Align recognition replay pressure with coach review continuity

UnrealHyperTwist/Source/UnrealHyperTwist/Private/HyperTwistTraining/HyperTwistTrainingCoachLibrary.cpp

@@ -2300,6 +2300,17 @@ namespace HyperTwistTrainingCoachLibraryInternal
 		});
 	}
 
+	bool ShouldPreferRecognitionReplayReview(
+		const FHyperTwistReplayReviewAnalytics& ReplayReviewAnalytics,
+		const FHyperTwistRecognitionReplaySummary& RecognitionReplaySummary)
+	{
+		return RecognitionReplaySummary.EpisodeCount > 0
+			&& (RecognitionReplaySummary.AmbiguousEpisodeCount > 0
+				|| ReplayReviewAnalytics.RecognitionCorrectionCount > 0
+				|| (RecognitionReplaySummary.AverageCommitConfidence >= 0.0f
+					&& RecognitionReplaySummary.AverageCommitConfidence < 0.7f));
+	}
+
 	void ApplyPrimarySignalToBrief(
 		FHyperTwistCoachBrief& Brief,
 		const FHyperTwistCoachSignal& PrimarySignal)
@@ -2848,11 +2859,9 @@ TArray<FHyperTwistCoachSignal> UHyperTwistTrainingCoachLibrary::DeriveCoachSigna
 		Signals.Add(Signal);
 	}
 
-	if (RecognitionReplaySummary.EpisodeCount > 0
-		&& (RecognitionReplaySummary.AmbiguousEpisodeCount > 0
-			|| ReplayReviewAnalytics.RecognitionCorrectionCount > 0
-			|| (RecognitionReplaySummary.AverageCommitConfidence >= 0.0f
-				&& RecognitionReplaySummary.AverageCommitConfidence < 0.7f)))
+	if (HyperTwistTrainingCoachLibraryInternal::ShouldPreferRecognitionReplayReview(
+			ReplayReviewAnalytics,
+			RecognitionReplaySummary))
 	{
 		const float ConfidencePenalty = RecognitionReplaySummary.AverageCommitConfidence >= 0.0f
 			? (1.0f - RecognitionReplaySummary.AverageCommitConfidence) * 25.0f
@@ -3014,9 +3023,10 @@ FHyperTwistCoachBrief UHyperTwistTrainingCoachLibrary::DeriveCoachBrief(
 	Brief.bNeedsCadenceRecovery =
 		CoachMemorySnapshot.bNeedsCadenceRecovery || CoachMemorySnapshot.bNeedsScheduleFrictionRecovery;
 	Brief.bNeedsVarietyRotation = CoachMemorySnapshot.bNeedsVarietyRotation;
-	Brief.bPreferRecognitionReplayReview = RecognitionReplaySummary.EpisodeCount > 0
-		&& (RecognitionReplaySummary.AmbiguousEpisodeCount > 0
-			|| ReplayReviewAnalytics.RecognitionCorrectionCount > 0);
+	Brief.bPreferRecognitionReplayReview =
+		HyperTwistTrainingCoachLibraryInternal::ShouldPreferRecognitionReplayReview(
+			ReplayReviewAnalytics,
+			RecognitionReplaySummary);
 
 	Brief.Signals = DeriveCoachSignals(
 		SessionSummary,

docs/arch/HYPERTWIST_PHASE4_RECOGNITION_REVIEW_CONTINUITY_PACKET_2026-05-06.md

@@ -0,0 +1,89 @@
+# HyperTwist Phase 4 recognition review continuity packet
+
+Created on `2026-05-06`
+
+Status:
+
+- first-party HyperTwist packet
+- bounded Phase `4` replay-to-coach validation slice
+
+## Purpose
+
+This packet closes a coherence gap between the already-landed recognition replay signal lane and the coach brief / follow-up lane.
+
+The open task is:
+
+- make low-confidence recognition replay cases preserve recognition-review preference all the way into the coach brief and downstream coach consumers
+
+It is not:
+
+- a new recognition transport packet
+- a dashboard redesign
+- a broad coaching heuristic rewrite
+- a speech or simulation packet
+
+## Scope
+
+Bounded lane:
+
+- unify the recognition-review preference predicate used by:
+  - recognition clarity signal derivation
+  - coach brief derivation
+- keep the threshold consistent for:
+  - ambiguous recognition episodes
+  - replay correction pressure
+  - low average commit confidence
+
+Out of scope:
+
+- new provider behavior
+- frame-submission logic changes
+- replay packet schema changes
+- broad queue or action-plan redesign
+
+## Why this was the right next packet
+
+Before this slice:
+
+- replay-derived recognition pressure could already emit a `recognition_clarity` coach signal
+- ambiguous episodes and correction-heavy replay already propagated into recognition-review preference
+- but low-confidence-only replay cases could still diverge:
+  - the signal lane treated them as recognition pressure
+  - the brief/follow-up lane did not consistently preserve that recognition-review preference
+
+So the next honest move was:
+
+- validate and tighten the replay-to-coach handoff so one recognition-pressure rule drives both signal generation and coach-review preference
+
+## What landed
+
+Primary code changes:
+
+- `UnrealHyperTwist/Source/UnrealHyperTwist/Private/HyperTwistTraining/HyperTwistTrainingCoachLibrary.cpp`
+  - added one shared recognition-review preference predicate
+  - recognition clarity signal derivation now uses that shared predicate
+  - coach brief derivation now uses the same shared predicate
+
+## Product effect
+
+Recognition replay pressure is now more coherent:
+
+- low-confidence recognition replay cases no longer stop at the signal layer
+- the coach brief, action-plan carry-forward, follow-up derivation, and dashboard inspection state now stay aligned with the same recognition-review preference rule
+- replay-to-coach validation is stronger without widening the runtime transport lane
+
+## Acceptance criteria
+
+- low-confidence recognition replay cases now set recognition-review preference even when ambiguity and correction counts stay at zero
+- recognition clarity signal and coach brief use the same predicate
+- the packet stays bounded to coach-library replay/brief continuity
+- full product build succeeds
+
+## Validation checklist
+
+1. build `UnrealHyperTwist.sln` / `UnrealHyperTwistEditor`
+2. confirm recognition clarity signal derivation still compiles and retains existing ambiguity/correction behavior
+3. confirm coach brief derivation now preserves recognition-review preference for low-confidence replay cases too
+4. confirm no recognition transport or dashboard-only packet is reopened
+
+That is the packet.