diff --git a/packages/process-daemon/README.md b/packages/process-daemon/README.md
index d8693b9bba61..b26a1fca9e47 100644
--- a/packages/process-daemon/README.md
+++ b/packages/process-daemon/README.md
@@ -68,7 +68,7 @@ that wakes partial requests and blocked subscription writes before joining.
 On Unix the daemon loop sleeps in `poll(2)` on the listener and a private wake
 socket pair until a connection arrives, a stop or ownership change wakes it, an
 ownership deadline is due, or a two-second safety net elapses. Windows still
-polls every 10 ms.
+polls every 25 ms.
 
 The Windows named-pipe backend keeps protocol 7 and the registration fields
 `instance_id`, `pid`, `protocol`, `socket`, and `token`. Treat `socket` as an opaque
@@ -332,7 +332,7 @@ coverage on their own.
 On Windows, normal root-child exit hands the ConPTY master to the existing
 child-wait worker for closing. The actor and reader continue draining until the
 real output-pipe EOF; only then is the final exit event published. The worker
-checks the root's wait handle at 10 ms intervals. After the master is handed
+checks the root's wait handle at 25 ms intervals. After the master is handed
 off, Windows input and resize requests fail with a child-exited error; final
 snapshots, rows, and replay remain readable until the terminal is removed.
 Unix post-exit PTY operations retain their existing behavior.
diff --git a/packages/process-daemon/src/service/windows.rs b/packages/process-daemon/src/service/windows.rs
index 81c8cd426514..3c69457ac457 100644
--- a/packages/process-daemon/src/service/windows.rs
+++ b/packages/process-daemon/src/service/windows.rs
@@ -9,10 +9,12 @@ use windows_sys::Win32::{
 
 use super::ActorMessage;
 
+const EXIT_POLL_INTERVAL: Duration = Duration::from_millis(25);
+
 /// Keep blocking ClosePseudoConsole off the actor/reader without introducing
 /// another worker. The actor hands its unique master here on root exit or on
 /// teardown; closing it then produces the real output-pipe EOF.
-/// Root-exit detection uses a 10 ms poll interval. After reaping, the
+/// Root-exit detection uses a 25 ms poll interval. After reaping, the
 /// worker blocks on the close channel rather than continuing to poll.
 pub(super) fn wait_and_close(
     mut child: Box<dyn Child + Send + Sync>,
@@ -37,7 +39,7 @@ pub(super) fn wait_and_close(
         let request = if reported {
             close.recv().map_err(|_| RecvTimeoutError::Disconnected)
         } else {
-            close.recv_timeout(Duration::from_millis(10))
+            close.recv_timeout(EXIT_POLL_INTERVAL)
         };
         match request {
             Ok(master) => {
diff --git a/packages/process-daemon/src/transport/mod.rs b/packages/process-daemon/src/transport/mod.rs
index a7ff72f22b5f..614c3dc38371 100644
--- a/packages/process-daemon/src/transport/mod.rs
+++ b/packages/process-daemon/src/transport/mod.rs
@@ -4,7 +4,7 @@
 //! frame before the peer has had a chance to read it.
 //!
 //! The daemon blocks in `Listener::wait` (Unix: poll(2) on the listener and a
-//! `Wake` socket pair; Windows: still a 10 ms poll) and then calls `accept` until
+//! `Wake` socket pair; Windows: still a 25 ms poll) and then calls `accept` until
 //! it returns `None`. Connection reads/writes run only on its handler threads. A
 //! separate `Cancellation` can wake both directions without waiting for either
 //! handler or peer. `finish_response` and a disconnect monitor's `finish` are
diff --git a/packages/process-daemon/src/transport/windows.rs b/packages/process-daemon/src/transport/windows.rs
index 1919ea15a91a..edfdc3d92df6 100644
--- a/packages/process-daemon/src/transport/windows.rs
+++ b/packages/process-daemon/src/transport/windows.rs
@@ -38,6 +38,8 @@ use security::PrivateSecurity;
 const BUFFER_BYTES: u32 = 64 * 1024;
 const CONNECT_TIMEOUT: Duration = Duration::from_secs(5);
 const COMPLETION_TIMEOUT: Duration = Duration::from_secs(2);
+/// Interim idle poll until `wait` blocks on the pending connect and a stop event.
+const POLL_INTERVAL: Duration = Duration::from_millis(25);
 
 pub(crate) struct Listener {
     endpoint: Vec<u16>,
@@ -83,9 +85,9 @@ impl Listener {
         self.pending.operation.pipe.cancel();
     }
 
-    /// Windows keeps the 10 ms poll for now; the caller then polls `accept`.
+    /// Windows keeps a 25 ms poll for now; the caller then polls `accept`.
     pub fn wait(&self, _wake: &Wake, listen: bool, timeout: Duration) -> io::Result<Readiness> {
-        thread::sleep(timeout.min(Duration::from_millis(10)));
+        thread::sleep(timeout.min(POLL_INTERVAL));
         // `timed_out` stays false: Windows cannot tell, and the missed-wake detector must stay off.
         Ok(Readiness {
             connection: listen,
@@ -102,7 +104,7 @@ pub(crate) struct Readiness {
     pub timed_out: bool,
 }
 
-/// Windows keeps the 10 ms poll for now, so notify is a no-op.
+/// Windows keeps a 25 ms poll for now, so notify is a no-op.
 pub(crate) struct Wake;
 
 impl Wake {