Switch benchmarks to warmup config; re-derive thresholds

Monitoring strategy with WarmupCount=1, IterationCount=4 — same total
work as before, but iteration 1's JIT/EF-model-build/cold-cache cost is
discarded instead of folded into the measurement, so StdDev tightens and
thresholds can come down to genuine 3σ above the with-index mean.

Both comment lines (baseline + with-index) are re-measured under this
config so the speedup comparison is apples-to-apples.

Author: myieye

backend/FwLite/FwLiteProjectSync.Tests/BenchmarkSupport.cs

@@ -13,8 +13,9 @@ namespace FwLiteProjectSync.Tests;
 internal static class BenchmarkSupport
 {
     /// <summary>
-    /// Standard config for sync benchmarks: in-process toolchain, cold iterations, and a logger
-    /// that pipes the BDN summary table into xUnit's test output.
+    /// Standard config for sync benchmarks: in-process toolchain, one warmup + four measurement
+    /// iterations (Monitoring strategy), and a logger that pipes the BDN summary table into
+    /// xUnit's test output.
     /// </summary>
     /// <remarks>
     /// The in-process toolchain is what lets a benchmark class read a static field set by its
@@ -25,14 +26,20 @@ internal static class BenchmarkSupport
     /// for multi-iteration sync benchmarks (e.g. delete-heavy alone is ~80s/iter and full-import
     /// setup adds ~50s/iter, so 5 iterations need ~11 min). 30 min covers the worst case with
     /// headroom.
+    ///
+    /// Monitoring + WarmupCount=1 means iteration 1 absorbs JIT + EF model build + first-touch
+    /// file cache and is discarded; iterations 2-5 are measured. ColdStart would skip warmup
+    /// entirely; for these slow ops the JIT/model-build noise in iteration 1 is large enough
+    /// that one discarded warmup tightens StdDev meaningfully without changing total CI time.
     /// </remarks>
     public static IConfig ConfigFor(ITestOutputHelper output)
     {
         var toolchain = new InProcessNoEmitToolchain(TimeSpan.FromMinutes(30), logOutput: false);
         return ManualConfig.CreateEmpty()
             .AddJob(Job.Default
-                .WithStrategy(RunStrategy.ColdStart)
-                .WithIterationCount(5)
+                .WithStrategy(RunStrategy.Monitoring)
+                .WithWarmupCount(1)
+                .WithIterationCount(4)
                 .WithToolchain(toolchain))
             .AddExporter(JsonExporter.FullCompressed)
             .AddColumnProvider(DefaultColumnProviders.Instance)

backend/FwLite/FwLiteProjectSync.Tests/SyncBenchmark.cs

@@ -52,9 +52,9 @@ public void First_Sync_Sena3()
 
 public class FirstSyncBench
 {
-    // CI initial result:           mean 49.5s, StdDev 2.4s (medium variance) => 57s (~3σ above mean)
-    // CI with commits order index: mean 44.5s, StdDev 3.5s (medium variance) => 55s (~3σ above mean)
-    public const double ThresholdSeconds = 55.0;
+    // CI baseline (no index):      mean 52.34s, StdDev 0.27s (low variance) => 53s (~3σ above mean)
+    // CI with commits order index: mean 44.35s, StdDev 1.10s (low variance) => 50s (~5σ above mean, generous for run-to-run drift)
+    public const double ThresholdSeconds = 50.0;
 
     internal static Sena3Fixture Fixture = null!;
 

backend/FwLite/FwLiteProjectSync.Tests/SyncMutationBenchmark.cs

@@ -69,20 +69,20 @@ public class MutationSyncBench
 {
     public static readonly IReadOnlyDictionary<string, double> ThresholdSecondsByProfile = new Dictionary<string, double>
     {
-        // CI 2026-05-06:               mean 58.3s, StdDev 4.1s (high variance) => 72s (~4σ above mean)
-        // CI with commits order index: mean 46.9s, StdDev 2.2s (medium variance) => 53s (~3σ above mean)
-        ["component-heavy"] = 53.0,
-        // CI 2026-05-06:               mean 94.3s, StdDev 5.7s (high variance) => 115s (~4σ above mean)
-        // CI with commits order index: mean 91.6s, StdDev 4.3s (high variance) => 105s (~4σ above mean)
-        ["delete-heavy"] = 105.0,
-        // CI 2026-05-06:               mean 36.2s, StdDev 3.3s (medium variance) => 45s (~3σ above mean)
-        // CI with commits order index: mean 33.4s, StdDev 0.8s (low variance) => 36s (~3σ above mean)
-        ["mixed-realistic"] = 36.0,
-        // CI 2026-05-06:               mean 5.05s, StdDev 0.4s (low variance) => 7s (generous margin since it's already pretty fast and we want to avoid false positives from noise).
-        // CI with commits order index: mean 3.7s, StdDev 0.1s (low variance) => 5s (pretty fast, so meh)
+        // CI baseline (no index):      mean 52.40s, StdDev 1.51s (low variance) => 61s (~5σ above mean)
+        // CI with commits order index: mean 50.49s, StdDev 0.87s (low variance) => 61s (~12σ above mean — kept generous, run-to-run drift can be ~3s here)
+        ["component-heavy"] = 61.0,
+        // CI baseline (no index):      mean 87.02s, StdDev 3.85s (medium variance) => 97s (~3σ above mean)
+        // CI with commits order index: mean 87.92s, StdDev 1.42s (low variance)    => 97s (~6σ above mean — kept generous, run-to-run drift can be ~2s here)
+        ["delete-heavy"] = 97.0,
+        // CI baseline (no index):      mean 32.99s, StdDev 0.77s (low variance) => 38s (~7σ above mean)
+        // CI with commits order index: mean 32.95s, StdDev 1.08s (low variance) => 38s (~5σ above mean)
+        ["mixed-realistic"] = 38.0,
+        // CI baseline (no index):      mean 4.52s, StdDev 0.08s (low variance) => 5s (generous margin since it's already pretty fast and we want to avoid false positives from noise)
+        // CI with commits order index: mean 3.53s, StdDev 0.10s (low variance) => 5s (pretty fast, so meh — same margin works)
         ["patch-heavy"] = 5.0,
-        // CI 2026-05-06:               mean 0.77s, StdDev 0.2s (low variance) => 3s (super fast, so meh)
-        // CI with commits order index: mean 0.58s, StdDev 0.02s (low variance) => 2s (super fast, so meh)
+        // CI baseline (no index):      mean 0.69s, StdDev 0.08s (low variance)  => 2s (super fast, so meh)
+        // CI with commits order index: mean 0.57s, StdDev 0.05s (low variance)  => 2s (super fast, so meh)
         ["reorder-heavy"] = 2.0,
     };