test: stabilize load-sensitive release contracts

docs/reviews/2026-08-18-r3-test-truth-ledger.md

@@ -134,7 +134,7 @@ identifiable in the lane report.
 | T-011 51 output-only methods | classified in batch C | The reviewed current set is 51 methods / 70 cases. All carry `Purpose=Diagnostic`, preserving the apparatus while removing it from release pass totals. Contract-shaped names remain explicitly flagged until a stable oracle exists. |
 | T-012 source-text freezes | requires semantic replacement map | Retain whole-tree dependency rules; remove exact-text freezes only when an equivalent semantic/behavioral guard is identified. |
 | T-013 controller self-comparison | high-confidence cleanup batch A | Capture the first controller next to the first body and compare every retry with that reference. |
-| T-014 seven load-sensitive tests | active; first mechanism reproduced | `PortalProjectionTests.ClipToRegion_FrameOwnedStore_ReusesExactResultArray` reported 6,904 B only while three App test hosts ran concurrently, then passed in an isolated run. Preserve both observations and remove ambient process-allocation sensitivity rather than loosening the zero-allocation oracle. |
+| T-014 seven load-sensitive tests | six mechanisms repaired in batch F; one product defect classified | Virtual/wall-clock mixing, tiered-JIT allocation noise, a live controller clock, and a ThreadPool-start timing oracle were removed without weakening behavioral contracts. `DatSoundCache` #321 is a real in-flight-entry race and now runs as `Status=KnownFailure` pending a product fix. |
 | T-015 four non-prerequisite skips | resolved in batches A/B | PVS scaffold deleted with rationale preserved; redundant chat/radar generators deleted in favor of the comprehensive Manual lane; tower oracle is `Status=KnownFailure`. |
 | T-016 historical test taxonomy | open | Rename/re-home only after each test's durable owner and oracle are established. |
 | T-017 Avalonia ownership | reopened and closed in batch D | The full gate exposed the same compositor ownership class between six newer `MainWindowViewTests` facts. Their six named assertion phases now run in one owned Avalonia application session; 25 fresh-process stress iterations and the complete gate pass. |
@@ -330,3 +330,43 @@ recognition. The public production factory remains wired directly to
 The focused class passes 4/4 in 17 ms with exact 10 ms and 600 ms virtual
 intervals. The complete Release build then passed with 0 warnings/errors, and
 the no-retry hermetic gate passed 14,392/14,392 with zero skips or failures.
+
+## Batch F load-sensitive contracts
+
+The seven cases in T-014 have now been separated into six unreliable test
+mechanisms and one real product defect:
+
+- #308 `NakEmissionTests.LossSoak...` no longer mixes its virtual transport
+  clock with a 60-second `DateTime.UtcNow` cutoff and fixed sleeps. Its two
+  convergence phases each permit exactly 120 half-second virtual steps and
+  yield to the background receiver. The full 10,000-message oracle is intact;
+  25 fresh-process repetitions pass.
+- #302 and #346, the two `PortalProjectionTests` allocation contracts, now
+  cross tiered-JIT/PGO thresholds before measuring and take the minimum of five
+  warmed 1,000-operation batches. A linear result-array regression would
+  allocate in every batch and still fail far above the existing ceilings.
+- #336 `RuntimeCollisionReportingStateTests.WarmedSteadyContactRefreshDoesNotAllocate`
+  now earns the word "warmed": it performs one complete 10,000-refresh warmup
+  and requires at least one of five subsequent 10,000-refresh batches to
+  allocate exactly zero bytes. It passes in 25 fresh processes.
+- #340 `StreamingWorkBudgetTests.DestinationAndEmptyUnloadPriorityNeverBypassPublicationBudget`
+  now supplies a frozen meter clock through an internal test-only constructor.
+  The public production constructor still supplies `Stopwatch.GetTimestamp`
+  and `Stopwatch.Frequency` exactly as before.
+- #402 `LandblockBuildFactoryTests.Build_UsesTheSuppliedSharedReaderGate` now
+  starts a dedicated thread and observes it in `WaitSleepJoin` on the held
+  monitor before checking that no DAT read occurred. It no longer treats a
+  ThreadPool start within two seconds or an arbitrary 100 ms wait as evidence.
+- #321 `DatSoundCacheTests.GetWave_ConcurrentSameId_PublishesOneCanonicalWaveAndDecodesOnce`
+  exposes a real race. A caller can pass the resident-cache check, pause, and
+  reach `_inflight` after the winning caller has admitted the wave and removed
+  its `Lazy`, producing a second decode. The assertion and rationale are
+  preserved under `Status=KnownFailure`; R3 makes no cache behavior change.
+
+The four changed App cases pass 100/100 across 25 fresh processes. The Runtime
+allocation case passes 25/25, and the Core.Net loss soak passes 25/25. Complete
+batch verification then passed: the 44-project Release build reports zero
+warnings/errors, the refreshed inventory reports two explicit known-failure
+methods and 15 remaining direct `Thread.Sleep` methods, and the no-retry
+hermetic gate passed 14,391/14,391 with zero skips or failures. The one-pass
+reduction is exactly #321 leaving the release lane.