diff --git a/CHANGELOG.md b/CHANGELOG.md
index 5eb5eb4..117b707 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -11,6 +11,17 @@ notes, so you can check for a newer version — and read what changed — withou
 
 ## [Unreleased]
 
+### Fixed
+
+- `aua record stop` no longer fails a recording whose screen never changed at all. A wholly
+  static window makes `screenrecord` emit a single frame, so its media length is exactly 0.0s,
+  and the aggregate "captured nothing at all" guard fired underneath the per-segment rule that
+  already excuses stillness: an 8.02s idle recording exited 3 with `recording_coverage_failed`
+  while holding a valid 37 KB MP4. A partly idle window passed and a wholly idle one did not.
+  Zero media now fails only where no `static_screen_no_frames` gap accounts for it. No segments,
+  a missing supervisor completion, an unreadable or unfinished segment, a non-zero recorder exit,
+  and stretches where the encoder was not running all still fail.
+
 ## [0.27.5] - 2026-09-16
 
 ### Fixed
diff --git a/docs/recording.md b/docs/recording.md
index 4747553..57b254f 100644
--- a/docs/recording.md
+++ b/docs/recording.md
@@ -33,9 +33,13 @@ positive check that the directory is absent. Unverified ownership retains the pe
 Inspect `segments`, `gaps`, `finish`, and `duration_check`. Lifecycle timestamps use target
 boot uptime, so changing the device wall clock cannot shorten a measured pause. The native
 movie duration is compared with the encoder process window and the requested recording
-window, with a two-second tolerance. A duration shortfall or missing supervisor completion
-returns `ok=false` after evidence collection. After encoder failure or incomplete supervisor
-lifecycle, stop exports all finalized segments in order into the requested playable MP4.
+window, with a two-second tolerance. Missing supervisor completion, an unreadable or unfinished
+segment, and stretches where the encoder was not running return `ok=false` after evidence
+collection. A shortfall while the encoder was running is reported as a `static_screen_no_frames`
+gap and `coverage_shortfall_s` instead: `screenrecord` emits a frame when the screen changes, so
+a window whose screen never changed is 0.0 seconds of media rather than lost footage. After
+encoder failure or incomplete supervisor lifecycle, stop exports all finalized segments in
+order into the requested playable MP4.
 Every lifecycle segment has a `status` (`finalized`, `missing`, or `corrupt`) and `exported`
 flag. Corrupt/partial files are copied unchanged into the segment directory. Missing footage
 is explicitly unexported; absence requires a successful inspection with a known missing-file
diff --git a/src/android_ui_analyser/platforms/android_recording.py b/src/android_ui_analyser/platforms/android_recording.py
index 5395f93..a8d53f0 100644
--- a/src/android_ui_analyser/platforms/android_recording.py
+++ b/src/android_ui_analyser/platforms/android_recording.py
@@ -544,13 +544,22 @@ def event_time(raw: str) -> float:
     if stop_uptime_s - previous > 0:
         gaps.append({"start_uptime_s": previous, "end_uptime_s": stop_uptime_s, "reason": "recording_ended_before_stop"})
     requested = max(0.0, stop_uptime_s - first)
-    # Two coverage results a static screen cannot explain, and only these still fail: capturing
-    # nothing at all, and stretches where the encoder was not running. Short media while the
+    # Only stretches where the encoder was not running still fail here. Short media while the
     # encoder WAS running is the static-screen case and is reported instead.
     not_running = {"encoder_startup", "segment_rotation", "recording_ended_before_stop"}
     dark = [gap for gap in gaps if gap["reason"] in not_running
             and (gap["end_uptime_s"] or 0) - (gap["start_uptime_s"] or 0) > _COVERAGE_TOLERANCE_S]
-    failed = failed or bool(dark) or (requested > _COVERAGE_TOLERANCE_S and media_total <= 0)
+    # "Captured nothing at all" is not a separate result from stillness. A window whose screen
+    # never changed once encodes a single frame, so its media length is exactly 0.0: a 2026-09-17
+    # idle run wrote a valid 37,320-byte MP4 with nb_frames=1 over 8.02s and was failed by this
+    # guard -- the very recording the per-segment rule above was written to keep. A partly idle
+    # window was tolerated and a wholly idle one was not. So zero media only fails where no
+    # static_screen_no_frames gap already accounts for the stillness. A recorder that really
+    # captured nothing is caught before this line: no segments, no finish event, an encoder
+    # failure, a missing or unreadable segment file, or a non-zero recorder exit.
+    stillness = any(gap["reason"] == "static_screen_no_frames" for gap in gaps)
+    captured_nothing = requested > _COVERAGE_TOLERANCE_S and media_total <= 0 and not stillness
+    failed = failed or bool(dark) or captured_nothing
     return {
         "mode": "native_segments", "state": "finalized", "segments": list(segments.values()),
         "requested_duration_s": requested, "media_duration_s": media_total,
diff --git a/tests/test_recording_coverage_is_judged_in_one_direction.py b/tests/test_recording_coverage_is_judged_in_one_direction.py
index b68e3c5..3e99d81 100644
--- a/tests/test_recording_coverage_is_judged_in_one_direction.py
+++ b/tests/test_recording_coverage_is_judged_in_one_direction.py
@@ -77,6 +77,21 @@ def test_a_shortfall_is_never_negative() -> None:
 # idle at home, so the check was guaranteed to fail exactly the runs it was meant to protect.
 #
 # Only a recording that captured nothing can still fail on coverage.
+#
+# --- 2026-09-17, third correction --------------------------------------------------------
+#
+# "Captured nothing" and "nothing happened" turned out to be the same number. A screen that
+# never changes at all makes `screenrecord` emit ONE frame, and one frame is 0.0 seconds of
+# media. An 8.02s recording of an idle screen wrote a valid 37,320-byte MP4 (`nb_frames=1`)
+# and still exited 3 with `recording_coverage_failed`, because the aggregate "media_total <= 0"
+# guard fired underneath the per-segment rule that had just excused the same stillness.
+#
+# So a partly idle window passed and a wholly idle one failed -- exactly inverted. Zero media
+# now fails only where no `static_screen_no_frames` gap accounts for it. Everything a dead
+# recorder actually produces (no segments, no finish event, an unreadable file, a non-zero
+# exit, an encoder that was not running) is judged before that point and still fails.
+
+IDLE_RUN = {"wall_s": 8.33, "media_s": 0.0}
 
 A1_RUN = {"wall_s": 168.17, "media_s": 130.001, "frames": 126}
 
@@ -103,10 +118,66 @@ def test_the_shortfall_is_still_reported_rather_than_hidden(monkeypatch):
     assert "static_screen_no_frames" in reasons
 
 
-def test_capturing_nothing_at_all_still_fails(monkeypatch):
-    # The one coverage result a screen that did not change cannot explain.
-    report = _timeline(media_s=0.0, monkeypatch=monkeypatch)
+def test_a_wholly_static_window_is_zero_media_and_still_passes(monkeypatch):
+    """The reported bug: one frame over the whole window is 0.0s of media, not lost evidence."""
+    report = _timeline(media_s=IDLE_RUN["media_s"], wall_s=IDLE_RUN["wall_s"], monkeypatch=monkeypatch)
+
+    assert report["media_duration_s"] == 0.0
+    assert report["duration_check"] == "passed"
+
+
+def test_a_wholly_static_window_still_reports_its_shortfall(monkeypatch):
+    """Passing is not hiding: the whole window is still named as stillness, not as coverage."""
+    report = _timeline(media_s=IDLE_RUN["media_s"], wall_s=IDLE_RUN["wall_s"], monkeypatch=monkeypatch)
+
+    assert report["coverage_shortfall_s"] == pytest.approx(IDLE_RUN["wall_s"], abs=0.05)
+    assert [gap["reason"] for gap in report["gaps"]] == ["static_screen_no_frames"]
+    assert report["encoder_idle_gaps"] == []
+
+
+def test_a_static_window_that_rotated_segments_also_passes(monkeypatch):
+    """Rotation is why the excuse is keyed to stillness found, not to stillness alone.
+
+    A window longer than the native segment limit rotates, and rotation adds its own gap. If a
+    zero-media recording were only excused when *every* gap is stillness, the same idle screen
+    would pass at 8 seconds and fail at 400.
+    """
+    from android_ui_analyser.platforms import android_recording as rec
+
+    monkeypatch.setattr(rec, "media_duration", lambda path: 0.0)
+    events = ("begin 0 100\nend 0 280 0\nbegin 1 280.4\nend 1 460.4 0\n"
+              "begin 2 460.9\nend 2 500.0 0\nfinish 500.0 stopped\n")
+    report = rec.timeline(events, [Path(f"segment-{i}.mp4") for i in range(3)],
+                          stop_uptime_s=500.0, start_uptime_s=100.0)
+
+    assert report["media_duration_s"] == 0.0
+    assert "segment_rotation" in {gap["reason"] for gap in report["gaps"]}
+    assert report["duration_check"] == "passed"
+
+
+def test_capturing_nothing_that_stillness_cannot_explain_still_fails(monkeypatch):
+    """Zero media across segments too short to raise a stillness gap keeps failing."""
+    from android_ui_analyser.platforms import android_recording as rec
+
+    monkeypatch.setattr(rec, "media_duration", lambda path: 0.0)
+    events = "begin 0 100\nend 0 101.2 0\nbegin 1 101.4\nend 1 102.6 0\nfinish 102.6 stopped\n"
+    report = rec.timeline(events, [Path("segment-0.mp4"), Path("segment-1.mp4")],
+                          stop_uptime_s=102.6, start_uptime_s=100.0)
+
+    assert not any(gap["reason"] == "static_screen_no_frames" for gap in report["gaps"])
+    assert report["duration_check"] == "failed"
+
+
+def test_a_recorder_that_produced_no_segment_at_all_still_fails():
+    """The dead recorder: the window was requested, nothing ever began, nothing can excuse it."""
+    from android_ui_analyser.platforms import android_recording as rec
+
+    report = rec.timeline("", [], stop_uptime_s=1000.0 + A1_RUN["wall_s"], start_uptime_s=1000.0)
+
+    assert report["segments"] == []
+    assert report["media_duration_s"] == 0.0
     assert report["duration_check"] == "failed"
+    assert report["encoder_idle_gaps"][0]["reason"] == "encoder_startup"
 
 
 def test_a_recorder_that_died_still_fails(monkeypatch):