diff --git a/quest/m0/README.md b/quest/m0/README.md
index 007b836..97e79a7 100644
--- a/quest/m0/README.md
+++ b/quest/m0/README.md
@@ -107,7 +107,6 @@ repository.
 - [moq-video's browser backends are pinned](/quest/m0/pin-browser-video.md) - the web studio's video slice and browser composition
 - [moq-audio's browser backends are pinned](/quest/m0/pin-browser-audio.md) - the web studio's audio slice
 - [moq-audio's source tap is pinned](/quest/m0/pin-audio-tap.md) - the samples clap-to-measure sync times a clap from
-- [A replayed media file never repeats a timestamp](/quest/m0/media-replay-timestamps.md) - a media source that starts over publishes strictly increasing timestamps, and its flaky test passes every run
 
 ## Related
 
diff --git a/quest/m0/media-replay-timestamps.md b/quest/m0/media-replay-timestamps.md
deleted file mode 100644
index ec72a1b..0000000
--- a/quest/m0/media-replay-timestamps.md
+++ /dev/null
@@ -1,25 +0,0 @@
-# [S] A replayed media file never repeats a timestamp
-
-## Goal
-
-A media source that starts over while on program publishes its pictures with
-strictly increasing timestamps every time, and
-`publish::tests::a_media_file_is_published_on_one_timeline`
-(`rs/moq-scene/src/publish.rs`) passes on every run.
-
-## Plan
-
-Planned 2026-11-07 (maintainer, /quest-complete): ✅ m0 (rejected: m1), since
-the test is in `just test`, so it flakes the merge gate for every PR, and the
-repeat may reach viewers. ✅ A quest of its own (rejected: fixing it in the moq
-pin bump, [moq-dev/studio#104](https://github.com/moq-dev/studio/pull/104)).
-
-The test fails about half its runs with two equal timestamps in a row: 4 of 8
-on `main` at d42fd70, 3 of 6 after the bump to moq 079a646. It came with
-[moq-dev/studio#44](https://github.com/moq-dev/studio/pull/44); name that PR
-under `## Caused by` if the bug is in its code. The test runs on a paused
-clock, so the variation comes from work outside it, such as the encoder's
-thread, racing the restart (`feed.play()`). Reproduce it first, then find
-whether the published track can repeat a picture's time at a restart (a real
-bug) or only the test reads too early, and fix that root cause rather than
-retry or loosen the assertion.
diff --git a/rs/moq-scene/src/source.rs b/rs/moq-scene/src/source.rs
index 44b8ca1..1c1749f 100644
--- a/rs/moq-scene/src/source.rs
+++ b/rs/moq-scene/src/source.rs
@@ -1701,7 +1701,13 @@ impl Driver {
 		if let (Some(frame), Some((_, config))) = (&picture, &video) {
 			self.open(own, frame, config).await;
 		}
-		let begin = begin.unwrap_or_else(|| kio::clock::now() + BUS_LATENCY);
+		// A cut's play begins after the last picture already published, even
+		// one cut off before it showed, so the track's time never repeats.
+		let begin = begin.unwrap_or_else(|| {
+			let soonest = kio::clock::now() + BUS_LATENCY;
+			own.stamped
+				.map_or(soonest, |stamped| soonest.max(stamped + Duration::from_millis(1)))
+		});
 		self.ran(play, |run| {
 			run.begun.get_or_insert(begin);
 		});
@@ -1796,6 +1802,7 @@ impl Driver {
 	) -> Result<(), Error> {
 		self.open(own, frame, config).await;
 		if let Some(video) = &mut own.track {
+			own.stamped = Some(shows);
 			let stamp = shows.saturating_duration_since(own.origin);
 			frame.timestamp =
 				moq_net::Timestamp::from_micros(stamp.as_micros() as u64).map_err(moq_video::Error::from)?;
@@ -1863,6 +1870,8 @@ struct Own<'a, 'b> {
 	/// Whether the track was opened, or failed to, so it is tried once.
 	opened: bool,
 	origin: Instant,
+	/// When the last picture encoded onto the track shows.
+	stamped: Option<Instant>,
 }
 
 impl<'a, 'b> Own<'a, 'b> {
@@ -1872,6 +1881,7 @@ impl<'a, 'b> Own<'a, 'b> {
 			track: None,
 			opened: false,
 			origin,
+			stamped: None,
 		}
 	}
 }