acdream/docs/superpowers/plans/2026-07-05-mp0-frame-profiler.md

# MP0 — Honest Frame Profiler + Baseline Capture Implementation Plan

> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.

**Goal:** Build the permanent frame profiler (CPU frame time, GPU frame time, per-stage CPU attribution, per-frame allocation) and capture the MP baseline report that gates the rest of the Modern Pipeline track.

**Architecture:** A pure, unit-testable stats core (`FrameStatsBuffer`) + a GL query wrapper (`GpuFrameTimer`) + a facade (`FrameProfiler`) wired into `GameWindow` at exactly one frame-boundary call plus three stage scopes. Toggle lives in `RenderingDiagnostics` (the established diagnostic-owner pattern), mirrored in the DebugPanel. Spec: `docs/superpowers/specs/2026-07-05-modern-pipeline-design.md` §5.

**Tech Stack:** .NET 10, Silk.NET OpenGL (`QueryTarget.TimeElapsed` ring, mirroring `WbDrawDispatcher`'s existing idiom), xunit in `tests/AcDream.App.Tests`.

**Spec deviations (recorded):**
1. Spec §5 lists per-stage GPU attribution via `glQueryCounter` (GL_TIMESTAMP) markers. Deferred: the measured GPU total is ~0.5 ms (the frame is CPU-bound), so MP0 ships per-stage **CPU** attribution + whole-frame GPU only. GPU timestamps get added in the MP phase whose gate first needs GPU-side attribution (likely MP3).
2. The spec says "Toggles via `RuntimeOptions` + DebugPanel". `RuntimeOptions`' own doc comment says runtime diagnostic toggles belong in diagnostic owner classes, NOT RuntimeOptions ([RuntimeOptions.cs:13-19](../../src/AcDream.App/RuntimeOptions.cs)). We follow the codebase rule: the flag goes in `RenderingDiagnostics` (env-seeded, runtime-toggleable). No RuntimeOptions change.

**Hard GL constraint (discovered during planning):** GL forbids two simultaneously active `TimeElapsed` queries. `WbDrawDispatcher` already brackets its opaque/transparent passes with `TimeElapsed` queries when `ACDREAM_WB_DIAG=1` ([WbDrawDispatcher.cs:1642](../../src/AcDream.App/Rendering/Wb/WbDrawDispatcher.cs)). Therefore the profiler's whole-frame GPU query is **disabled (with a one-time console notice) when `ACDREAM_WB_DIAG=1`**. This is also the 2026-06-23 measurement lesson (separate flags for whole-frame vs per-pass) enforced in code.

**Threading assumption:** Silk.NET drives `OnUpdate` and `OnRender` on the same loop thread in this app. `FrameProfiler` captures the first caller's managed thread id and emits a one-time warning if any later call arrives on a different thread (guards the `GC.GetAllocatedBytesForCurrentThread` validity).

---

## File map

| File | Action | Responsibility |
|---|---|---|
| `src/AcDream.App/Diagnostics/FrameStatsBuffer.cs` | Create | Pure ring-buffer stats: samples, percentiles, stage accumulators, report line. Zero GL, zero statics. |
| `src/AcDream.App/Diagnostics/GpuFrameTimer.cs` | Create | GL `TimeElapsed` query ring (depth 4, `Begun` flags per the #125 lesson). |
| `src/AcDream.App/Diagnostics/FrameProfiler.cs` | Create | Facade: frame boundary, stage scopes, GC counters, 5-second `[frame-prof]` report. |
| `src/AcDream.Core/Rendering/RenderingDiagnostics.cs` | Modify | Add `FrameProfEnabled` flag (env `ACDREAM_FRAME_PROF`). |
| `src/AcDream.App/Rendering/GameWindow.cs` | Modify | One field + one call at `OnRender` top (~:8757), stage scope around `_wbMeshAdapter?.Tick()` (~:8813), around `_imguiBootstrap.Render()` (~:9821), and at `OnUpdate` top (~:7952). Wiring only — no logic in GameWindow. |
| `src/AcDream.UI.Abstractions/Panels/Debug/DebugVM.cs` | Modify | `FrameProf` mirror property (copy the `ProbeResolve` pattern at :248). |
| `src/AcDream.UI.ImGui/` (renderer for DebugPanel) | Modify | Checkbox for the mirror (locate via `grep -rn "ProbeResolve" src/AcDream.UI.ImGui/`). |
| `tests/AcDream.App.Tests/FrameStatsBufferTests.cs` | Create | Percentiles, ring wrap, stage reset, report format. |
| `docs/research/2026-07-XX-mp0-baseline.md` | Create (Task 5) | The baseline report — MP0's deliverable. |

Line numbers drift; match by the quoted anchor code, not the number.

---

### Task 1: FrameStatsBuffer (pure stats core, TDD)

**Files:**
- Create: `src/AcDream.App/Diagnostics/FrameStatsBuffer.cs`
- Test: `tests/AcDream.App.Tests/FrameStatsBufferTests.cs`

- [ ] **Step 1: Write the failing tests**

```csharp
using AcDream.App.Diagnostics;
using Xunit;

namespace AcDream.App.Tests;

public class FrameStatsBufferTests
{
    [Fact]
    public void Percentiles_OnKnownDistribution_AreExact()
    {
        var buf = new FrameStatsBuffer(capacity: 100);
        // 1..100 µs — p50 = 50, p95 = 95, p99 = 99, max = 100.
        for (long i = 1; i <= 100; i++) buf.Push(i);

        Assert.Equal(50, buf.Percentile(0.50));
        Assert.Equal(95, buf.Percentile(0.95));
        Assert.Equal(99, buf.Percentile(0.99));
        Assert.Equal(100, buf.Max());
    }

    [Fact]
    public void Push_PastCapacity_KeepsOnlyNewestWindow()
    {
        var buf = new FrameStatsBuffer(capacity: 4);
        foreach (long v in new long[] { 1000, 1000, 1000, 1000, 1, 2, 3, 4 })
            buf.Push(v);
        // The four 1000s were overwritten; window is {1,2,3,4}.
        Assert.Equal(4, buf.Count);
        Assert.Equal(4, buf.Max());
        Assert.Equal(2, buf.Percentile(0.50));
    }

    [Fact]
    public void Percentile_Empty_ReturnsZero()
    {
        var buf = new FrameStatsBuffer(capacity: 8);
        Assert.Equal(0, buf.Percentile(0.95));
        Assert.Equal(0, buf.Max());
        Assert.Equal(0, buf.Count);
    }

    [Fact]
    public void Reset_ClearsWindow()
    {
        var buf = new FrameStatsBuffer(capacity: 8);
        buf.Push(5); buf.Push(7);
        buf.Reset();
        Assert.Equal(0, buf.Count);
        Assert.Equal(0, buf.Percentile(0.5));
    }
}
```

- [ ] **Step 2: Run tests to verify they fail**

Run: `dotnet test tests/AcDream.App.Tests --filter FrameStatsBufferTests -v minimal`
Expected: FAIL — `FrameStatsBuffer` does not exist (compile error).

- [ ] **Step 3: Implement FrameStatsBuffer**

```csharp
using System;

namespace AcDream.App.Diagnostics;

/// <summary>
/// MP0 (2026-07-05) — fixed-capacity ring buffer of long samples
/// (microseconds or bytes) with percentile/max over the current window.
/// Pure and allocation-free after construction: <see cref="Percentile"/>
/// sorts into a preallocated scratch array, so the 5-second report path
/// allocates nothing. Not thread-safe — owned by the window loop thread.
/// Spec: docs/superpowers/specs/2026-07-05-modern-pipeline-design.md §5.
/// </summary>
public sealed class FrameStatsBuffer
{
    private readonly long[] _samples;
    private readonly long[] _scratch;
    private int _cursor;
    private int _count;

    public FrameStatsBuffer(int capacity)
    {
        if (capacity <= 0) throw new ArgumentOutOfRangeException(nameof(capacity));
        _samples = new long[capacity];
        _scratch = new long[capacity];
    }

    public int Count => _count;

    public void Push(long value)
    {
        _samples[_cursor] = value;
        _cursor = (_cursor + 1) % _samples.Length;
        if (_count < _samples.Length) _count++;
    }

    public void Reset()
    {
        _cursor = 0;
        _count = 0;
    }

    /// <summary>
    /// Nearest-rank percentile over the current window: element at
    /// ceil(q·n) in the ascending sort (1-based), 0 when empty.
    /// </summary>
    public long Percentile(double q)
    {
        if (_count == 0) return 0;
        Array.Copy(_samples, _scratch, _count);
        Array.Sort(_scratch, 0, _count);
        int rank = (int)Math.Ceiling(q * _count);   // 1-based nearest rank
        if (rank < 1) rank = 1;
        if (rank > _count) rank = _count;
        return _scratch[rank - 1];
    }

    public long Max()
    {
        long max = 0;
        for (int i = 0; i < _count; i++)
            if (_samples[i] > max) max = _samples[i];
        return max;
    }
}
```

- [ ] **Step 4: Run tests to verify they pass**

Run: `dotnet test tests/AcDream.App.Tests --filter FrameStatsBufferTests -v minimal`
Expected: 4 passed.

- [ ] **Step 5: Commit**

```bash
git add src/AcDream.App/Diagnostics/FrameStatsBuffer.cs tests/AcDream.App.Tests/FrameStatsBufferTests.cs
git commit -m "feat(pipeline): MP0 - FrameStatsBuffer ring/percentile core"
```

---

### Task 2: RenderingDiagnostics flag + GpuFrameTimer

**Files:**
- Modify: `src/AcDream.Core/Rendering/RenderingDiagnostics.cs` (append inside the class, before the closing brace)
- Create: `src/AcDream.App/Diagnostics/GpuFrameTimer.cs`

- [ ] **Step 1: Add the flag to RenderingDiagnostics**

Append inside `public static class RenderingDiagnostics` (after `ShouldRenderIndoor`, before the final `}`):

```csharp
    /// <summary>
    /// MP0 (2026-07-05) — master toggle for the permanent frame profiler
    /// (<c>AcDream.App.Diagnostics.FrameProfiler</c>): CPU frame time
    /// (swap-to-swap), whole-frame GPU time, per-stage CPU attribution,
    /// per-frame allocation counters, reported as one <c>[frame-prof]</c>
    /// line every ~5 s. Permanent apparatus (every MP-track gate reads it) —
    /// do NOT strip with session probes. The whole-frame GPU query is
    /// self-disabled while <c>ACDREAM_WB_DIAG=1</c> (GL forbids nested
    /// TimeElapsed queries; WbDrawDispatcher owns per-pass queries under
    /// that flag — the 2026-06-23 "separate flags" measurement lesson).
    /// Initial state from <c>ACDREAM_FRAME_PROF=1</c>; runtime-toggleable
    /// via the DebugPanel mirror (<c>DebugVM.FrameProf</c>).
    /// Spec: docs/superpowers/specs/2026-07-05-modern-pipeline-design.md §5.
    /// </summary>
    public static bool FrameProfEnabled { get; set; } =
        Environment.GetEnvironmentVariable("ACDREAM_FRAME_PROF") == "1";
```

- [ ] **Step 2: Create GpuFrameTimer**

```csharp
using System;
using Silk.NET.OpenGL;

namespace AcDream.App.Diagnostics;

/// <summary>
/// MP0 (2026-07-05) — whole-frame GPU time via a ring of
/// <see cref="QueryTarget.TimeElapsed"/> queries (depth 4, so results are
/// read ~3 frames late and never stall). Mirrors WbDrawDispatcher's query
/// idiom including the #125 lesson: a glGenQueries name is not a query
/// OBJECT until first glBeginQuery, so never-begun slots are skipped via
/// the Begun flags.
///
/// <para>MUST NOT be active while ACDREAM_WB_DIAG=1: GL forbids two
/// simultaneously active TimeElapsed queries and WbDrawDispatcher brackets
/// its passes with them under that flag. The caller (FrameProfiler)
/// enforces the exclusion; this class just does the ring.</para>
/// </summary>
internal sealed class GpuFrameTimer : IDisposable
{
    private const int RingDepth = 4;

    private readonly GL _gl;
    private readonly uint[] _queries = new uint[RingDepth];
    private readonly bool[] _begun = new bool[RingDepth];
    private int _frameIndex;
    private bool _queryActive;

    public GpuFrameTimer(GL gl)
    {
        _gl = gl ?? throw new ArgumentNullException(nameof(gl));
        for (int i = 0; i < RingDepth; i++)
            _queries[i] = _gl.GenQuery();
    }

    /// <summary>
    /// Call once per frame at the frame boundary. Ends the previous
    /// frame's query, polls the oldest slot non-blocking, begins this
    /// frame's query. Returns the completed GPU time in microseconds for
    /// a ~RingDepth-frames-old frame, or null when no result is ready.
    /// </summary>
    public long? FrameBoundary()
    {
        if (_queryActive)
        {
            _gl.EndQuery(QueryTarget.TimeElapsed);
            _queryActive = false;
        }

        long? completedUs = null;
        int readSlot = _frameIndex % RingDepth;   // about to be reused — oldest
        if (_begun[readSlot])
        {
            _gl.GetQueryObject(_queries[readSlot], QueryObjectParameterName.ResultAvailable, out int avail);
            if (avail != 0)
            {
                _gl.GetQueryObject(_queries[readSlot], QueryObjectParameterName.Result, out ulong ns);
                completedUs = (long)(ns / 1000UL);
            }
            // Not available ⇒ sample silently dropped (same policy as
            // WbDrawDispatcher) — percentiles tolerate missing samples.
        }

        _gl.BeginQuery(QueryTarget.TimeElapsed, _queries[readSlot]);
        _begun[readSlot] = true;
        _queryActive = true;
        _frameIndex++;
        return completedUs;
    }

    /// <summary>End any active query without beginning a new one (used when the profiler is toggled off mid-session).</summary>
    public void Stop()
    {
        if (_queryActive)
        {
            _gl.EndQuery(QueryTarget.TimeElapsed);
            _queryActive = false;
        }
    }

    public void Dispose()
    {
        Stop();
        for (int i = 0; i < RingDepth; i++)
            _gl.DeleteQuery(_queries[i]);
    }
}
```

Note: `_gl.GenQuery()` / `_gl.DeleteQuery(uint)` are the Silk.NET single-name helpers; if this overload doesn't resolve on the project's Silk.NET version, use the array forms exactly as `WbDrawDispatcher` does (`grep -n "GenQueries" src/AcDream.App/Rendering/Wb/WbDrawDispatcher.cs`) and copy that call shape.

- [ ] **Step 3: Build**

Run: `dotnet build`
Expected: green (no tests possible for the GL class headless; correctness rides on the copied WbDrawDispatcher idiom + Task 5's live run).

- [ ] **Step 4: Commit**

```bash
git add src/AcDream.Core/Rendering/RenderingDiagnostics.cs src/AcDream.App/Diagnostics/GpuFrameTimer.cs
git commit -m "feat(pipeline): MP0 - FrameProfEnabled flag + GPU TimeElapsed ring"
```

---

### Task 3: FrameProfiler facade

**Files:**
- Create: `src/AcDream.App/Diagnostics/FrameProfiler.cs`
- Test: `tests/AcDream.App.Tests/FrameProfilerReportTests.cs`

- [ ] **Step 1: Write the failing test (report formatting is the testable part)**

```csharp
using AcDream.App.Diagnostics;
using Xunit;

namespace AcDream.App.Tests;

public class FrameProfilerReportTests
{
    [Fact]
    public void FormatReport_IsInvariantAndComplete()
    {
        var cpu = new FrameStatsBuffer(16);
        var gpu = new FrameStatsBuffer(16);
        var alloc = new FrameStatsBuffer(16);
        var stages = new[] { new FrameStatsBuffer(16), new FrameStatsBuffer(16), new FrameStatsBuffer(16) };
        for (long i = 1; i <= 10; i++)
        {
            cpu.Push(i * 1000);      // 1..10 ms in µs
            gpu.Push(i * 100);
            alloc.Push(i * 1024);    // bytes
            stages[0].Push(i * 200);
            stages[1].Push(i * 50);
            stages[2].Push(i * 10);
        }

        string line = FrameProfiler.FormatReport(
            frameCount: 10, cpu: cpu, gpu: gpu, gpuActive: true,
            alloc: alloc, gc0: 3, gc1: 1, gc2: 0, stages: stages);

        Assert.StartsWith("[frame-prof]", line);
        Assert.Contains("n=10", line);
        Assert.Contains("cpu_ms p50=5.0 p95=10.0 p99=10.0 max=10.0", line);
        Assert.Contains("gpu_ms p50=0.5", line);
        Assert.Contains("alloc_kb p50=5.0 max=10.0", line);
        Assert.Contains("gc=3/1/0", line);
        Assert.Contains("upd p50=1.0", line);   // stage 0: 200µs·5 = 1.0 ms
        Assert.DoesNotContain(",0", line.Replace("gc=3/1/0", ""));  // no comma decimals (invariant culture)
    }

    [Fact]
    public void FormatReport_GpuInactive_SaysWhy()
    {
        var empty = new FrameStatsBuffer(4);
        string line = FrameProfiler.FormatReport(
            frameCount: 0, cpu: empty, gpu: empty, gpuActive: false,
            alloc: empty, gc0: 0, gc1: 0, gc2: 0,
            stages: new[] { empty, empty, empty });
        Assert.Contains("gpu=off(wbdiag)", line);
    }
}
```

- [ ] **Step 2: Run tests to verify they fail**

Run: `dotnet test tests/AcDream.App.Tests --filter FrameProfilerReportTests -v minimal`
Expected: FAIL — `FrameProfiler` does not exist.

- [ ] **Step 3: Implement FrameProfiler**

```csharp
using System;
using System.Diagnostics;
using System.Globalization;
using System.Text;
using AcDream.Core.Rendering;
using Silk.NET.OpenGL;

namespace AcDream.App.Diagnostics;

/// <summary>Stage indices for per-frame CPU attribution.</summary>
public enum FrameStage
{
    /// <summary>Whole OnUpdate body (simulation + streaming apply).</summary>
    Update = 0,
    /// <summary>WbMeshAdapter.Tick — staged mesh/texture GPU upload drain.</summary>
    Upload = 1,
    /// <summary>ImGui Render (dev overlay).</summary>
    ImGui = 2,
}

/// <summary>
/// MP0 (2026-07-05) — the permanent honest frame profiler. One
/// <c>FrameBoundary</c> call at the top of <c>GameWindow.OnRender</c>
/// measures CPU frame time as the delta between consecutive boundaries
/// (captures the FULL frame including present), brackets the frame in a
/// GPU <c>TimeElapsed</c> query (via <see cref="GpuFrameTimer"/>), and
/// samples per-frame allocated bytes + GC collection counts. Stage scopes
/// (<see cref="BeginStage"/>) attribute CPU time to Update / Upload /
/// ImGui. Emits one <c>[frame-prof]</c> line every ~5 s while
/// <see cref="RenderingDiagnostics.FrameProfEnabled"/> is true; costs one
/// bool check per frame when off.
///
/// <para>Permanent apparatus — every MP-track gate reads it; do not strip.
/// Whole-frame GPU timing self-disables under <c>ACDREAM_WB_DIAG=1</c>
/// (nested TimeElapsed is illegal GL; see GpuFrameTimer).</para>
/// Spec: docs/superpowers/specs/2026-07-05-modern-pipeline-design.md §5.
/// </summary>
public sealed class FrameProfiler : IDisposable
{
    private const int WindowCapacity = 2048;              // ~12 s at 165 fps
    private const long ReportIntervalTicks = 5 * TimeSpan.TicksPerSecond;
    private static readonly int StageCount = Enum.GetValues<FrameStage>().Length;

    private readonly FrameStatsBuffer _cpuUs = new(WindowCapacity);
    private readonly FrameStatsBuffer _gpuUs = new(WindowCapacity);
    private readonly FrameStatsBuffer _allocBytes = new(WindowCapacity);
    private readonly FrameStatsBuffer[] _stageUs;
    private readonly long[] _stageAccumTicks;
    private readonly bool _wbDiagActive =
        Environment.GetEnvironmentVariable("ACDREAM_WB_DIAG") == "1";

    private GpuFrameTimer? _gpuTimer;
    private long _lastBoundaryTimestamp;
    private long _lastAllocBytes;
    private long _lastReportTicks;
    private int _gc0Base, _gc1Base, _gc2Base;
    private int _framesInWindow;
    private int _ownerThreadId;
    private bool _threadWarned;
    private bool _wbDiagNoticePrinted;
    private bool _wasEnabled;

    public FrameProfiler()
    {
        _stageUs = new FrameStatsBuffer[StageCount];
        for (int i = 0; i < StageCount; i++) _stageUs[i] = new FrameStatsBuffer(WindowCapacity);
        _stageAccumTicks = new long[StageCount];
    }

    /// <summary>
    /// Call as the FIRST statement of <c>GameWindow.OnRender</c>.
    /// </summary>
    public void FrameBoundary(GL gl)
    {
        bool enabled = RenderingDiagnostics.FrameProfEnabled;
        if (!enabled)
        {
            if (_wasEnabled) { _gpuTimer?.Stop(); _wasEnabled = false; _lastBoundaryTimestamp = 0; }
            return;
        }

        if (_ownerThreadId == 0) _ownerThreadId = Environment.CurrentManagedThreadId;
        else if (!_threadWarned && _ownerThreadId != Environment.CurrentManagedThreadId)
        {
            _threadWarned = true;
            Console.WriteLine("[frame-prof] WARNING: frame boundary crossed threads; alloc counter is per-thread and now unreliable");
        }

        long now = Stopwatch.GetTimestamp();
        long allocNow = GC.GetAllocatedBytesForCurrentThread();

        if (!_wasEnabled)
        {
            // First enabled frame (startup or runtime toggle-on): establish
            // baselines, emit nothing.
            _wasEnabled = true;
            _lastReportTicks = DateTime.UtcNow.Ticks;
            _gc0Base = GC.CollectionCount(0); _gc1Base = GC.CollectionCount(1); _gc2Base = GC.CollectionCount(2);
            if (_gpuTimer is null && !_wbDiagActive) _gpuTimer = new GpuFrameTimer(gl);
            if (_wbDiagActive && !_wbDiagNoticePrinted)
            {
                _wbDiagNoticePrinted = true;
                Console.WriteLine("[frame-prof] GPU frame timing OFF: ACDREAM_WB_DIAG=1 owns TimeElapsed queries (nested queries are illegal GL)");
            }
        }
        else
        {
            long cpuUs = (now - _lastBoundaryTimestamp) * 1_000_000L / Stopwatch.Frequency;
            _cpuUs.Push(cpuUs);
            _allocBytes.Push(allocNow - _lastAllocBytes);
            for (int i = 0; i < StageCount; i++)
            {
                _stageUs[i].Push(_stageAccumTicks[i] * 1_000_000L / Stopwatch.Frequency);
                _stageAccumTicks[i] = 0;
            }
            _framesInWindow++;
        }

        _lastBoundaryTimestamp = now;
        _lastAllocBytes = allocNow;

        if (_gpuTimer?.FrameBoundary() is long gpuUs)
            _gpuUs.Push(gpuUs);

        long nowTicks = DateTime.UtcNow.Ticks;
        if (nowTicks - _lastReportTicks >= ReportIntervalTicks && _framesInWindow > 0)
        {
            int gc0 = GC.CollectionCount(0) - _gc0Base;
            int gc1 = GC.CollectionCount(1) - _gc1Base;
            int gc2 = GC.CollectionCount(2) - _gc2Base;
            Console.WriteLine(FormatReport(_framesInWindow, _cpuUs, _gpuUs,
                gpuActive: _gpuTimer is not null, _allocBytes, gc0, gc1, gc2, _stageUs));
            _lastReportTicks = nowTicks;
            _gc0Base += gc0; _gc1Base += gc1; _gc2Base += gc2;
            _framesInWindow = 0;
            _cpuUs.Reset(); _gpuUs.Reset(); _allocBytes.Reset();
            for (int i = 0; i < StageCount; i++) _stageUs[i].Reset();
        }
    }

    /// <summary>
    /// Attribute the enclosed CPU time to <paramref name="stage"/>.
    /// Usage: <c>using var _ = profiler.BeginStage(FrameStage.Update);</c>.
    /// Zero-cost (default scope) when the profiler is off.
    /// </summary>
    public StageScope BeginStage(FrameStage stage)
        => RenderingDiagnostics.FrameProfEnabled
            ? new StageScope(this, stage, Stopwatch.GetTimestamp())
            : default;

    internal void EndStage(FrameStage stage, long startTimestamp)
        => _stageAccumTicks[(int)stage] += Stopwatch.GetTimestamp() - startTimestamp;

    /// <summary>Pure report formatter — unit-tested; invariant culture.</summary>
    public static string FormatReport(
        int frameCount,
        FrameStatsBuffer cpu, FrameStatsBuffer gpu, bool gpuActive,
        FrameStatsBuffer alloc, int gc0, int gc1, int gc2,
        FrameStatsBuffer[] stages)
    {
        var ci = CultureInfo.InvariantCulture;
        var sb = new StringBuilder(256);
        sb.Append("[frame-prof] n=").Append(frameCount);
        sb.AppendFormat(ci, " | cpu_ms p50={0:0.0} p95={1:0.0} p99={2:0.0} max={3:0.0}",
            cpu.Percentile(0.50) / 1000.0, cpu.Percentile(0.95) / 1000.0,
            cpu.Percentile(0.99) / 1000.0, cpu.Max() / 1000.0);
        if (gpuActive)
            sb.AppendFormat(ci, " | gpu_ms p50={0:0.0} p95={1:0.0}",
                gpu.Percentile(0.50) / 1000.0, gpu.Percentile(0.95) / 1000.0);
        else
            sb.Append(" | gpu=off(wbdiag)");
        sb.AppendFormat(ci, " | alloc_kb p50={0:0.0} max={1:0.0} gc={2}/{3}/{4}",
            alloc.Percentile(0.50) / 1024.0, alloc.Max() / 1024.0, gc0, gc1, gc2);
        string[] names = { "upd", "upl", "imgui" };
        for (int i = 0; i < stages.Length && i < names.Length; i++)
            sb.AppendFormat(ci, " | {0} p50={1:0.0} p95={2:0.0}",
                names[i], stages[i].Percentile(0.50) / 1000.0, stages[i].Percentile(0.95) / 1000.0);
        return sb.ToString();
    }

    public void Dispose() => _gpuTimer?.Dispose();
}

/// <summary>Disposable stage scope; default instance is a no-op.</summary>
public readonly struct StageScope : IDisposable
{
    private readonly FrameProfiler? _owner;
    private readonly FrameStage _stage;
    private readonly long _start;

    internal StageScope(FrameProfiler owner, FrameStage stage, long start)
    {
        _owner = owner; _stage = stage; _start = start;
    }

    public void Dispose() => _owner?.EndStage(_stage, _start);
}
```

- [ ] **Step 4: Run tests to verify they pass**

Run: `dotnet test tests/AcDream.App.Tests --filter FrameProfilerReportTests -v minimal`
Expected: 2 passed. (If the `,0` invariant assertion trips on stage names, adjust only the assertion's exclusion — the formatter must keep `CultureInfo.InvariantCulture` everywhere.)

- [ ] **Step 5: Commit**

```bash
git add src/AcDream.App/Diagnostics/FrameProfiler.cs tests/AcDream.App.Tests/FrameProfilerReportTests.cs
git commit -m "feat(pipeline): MP0 - FrameProfiler facade (CPU/GPU/alloc/stages, 5s report)"
```

---

### Task 4: GameWindow wiring + DebugPanel mirror

**Files:**
- Modify: `src/AcDream.App/Rendering/GameWindow.cs`
- Modify: `src/AcDream.UI.Abstractions/Panels/Debug/DebugVM.cs`
- Modify: the DebugPanel ImGui renderer (locate in step 4)

- [ ] **Step 1: Add the field and frame-boundary call**

Near GameWindow's other diagnostic fields (search for `_frameDiag` field declaration to co-locate):

```csharp
    // MP0 (2026-07-05): permanent frame profiler — one FrameBoundary call
    // per OnRender + three stage scopes. All logic lives in
    // AcDream.App.Diagnostics.FrameProfiler (structure rule 1).
    private readonly AcDream.App.Diagnostics.FrameProfiler _frameProfiler = new();
```

At the very top of `OnRender` (anchor: `private void OnRender(double deltaSeconds)` ~line 8756 — the call goes BEFORE the `var kf = WorldTime.CurrentSky;` line):

```csharp
        _frameProfiler.FrameBoundary(_gl!);
```

In `Dispose`/`OnClosing` teardown (anchor: where `_wbMeshAdapter` is disposed — profiler goes before `_dats`):

```csharp
        _frameProfiler.Dispose();
```

- [ ] **Step 2: Add the three stage scopes**

`OnUpdate` (anchor: `private void OnUpdate(double dt)` ~line 7952) — first statement of the body:

```csharp
        using var _updStage = _frameProfiler.BeginStage(AcDream.App.Diagnostics.FrameStage.Update);
```

Upload drain (anchor: `_wbMeshAdapter?.Tick();` ~line 8813 — wrap it, keeping the existing `[FRAME-DIAG]` lines untouched):

```csharp
        using (var _uplStage = _frameProfiler.BeginStage(AcDream.App.Diagnostics.FrameStage.Upload))
        {
            _wbMeshAdapter?.Tick();
        }
```

ImGui render (anchor: `_imguiBootstrap.Render();` ~line 9821):

```csharp
            using (var _imguiStage = _frameProfiler.BeginStage(AcDream.App.Diagnostics.FrameStage.ImGui))
            {
                _imguiBootstrap.Render();
            }
```

- [ ] **Step 3: Add the DebugVM mirror**

In `src/AcDream.UI.Abstractions/Panels/Debug/DebugVM.cs`, after the `ProbeResolve` property (~line 252), following the same pattern:

```csharp
    /// <summary>
    /// Runtime mirror of <c>RenderingDiagnostics.FrameProfEnabled</c>
    /// (env var <c>ACDREAM_FRAME_PROF</c>). Toggling here starts/stops the
    /// [frame-prof] 5-second report live — no relaunch required.
    /// </summary>
    public bool FrameProf
    {
        get => AcDream.Core.Rendering.RenderingDiagnostics.FrameProfEnabled;
        set => AcDream.Core.Rendering.RenderingDiagnostics.FrameProfEnabled = value;
    }
```

- [ ] **Step 4: Add the DebugPanel checkbox**

Locate the renderer: `grep -rn "ProbeResolve" src/AcDream.UI.ImGui/` — expected: one hit in the DebugPanel renderer drawing a checkbox bound to `vm.ProbeResolve`. Add an identical checkbox line for `vm.FrameProf` labeled `"Frame profiler ([frame-prof])"` adjacent to it, copying the exact checkbox idiom used there.

- [ ] **Step 5: Build + full test suite**

Run: `dotnet build`
Expected: green.
Run: `dotnet test`
Expected: all green (no existing test touches these paths).

- [ ] **Step 6: Commit**

```bash
git add src/AcDream.App/Rendering/GameWindow.cs src/AcDream.UI.Abstractions/Panels/Debug/DebugVM.cs src/AcDream.UI.ImGui/
git commit -m "feat(pipeline): MP0 - wire FrameProfiler into GameWindow + DebugPanel toggle"
```

---

### Task 5: Baseline capture (user-driven; MP0's deliverable + gate)

**Files:**
- Create: `docs/research/2026-07-XX-mp0-baseline.md` (use the actual capture date)

- [ ] **Step 1: Confirm Release build is green**

Run: `dotnet build -c Release`
Expected: green. Never launch on a red build.

- [ ] **Step 2: Launch with the profiler on (user drives)**

```powershell
$env:ACDREAM_DAT_DIR="$env:USERPROFILE\Documents\Asheron's Call"; $env:ACDREAM_LIVE="1"
$env:ACDREAM_TEST_HOST="127.0.0.1"; $env:ACDREAM_TEST_PORT="9000"
$env:ACDREAM_TEST_USER="testaccount"; $env:ACDREAM_TEST_PASS="testpassword"
$env:ACDREAM_FRAME_PROF="1"    # NOT ACDREAM_WB_DIAG — mutually exclusive GPU timing
dotnet run --project src\AcDream.App\AcDream.App.csproj --no-build -c Release 2>&1 | Tee-Object -FilePath "mp0-baseline.log"
```

Launch in the background (`run_in_background: true`); the user manages the client lifecycle per CLAUDE.md. Ask the user to run this route, pausing ~30 s (≥5 report lines) in each state:
1. Dense town, standing still, facing the dense view (Holtburg center; Fort Tethana/Arwic if reachable).
2. Dense town, panning the camera continuously.
3. Walking a full town traversal (streaming active).
4. 2–3 portal hops (the hitch scenario).
5. A dungeon interior walk.

- [ ] **Step 3: Extract the numbers**

Run: `Select-String -Path mp0-baseline.log -Pattern "\[frame-prof\]"`
Expected: one line per 5 s. Map line timestamps/order to the route states (the user says when each state started; correlate by order).

- [ ] **Step 4: Write the baseline report**

Create `docs/research/2026-07-XX-mp0-baseline.md` containing, per route state: cpu_ms p50/p95/p99/max, gpu_ms p50/p95, alloc_kb p50/max, gc counts, stage split (upd/upl/imgui p50/p95) — plus:
- **The attribution verdict:** where do the ~6 dense-town CPU ms actually go (update vs upload vs imgui vs unattributed-render)? Does it confirm the spec §1 split (render-submission dominant)?
- **The gate decision (spec §5):** if attribution confirms the assumed split → proceed to MP1 planning unchanged. If it contradicts (e.g. Update dominates) → amend the spec's phase order BEFORE MP1, in the same session, and say so in this doc.
- The hitch profile: worst max/p99 during portal hops and traversal — this is MP1's "before" number.

- [ ] **Step 5: Commit + update the spec status**

Mark the spec's MP0 row as baseline-captured (one-line edit in `docs/superpowers/specs/2026-07-05-modern-pipeline-design.md` §5 gate line: append "— PASSED <date>, see baseline doc").

```bash
git add docs/research/2026-07-XX-mp0-baseline.md docs/superpowers/specs/2026-07-05-modern-pipeline-design.md
git commit -m "docs(pipeline): MP0 baseline captured - attribution verdict + gate decision"
```

---

## Post-plan notes for the executor

- **Do not strip `[frame-prof]`** in future probe-cleanup commits — it is permanent apparatus (the flag doc comment says so; so does spec §5).
- The existing `[FRAME-DIAG]` / `[WB-DIAG]` apparatus is untouched by this plan — the distance-degrade handoff still owns its strip decision.
- The roadmap "Modern Pipeline side track" section + milestones-doc freeze-exception paragraph (spec §12) should be added in the same commit as Task 5's baseline doc if not already present.