test(ios): deterministic injection seam for recorded-tap-failure corroboration (#1605 merge gate)

The field failure cannot be reproduced on this head: the tap false-failures
were a downstream symptom of XCTest-channel saturation, which the #1587
capture fixes removed. The seam records a real XCTIssue AFTER the real
gesture inside the per-command failure-count window, so
xctestRecordedFailureResponse and target invalidation fire byte-for-byte
like the field failure. Armed via a decrementing /tmp flag file (the daemon
regenerates tampered xctestrun templates, so env plumbing cannot reach a
daemon-spawned runner); compiled only under AGENT_DEVICE_RUNNER_UNIT_TESTS.

Live evidence on a daemon-spawned runner (Bluesky, ad-bsky-repro sim):
- landed case: injected failure on a real Search-tab tap -> success with
  the corroboration warning, screen verifiably on Search, no redispatch,
  runner serving next commands; flag consumed exactly once.
- unchanged case: injected failure on a dead-coordinate tap -> capture
  unchanged -> XCTEST_RECORDED_FAILURE preserved with the new honest hint;
  runner still usable.
- field-shape race (relaunch -> full snapshot -> immediate press, 5
  attempts): no natural recorded failure occurs on this head — the hostile
  tree needed for channel saturation is gone, corroborating the causal
  story.

Author: thymikee

apple/runner/AgentDeviceRunner/AgentDeviceRunnerUITests/RunnerTests+CommandExecution.swift

@@ -439,6 +439,16 @@ extension RunnerTests {
   }
 #endif
 
+  func testInjectedTapRecordedFailureGateIsTapOnlyAndCountGated() {
+    // The seam's recording side cannot run in-bundle (a real XCTIssue would
+    // fail this very test run — same constraint the record(_:) suppression
+    // tests document); the live daemon proof covers it. This pins the gate.
+    XCTAssertFalse(RunnerTests.shouldInjectTapRecordedFailure(command: .tap, remaining: 0))
+    XCTAssertTrue(RunnerTests.shouldInjectTapRecordedFailure(command: .tap, remaining: 1))
+    XCTAssertFalse(RunnerTests.shouldInjectTapRecordedFailure(command: .type, remaining: 1))
+    XCTAssertFalse(RunnerTests.shouldInjectTapRecordedFailure(command: .snapshot, remaining: 1))
+  }
+
   func testXCTestRecordedFailureResponseFailsMutatingSuccesses() throws {
     let command = try runnerCommandFixture(#"{"command":"tap","commandId":"tap-1"}"#)
     let response = Response(ok: true, data: DataPayload(message: "tapped"))
@@ -1227,6 +1237,20 @@ extension RunnerTests {
           userInfo: [NSLocalizedDescriptionKey: "command returned no response"]
         )
       }
+#if AGENT_DEVICE_RUNNER_UNIT_TESTS
+      // #1605 merge gate: the REAL gesture already executed above; recording a
+      // production-shaped issue here makes the per-command failure-count
+      // conversion below fire exactly as in the field (bsky-24: activation
+      // lands, bookkeeping records a failure). Compiled out of production.
+      if consumeInjectedTapRecordedFailureForTesting(command: command.command) {
+        record(
+          XCTIssue(
+            type: .assertionFailure,
+            compactDescription: "Injected tap recorded-failure (#1605 corroboration merge gate)"
+          )
+        )
+      }
+#endif
       if didRecordXCTestFailure(since: failureCountBefore),
         let failureResponse = xctestRecordedFailureResponse(command: command, response: response)
       {

apple/runner/AgentDeviceRunner/AgentDeviceRunnerUITests/RunnerTests.swift

@@ -120,6 +120,42 @@ final class RunnerTests: XCTestCase {
   // seconds on remote-hosted consent dialogs and bypass the plan budget (#1244).
   let systemModalProbeBudget: TimeInterval = 4
   #if AGENT_DEVICE_RUNNER_UNIT_TESTS
+  // #1605 merge gate: deterministic live reproduction of the field ambiguity —
+  // a tap whose coordinate activation LANDS while XCTest bookkeeping records a
+  // failure. Armed by writing a decrementing count to the flag file below
+  // (the daemon regenerates tampered xctestrun templates, so env plumbing
+  // cannot reach a daemon-spawned runner); consumed one injection per tap.
+  // The injection records a real XCTIssue AFTER the real gesture, so
+  // `xctestRecordedFailureResponse` and target invalidation fire byte-for-byte
+  // like a field failure. Production builds compile none of this.
+  static let injectedTapFailureFlagPathForTesting =
+    "/tmp/agent-device-inject-tap-recorded-failure-for-testing"
+
+  static func shouldInjectTapRecordedFailure(command: CommandType, remaining: Int) -> Bool {
+    command == .tap && remaining > 0
+  }
+
+  func consumeInjectedTapRecordedFailureForTesting(command: CommandType) -> Bool {
+    guard
+      let raw = try? String(
+        contentsOfFile: Self.injectedTapFailureFlagPathForTesting,
+        encoding: .utf8
+      ),
+      let remaining = Int(raw.trimmingCharacters(in: .whitespacesAndNewlines))
+    else {
+      return false
+    }
+    guard Self.shouldInjectTapRecordedFailure(command: command, remaining: remaining) else {
+      return false
+    }
+    try? String(remaining - 1).write(
+      toFile: Self.injectedTapFailureFlagPathForTesting,
+      atomically: true,
+      encoding: .utf8
+    )
+    return true
+  }
+
   // Unit-test-only injectable override for the system-modal probe (see
   // `boundedBlockingSystemAlertSnapshot` in RunnerTests+Snapshot.swift): when set, a test's probe
   // body runs in place of `blockingSystemAlertSnapshot` so it can force a real timeout without a