Make display confirmation test deterministic

tests/quickshell/display-transaction-contract.sh

@@ -209,6 +209,8 @@ stale_json="$(jq '.[1].y = 250' "$monitor_state")"
 run ipc --pid "$harness_pid" call displays-test injectReadback "$stale_json" "$((generation - 1))" >/dev/null
 jq -e '.canConfirm == false' <<<"$(transaction_status)" >/dev/null \
     || fail 'stale readback confirmed a newer operation'
+[[ "$(run ipc --pid "$harness_pid" call displays-test confirmChange)" == "false" ]] \
+    || fail 'Keep accepted a display change without a generation-matched readback'
 rm -f "$fixture/no-apply"
 run ipc --pid "$harness_pid" call displays-test expireApplyVerification >/dev/null
 wait_for '.busy == false and .awaiting == false' >/dev/null

tests/quickshell/displays-contract.sh

@@ -335,9 +335,14 @@ done
 [[ "$(status | jq -r .overridden)" == "false" ]] || fail 'an unconfirmed change was written to the settings store'
 
 # ── A confirmed change is what writes ────────────────────────────────────────
-run ipc call displays-test applyScale "$target_scale" >/dev/null
-[[ "$(run ipc call displays-test confirmChange)" == "false" ]] \
-    || fail 'Keep accepted a display change before compositor readback'
+[[ "$(run ipc call displays-test applyScale "$target_scale")" == "true" ]] \
+    || fail 'the confirmed-change fixture could not apply'
+
+# Hyprland can apply and read back a change faster than two IPC round trips, so
+# this live test cannot reliably observe the pre-readback state. The
+# fake-compositor contract deterministically holds that boundary open and
+# proves Keep refuses it; this path proves a real readback eventually enables
+# and persists Keep on the physical display.
 verified=false
 for _ in $(seq 1 30); do
     [[ "$(status | jq -r .canConfirm)" == "true" ]] && { verified=true; break; }