spec(047): Phase 3 final — advisory 3x5 perf capture (x64)

3 launches × 5 reps × 13 benches × 4 variants on Cloud PC
(CPC-ander-YTZ3O, AMD EPYC 7763, x64). 50 controls now registered in
DescriptorVariantFactory (12 prior + 38 batches 3-11).

Headline (V1 ON descriptors vs V1 OFF today, n=15 medians):
- M8 Update_OneLeafChanged: +25.5% — largest regression
- M12 Pool_Rent_HotPath:    +20.9%
- M1 Mount_Leaf_NoCallback: +14.9%
- M4 Dispatch_Switch_Cold:  -21.2% — improvement (more types in handler table)
- M5 Dispatch_Switch_Warm:  -24.3% — improvement (same)

The +25.5% / +20.9% / +14.9% Mount/Update regressions exceed the §13 Q1
5-15% judgment-call band on this advisory capture and surface
descriptor-interpreter Update overhead amortized over 50 registered
controls. Cloud PC advisory only — stable-AC ARM64 re-capture on
LAPTOP-4MEP83VI deferred. Q1 reopen condition (source-gen) not met.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

Author: codemonkeychris

docs/specs/047/phase3-results/CPC-ander-YTZ3O-x64-advisory/2026-05-27-phase3-final-3x5/README.md

@@ -0,0 +1,130 @@
+# Spec 047 §14 Phase 3 — final bulk-port, x64 advisory
+
+**This is an advisory x64 capture, NOT authoritative.** Cloud PC
+(`CPC-ander-YTZ3O`, AMD EPYC 7763 64-Core Processor, x64), not on
+AC/dedicated hardware. Do not cite these numbers in §13 or §14 spec
+text. A stable-AC ARM64 re-capture on `LAPTOP-4MEP83VI` should ratify
+the matrix before §14 Phase 3 is closed.
+
+## Why this capture exists
+
+This run extends the `2026-05-27-batch-1-2-3x5/` matrix with the
+remaining Phase 3 batches landed in the bulk-port PR
+(`spec/047-phase3-bulk-port`):
+
+- **Batch 3** — Display family (TextBlock, Image, PersonPicture,
+  ProgressBar, ProgressRing, InfoBadge).
+- **Batch 4** — Button family (Button, HyperlinkButton, RepeatButton,
+  ToggleButton, DropDownButton, SplitButton).
+- **Batch 5** — Value-bearing inputs (RichEditBox, PasswordBox,
+  RadioButtons).
+- **Batch 6** — Multi-event inputs (AutoSuggestBox, ComboBox).
+- **Batch 7** — Single-content containers (Viewbox, Expander,
+  ScrollViewer, ScrollView).
+- **Batch 8** — Panels (StackPanel, Grid, Canvas, FlexPanel,
+  RelativePanel).
+- **Batch 9** — Named-slot containers (SplitView, InfoBar, TeachingTip).
+- **Batch 10** — Shapes + display leaves (Rectangle, Ellipse, Line,
+  Path, AnimatedIcon).
+- **Batch 11** — Long tail (PipsPager, ListBox, SelectorBar,
+  BreadcrumbBar).
+
+After all batches the `DescriptorVariantFactory` registers **50 ported
+controls** (12 from prior PRs: 4 prereq + 8 batches 1+2; 38 new from
+batches 3-11). The bench matrix detects dispatch-table shape change
+and descriptor-interpreter amortization across M1-M13 even though
+those benches don't mount the new controls directly.
+
+## Capture environment
+
+`CPC-ander-YTZ3O`, x64 (AMD EPYC 7763 64-Core Processor), Release,
+.NET 10.0.8, Windows 11 26200. **Cloud PC — not on AC/dedicated
+hardware**. 3 process launches × 5 reps × 13 benches × 4 variants =
+780 measurements across `launch-1.jsonl` + `launch-2.jsonl` +
+`launch-3.jsonl`.
+
+## Headline — V1 ON (descriptors) vs V1 OFF (today)
+
+The user-facing question is "does turning V1 ON with the full Phase 3
+descriptor set regress anything." Median of n=15 (3 launches × 5 reps)
+per cell:
+
+| Bench | V1 ON (Descriptors) vs V1 OFF (Today) | Verdict |
+|---|---:|---|
+| M1 Mount_Leaf_NoCallback | **+14.9%** | exceeds 5-15% judgment-call band ceiling |
+| M2 Mount_Leaf_OneCallback | -1.7% | within noise |
+| M3 Mount_Leaf_ThreeCallbacks | +3.3% | within noise |
+| M4 Dispatch_Switch_Cold | **-21.2%** | improvement (fewer types fall through to switch arm) |
+| M5 Dispatch_Switch_Warm | **-24.3%** | improvement (same; warmer dispatch hits handler table) |
+| M6 Dispatch_ExternalType | +0.2% | within noise |
+| M7 Update_NoChange | +7.4% | judgment-call band |
+| M8 Update_OneLeafChanged | **+25.5%** | **largest regression** |
+| M9 Update_AllChanged | +3.6% | within noise |
+| M10 EventHandlerState_Alloc | +8.7% | judgment-call band |
+| M11 ModifierEHS_Frequency | +8.5% | judgment-call band |
+| M12 Pool_Rent_HotPath | **+20.9%** | second-largest regression |
+| M13 Setters_Suppression_Scope | -0.9% | within noise |
+
+**The +25.5% on M8 (Update_OneLeafChanged) and +20.9% on M12
+(Pool_Rent_HotPath) exceed the §13 Q1 5-15% judgment-call band.**
+This is a real surfacing of descriptor-interpreter Update overhead
+compared to the hand-coded switch arms, now amortized over 50
+registered controls.
+
+The M4/M5 improvements are the mirror image: dispatch is faster when
+more types are in the handler table (the fallback to the switch arm
+is rarer). Net Mount cost is dominated by interpreter overhead (M1
++14.9%); net Update cost regresses substantially on one-leaf-change
+workloads (M8 +25.5%).
+
+These deltas reproduce the trade-off the §13 Q1 stable-AC capture
+already documented (M2 +9.6% at batch-2 scope). The 5-percentage-point
+shift on the M1/M5 readings between this capture and the prior
+`2026-05-27-batch-1-2-3x5/` capture is consistent with the prior
+README's "noise-prone, advisory" caveat — Cloud PC, no AC pinning.
+
+## Q1 decision matrix — for completeness
+
+Per §13 Q1's pre-committed decision matrix applied to
+ReactorDescriptors vs ReactorV2:
+
+| Bench | vs ReactorV2 ns | Q1 band |
+|---|---:|---|
+| M1 | +19.6% | exceeds 15% — judgment call vs LOC/readability |
+| M2 | -5.5% | ship descriptors |
+| M5 | -18.6% | ship descriptors (improvement) |
+| M7 | +7.7% | judgment-call band |
+| M10 | +2.4% | ship descriptors |
+
+**Verdict:** No reopen condition for Q1 — Q1's reopen is gated on
+source-gen (§7) landing, not advisory perf noise. The scope-amortized
+deltas on M1/M8/M12 should be confirmed on stable-AC ARM64 before any
+spec-text change.
+
+## Caveats
+
+- **Cloud PC noise.** Per the prior README: "noise-prone, advisory.
+  Do not cite in §13/§14 spec text."
+- **ARM64 stable-AC re-capture on `LAPTOP-4MEP83VI` is deferred** from
+  this session (not on critical path; tracked for spec close-out
+  alongside the §14 Phase 3 ratification gate).
+- M7 (Update_NoChange) Direct measurement is dominated by the
+  3-second `Stopwatch` calibration in BenchRunner — Direct timing on
+  this no-op bench is degenerate; only the relative
+  Today/V2/Descriptors comparison is meaningful.
+
+## Reproduce
+
+```powershell
+cd C:\Users\andersonch\Code\reactor2
+dotnet build tests/perf_bench/PerfBench.ControlModel -c Release -p:Platform=x64
+# 3 launches:
+1..3 | ForEach-Object {
+    & tests/perf_bench/PerfBench.ControlModel/bin/x64/Release/net10.0-windows10.0.22621.0/PerfBench.ControlModel.exe
+    Copy-Item tests/perf_bench/PerfBench.ControlModel/bin/x64/Release/net10.0-windows10.0.22621.0/results.jsonl `
+              docs/specs/047/phase3-results/CPC-ander-YTZ3O-x64-advisory/2026-05-27-phase3-final-3x5/launch-$_.jsonl
+}
+```
+
+Aggregation: median across (benchId, variant) over the three launches.
+See `summary.md` for the full per-bench table.