diff --git a/docs/adr/0013-unified-gesture-plans.md b/docs/adr/0013-unified-gesture-plans.md
index d9ab8c563..fd1d5c458 100644
--- a/docs/adr/0013-unified-gesture-plans.md
+++ b/docs/adr/0013-unified-gesture-plans.md
@@ -114,7 +114,10 @@ Platform adapters consume the canonical plan:
   roughly 16 ms samples using smoothstep `s(t) = 3t² - 2t³`. A linear segment has constant movement
   velocity through its endpoint unless a destination hold follows it; smoothstep has zero slope at
   both endpoints and a peak velocity 1.5 times its average. Identical endpoints and total durations
-  can consequently produce different recognizer and deceleration outcomes.
+  can consequently produce different recognizer and deceleration outcomes. Neither a destination
+  hold nor an analytically zero endpoint slope proves a controlled release by itself: XCTest event
+  sampling and app recognizer thresholds can still leave observable post-lift motion, so live
+  evidence must measure the resulting content offset after pointer-up.
 
   Live iOS characterization in [issue #1586](https://github.com/callstack/agent-device/issues/1586)
   confirmed that distinction: the schedules crossed the same fling-recognizer thresholds in the
@@ -192,8 +195,9 @@ resolution and recording contracts are tested at that composition seam.
   plans remain cadence-bounded because their synchronized geometry is part of their contract.
 - Unit tests cover canonical plan shape, Android lowering, helper/provider payloads, and WebDriver
   action construction. They cannot prove timing or event delivery inside the private XCTest bridge,
-  so iOS timing changes require live simulator evidence that observes the requested content change
-  and records the runner start/end uptime delta alongside the requested duration.
+  so iOS timing changes require live simulator evidence from an independent app-observed
+  postcondition. Runner start/end uptime deltas can supplement that evidence, but request echoes or
+  internal timing alone do not prove that the app recognized the intended gesture.
 
 ## Alternatives Considered
 
diff --git a/docs/agents/testing.md b/docs/agents/testing.md
index 2eb5915f7..e0d562f3a 100644
--- a/docs/agents/testing.md
+++ b/docs/agents/testing.md
@@ -87,6 +87,16 @@ run the smallest owning test, record the failing test count, then restore it. Ap
 test relocation and structural gates: plant a type error or violation and verify the intended gate
 discovers and names it.
 
+A callback-based canary must observe the subject's semantic success state, not merely lifecycle
+completion. For example, React Native Gesture Handler's
+[`onFinalize`](https://docs.swmansion.com/react-native-gesture-handler/docs/fundamentals/callbacks-events/)
+also runs when recognition fails or is interrupted; use an activation-dependent callback or assert
+the callback's success state before publishing a passing result.
+
+A device replay is automatic regression coverage only when an automatic PR or scheduled lane selects
+and executes it. Name the owning lane and confirm the scenario ran on the exact PR head; placing a
+replay in a manual or otherwise unselected tier is test material, not automatic evidence.
+
 For structured classifiers, pair the positive case with the closest negative case. When an error
 message can be identical with and without a typed reason, the negative test must prove the message
 alone cannot activate retry, fallback, or recovery.