill.py",
+          "src/vtx/sdk/skills.py",
+          "src/vtx/context/loader.py",
+          "Site/src/content/blog/agentskill.ts",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/app.py",
+          "src/vtx/agents/registry.py",
+          "src/vtx/context/skills.py",
+          "src/vtx/tools/skill.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0226
+      },
+      {
+        "query": "skill slash command frontmatter parsing",
+        "description": "Skill frontmatter",
+        "expected_file": "builtin_skill",
+        "latency_ms": 5.74,
+        "top1_file": "src/vtx/context/skills.py",
+        "top5_files": [
+          "src/vtx/context/skills.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/app.py",
+          "Site/src/content/blog/agentskill.ts",
+          "src/vtx/ui/autocomplete.py"
+        ],
+        "top10_files": [
+          "src/vtx/context/skills.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/ui/app.py",
+          "Site/src/content/blog/agentskill.ts",
+          "src/vtx/ui/autocomplete.py",
+          "src/vtx/tools/skill.py",
+          "src/vtx/ui/input.py",
+          "src/vtx/sdk/skills.py",
+          "src/vtx/ui/commands/agents.py",
+          "src/vtx/context/skills.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0212
+      },
+      {
+        "query": "subagent invocation and communication",
+        "description": "Subagent invocation",
+        "expected_file": "agents",
+        "latency_ms": 6.02,
+        "top1_file": "src/vtx/tools/task.py",
+        "top5_files": [
+          "src/vtx/tools/task.py",
+          "src/vtx/config.py",
+          "src/vtx/dispatcher.py",
+          "benchmark.py",
+          "src/vtx/tools/background.py"
+        ],
+        "top10_files": [
+          "src/vtx/tools/task.py",
+          "src/vtx/config.py",
+          "src/vtx/dispatcher.py",
+          "benchmark.py",
+          "src/vtx/tools/background.py",
+          "src/vtx/extensions.py",
+          "src/vtx/tools/ask_user.py",
+          "src/vtx/ui/chat.py",
+          "src/vtx/loop.py",
+          "src/vtx/llm/sdk/supercode.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.017
+      },
+      {
+        "query": "agent message passing protocol",
+        "description": "Agent messaging",
+        "expected_file": "agents",
+        "latency_ms": 7.92,
+        "top1_file": "src/vtx/ui/commands/agents.py",
+        "top5_files": [
+          "src/vtx/ui/commands/agents.py",
+          "src/vtx/agents/loader.py",
+          "src/vtx/ui/app_protocol.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/agents/api.py"
+        ],
+        "top10_files": [
+          "src/vtx/ui/commands/agents.py",
+          "src/vtx/agents/loader.py",
+          "src/vtx/ui/app_protocol.py",
+          "src/vtx/sdk/runner.py",
+          "src/vtx/agents/api.py",
+          "src/vtx/agent_runner.py",
+          "src/vtx/runtime.py",
+          "src/vtx/hooks/agent_hook.py",
+          "src/vtx/agents/registry.py",
+          "src/vtx/ui/agent_runner.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0198
+      },
+      {
+        "query": "pre-commit hook integration",
+        "description": "Pre-commit hooks",
+        "expected_file": "hooks",
+        "latency_ms": 7.1,
+        "top1_file": "src/vtx/hooks/bridge.py",
+        "top5_files": [
+          "src/vtx/hooks/bridge.py",
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/types.py",
+          "src/vtx/hooks/agent_hook.py",
+          "src/vtx/hooks/registry.py"
+        ],
+        "top10_files": [
+          "src/vtx/hooks/bridge.py",
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/types.py",
+          "src/vtx/hooks/agent_hook.py",
+          "src/vtx/hooks/registry.py",
+          "Site/scripts/prerender.mjs",
+          "src/vtx/hooks/runtime.py",
+          "src/vtx/hooks/__init__.py",
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/bridge.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0224
+      },
+      {
+        "query": "async utility helpers gather timeout",
+        "description": "Async utilities",
+        "expected_file": "async_utils",
+        "latency_ms": 11.93,
+        "top1_file": "src/vtx/turn.py",
+        "top5_files": [
+          "src/vtx/turn.py",
+          "src/vtx/hooks/bridge.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/llm/sdk/anthropic.py",
+          "src/vtx/tools/background.py"
+        ],
+        "top10_files": [
+          "src/vtx/turn.py",
+          "src/vtx/hooks/bridge.py",
+          "src/vtx/tools/bash.py",
+          "src/vtx/llm/sdk/anthropic.py",
+          "src/vtx/tools/background.py",
+          "src/vtx/tools/web.py",
+          "src/vtx/tools/_tool_utils.py",
+          "src/vtx/llm/oauth/copilot.py",
+          "src/vtx/llm/providers/mock.py",
+          "benchmark.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": false,
+        "hit_at_10": false,
+        "top1_score": 0.0181
+      },
+      {
+        "query": "headless non-interactive execution mode",
+        "description": "Headless mode",
+        "expected_file": "headless",
+        "latency_ms": 15.42,
+        "top1_file": "src/vtx/headless.py",
+        "top5_files": [
+          "src/vtx/headless.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "Site/src/hooks/useRouteMeta.ts",
+          "src/vtx/prompts/identity.py",
+          "Site/src/components/Capabilities.tsx"
+        ],
+        "top10_files": [
+          "src/vtx/headless.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "Site/src/hooks/useRouteMeta.ts",
+          "src/vtx/prompts/identity.py",
+          "Site/src/components/Capabilities.tsx",
+          "src/vtx/cli.py",
+          "src/vtx/ui/commands/settings.py",
+          "src/vtx/ui/app.py",
+          "benchmark.py",
+          "src/vtx/runtime.py"
+        ],
+        "hit_at_1": true,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0198
+      },
+      {
+        "query": "error handling exception retry logic",
+        "description": "Generic: error handling (open)",
+        "expected_file": null,
+        "latency_ms": 14.02,
+        "top1_file": "src/vtx/llm/rate_limit.py",
+        "top5_files": [
+          "src/vtx/llm/rate_limit.py",
+          "src/vtx/core/errors.py",
+          "src/vtx/turn.py",
+          "src/vtx/llm/sdk/openai.py",
+          "src/vtx/llm/sdk/anthropic.py"
+        ],
+        "top10_files": [
+          "src/vtx/llm/rate_limit.py",
+          "src/vtx/core/errors.py",
+          "src/vtx/turn.py",
+          "src/vtx/llm/sdk/openai.py",
+          "src/vtx/llm/sdk/anthropic.py",
+          "src/vtx/llm/providers/openai_sdk.py",
+          "src/vtx/llm/providers/anthropic_sdk.py",
+          "src/vtx/events.py",
+          "src/vtx/llm/base.py",
+          "src/vtx/tools/background.py"
+        ],
+        "hit_at_1": null,
+        "hit_at_5": null,
+        "hit_at_10": null,
+        "top1_score": 0.0177
+      },
+      {
+        "query": "logging setup structured log output",
+        "description": "Generic: logging (open)",
+        "expected_file": null,
+        "latency_ms": 10.64,
+        "top1_file": "src/vtx/loop.py",
+        "top5_files": [
+          "src/vtx/loop.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/registry.py",
+          "Site/src/components/FeaturesPage.tsx"
+        ],
+        "top10_files": [
+          "src/vtx/loop.py",
+          "Site/src/content/blog/memory-compaction.ts",
+          "src/vtx/hooks/loader.py",
+          "src/vtx/hooks/registry.py",
+          "Site/src/components/FeaturesPage.tsx",
+          "src/vtx/llm/sdk/openai.py",
+          ".agents/skills/vtx-tmux-test/setup-test-project.sh",
+          "src/vtx/ui/session_ui.py",
+          "src/vtx/ui/app.py",
+          "scripts/install.sh"
+        ],
+        "hit_at_1": null,
+        "hit_at_5": null,
+        "hit_at_10": null,
+        "top1_score": 0.0179
+      },
+      {
+        "query": "test fixtures mock patching pytest",
+        "description": "Test utilities",
+        "expected_file": "tests",
+        "latency_ms": 6.85,
+        "top1_file": "benchmark.py",
+        "top5_files": [
+          "benchmark.py",
+          "tests/llm/test_mock_provider.py",
+          "tests/ui/test_shell_command_detection.py",
+          "tests/test_rate_limit.py",
+          "tests/tools/test_web_expand_collapse.py"
+        ],
+        "top10_files": [
+          "benchmark.py",
+          "tests/llm/test_mock_provider.py",
+          "tests/ui/test_shell_command_detection.py",
+          "tests/test_rate_limit.py",
+          "tests/tools/test_web_expand_collapse.py",
+          "tests/tools/test_subprocess_cancellation.py",
+          "tests/sdk/test_provider_field.py",
+          "tests/sdk/conftest.py",
+          "tests/tools/test_read.py",
+          "tests/test_grep_tool.py"
+        ],
+        "hit_at_1": false,
+        "hit_at_5": true,
+        "hit_at_10": true,
+        "top1_score": 0.0085
+      }
+    ],
+    "recall_at_1": 0.4643,
+    "recall_at_5": 0.6429,
+    "recall_at_10": 0.7143,
+    "mean_latency_ms": 7.28,
+    "p50_latency_ms": 6.34,
+    "p95_latency_ms": 11.93,
+    "p99_latency_ms": 14.02,
+    "evaluated_queries": 28,
+    "open_queries": 2
+  }
+]