iOS agent-device press accepts stale ref after navigation (12-stale-validate-back)

[nmel-cs]

Description
While running negative-scenario QA on iOS simulator via agent-device, pressing a saved element ref from a previous screen succeeds after navigation even though that element no longer exists on the current surface. The same flow on Android correctly rejects the stale ref.

This is scenario 12-stale-validate-back from the Agent-Device Negative Scenarios sign-in suite.

What was observed:

• iOS (FAIL): After saving Verify ref @e13 on the magic-code screen, going Back to Home, and pressing @e13 without a fresh snapshot, agent-device reports a successful tap (Tapped @e13 (201, 804)) and exits 0 (STALE_EXIT: 0).
• Android (PASS / expected behavior): Same stale press returns Error (COMMAND_FAILED): Ref @e50 not found or has no bounds and exits 1 (STALE_EXIT: 1).
• Cross-platform parity gap: iOS does not reject a ref that is invalid for the current UI tree after screen change.
• Full iOS report: https://app.notion.com/p/callstack/Agent-Device-Negative-Scenarios-iOS-full-report-3985d027c0f881869cdaf60bb00f6ae0
• Android baseline report: https://app.notion.com/p/callstack/Agent-Device-Negative-Scenarios-full-report-3985d027c0f8812a8551eaa94f2bb580

Expected vs. actual behavior

Expected	After navigation away from the magic-code screen, press @e13 (saved Verify ref) should fail because Verify is not on Home — same as Android: COMMAND_FAILED: Ref @e13 not found or has no bounds, non-zero exit.
Actual	iOS accepts the stale ref and performs a tap at cached coordinates (Tapped @e13 (201, 804)), exit code 0. Automation may continue as if Verify was pressed on Home, producing false-positive tool behavior and hiding stale-ref bugs.

Repro steps
prompt:

Fresh install. Sign in with a new random Gmail until Home. Open **Validate your account** to reach the magic-code screen.

On that screen, note the **Verify** button ref (e.g. `@e13`). Go **Back** to Home.

Without re-checking the screen, press the saved Verify ref again.

Expect fail: Verify does not exist on Home — tool should reject the stale ref.

minimal commands (iOS):

# After reaching magic-code screen and snapshot -i
VERIFY_REF=@e13   # from snapshot: @e13 [button] "Verify"

agent-device press label="Back" --session <session>
agent-device press @e13 --session <session>   # stale press on Home
echo $?   # observed: 0 (should be 1)

key log excerpt (iOS, 2026-07-10):

VERIFY_REF=@e13
Tapped (28, 98)                    # Back
Tapped @e13 (201, 804)              # stale press on Home
STALE_EXIT: 0
RESULT: FAIL

key log excerpt (Android, baseline):

VERIFY_REF=@e50
Error (COMMAND_FAILED): Ref @e50 not found or has no bounds
STALE_EXIT: 1

Environment:

Field	Value
Device	iPhone 16 Pro simulator (UDID 4700A4D2-8FB1-4089-94B0-C2A838E965B1)
OS	iOS 18.0 (simulator)
App	com.expensify.chat.dev — New Expensify Dev standalone
App path	~/App/ios/build/Build/Products/DebugDevelopment-iphonesimulator/New Expensify Dev.app
agent-device	0.18.3
Android comparison device	Google Pixel 7 (physical), Android 16, org.me.mobiexpensifyg.adhoc
Scenario	12-stale-validate-back
Diagnostics (iOS)	mreqf3bf-6dbc7191 (session ios-stale-12-122511)

Severity / impact
Severity: Medium
Impact: Stale-ref presses that should fail are treated as successful on iOS. Agents and automation can tap wrong coordinates on a new screen without an error, undermining negative-test reliability and cross-platform parity. False passes can mask navigation/ref-management bugs in AI-driven mobile testing flows.