diff --git a/.github/workflows/project-build-test.yaml b/.github/workflows/project-build-test.yaml
index 55116aa45..55fc7ebe6 100644
--- a/.github/workflows/project-build-test.yaml
+++ b/.github/workflows/project-build-test.yaml
@@ -122,9 +122,18 @@ jobs:
       if: ${{ github.event_name == 'push' || startsWith(github.ref, 'refs/tags/') }}
       env:
         tag: ${{ env.REGISTRY }}/${{ github.repository }}/${{ matrix.project }}:${{ github.ref_name }}
+        # online's environment is conda-based, and its libraries need to
+        # come ahead of the system's. docker import drops the def's
+        # %environment, and apptainer ignores a docker LD_LIBRARY_PATH
+        # but reads PREPEND_LD_LIBRARY_PATH and prepends it.
+        ld_prepend: ${{ matrix.project == 'online' && '/opt/env/lib' || '' }}
       run: |
         export TAG_LC=${tag,,}
-        cat app.tar.gz | docker import --change "ENV PATH=${{ env.PATH }}" - $TAG_LC
+        changes=(--change "ENV PATH=${{ env.PATH }}")
+        if [ -n "$ld_prepend" ]; then
+          changes+=(--change "ENV PREPEND_LD_LIBRARY_PATH=$ld_prepend")
+        fi
+        cat app.tar.gz | docker import "${changes[@]}" - $TAG_LC
         docker push $TAG_LC
 
 
diff --git a/projects/online/online/monitor/pages/summary.py b/projects/online/online/monitor/pages/summary.py
index 87e607616..f026fa31e 100644
--- a/projects/online/online/monitor/pages/summary.py
+++ b/projects/online/online/monitor/pages/summary.py
@@ -16,6 +16,7 @@
     compute_duty_cycle,
     current_status,
     downtime_breakdown,
+    find_overlaps,
 )
 from online.utils.timing import gps_now
 
@@ -68,6 +69,7 @@ def __init__(
         self.html_file = self.out_dir / "summary.html"
         self.segments = None
         self.stats = None
+        self.overlaps = None
 
     @property
     def plot_name_dict(self) -> dict:
@@ -81,6 +83,22 @@ def plot_name_dict(self) -> dict:
             "duty_cycle_trend": "Hourly search duty cycle",
         }
 
+    def overlap_html(self) -> str:
+        """Writes a warning if any segment overlaps another segment"""
+        if self.overlaps.empty:
+            return ""
+
+        date_format = "%Y-%m-%d %H:%M:%S"
+        first = tconvert(self.overlaps["start"].iloc[0]).strftime(date_format)
+        duration = format_duration(self.overlaps["overlap"].sum())
+        return f"""
+            <p class="red" style="max-width: 820px; margin: 0 auto;">
+            The segment record overlaps itself, counting {duration} of
+            time more than once, first at {first} UTC. The duty cycle and
+            uptime below may be wrong.
+            </p>
+        """
+
     def duty_cycle_html(self) -> str:
         """
         Duty cycle stats and a table of the downtime we were
@@ -106,7 +124,8 @@ def duty_cycle_html(self) -> str:
             ]
             for window, stats in windows.items()
         ]
-        html = self.html_table(
+        html = self.overlap_html()
+        html += self.html_table(
             [
                 "Window",
                 "Duty cycle",
@@ -238,5 +257,6 @@ def create(self, segments: pd.DataFrame) -> None:
         """
         self.segments = segments
         self.stats = compute_duty_cycle(segments)
+        self.overlaps = find_overlaps(segments)
         self.update_summary_plots()
         self.write_html()
diff --git a/projects/online/online/monitor/utils/plotting.py b/projects/online/online/monitor/utils/plotting.py
index 92ce95d51..f4d94480e 100644
--- a/projects/online/online/monitor/utils/plotting.py
+++ b/projects/online/online/monitor/utils/plotting.py
@@ -148,9 +148,8 @@ def latency_plot(plotsdir: Path, df: pd.DataFrame) -> None:
     latency = df["aframe latency"].dropna()
     median = latency.median()
     ninetieth_percentile = latency.quantile(0.9)
-    bins = np.logspace(
-        np.log10(latency.min()), np.log10(latency.max()), num=30
-    )
+    # Pad the range so that the fastest and slowest events are inside the bins
+    bins = np.geomspace(0.95 * latency.min(), 1.05 * latency.max(), num=30)
     plt.hist(latency, bins=bins, alpha=0.7)
     plt.axvline(
         median, color="red", linestyle="--", label=f"Median: {median:.2f} s"
diff --git a/projects/online/online/monitor/utils/segments.py b/projects/online/online/monitor/utils/segments.py
index 186c763db..e1ef33027 100644
--- a/projects/online/online/monitor/utils/segments.py
+++ b/projects/online/online/monitor/utils/segments.py
@@ -21,6 +21,9 @@
 # How long the heartbeat can go unwritten before calling the search dead.
 STALE_SECONDS = 30.0
 
+# Times are written to 5 decimal places, so ignore overlap below this
+OVERLAP_TOLERANCE = 1e-3
+
 
 def segment_dir(run_dir: Path) -> Path:
     return run_dir / "output" / "segments"
@@ -111,6 +114,26 @@ def load_segments(
     return df.reset_index(drop=True)
 
 
+def find_overlaps(df: pd.DataFrame) -> pd.DataFrame:
+    """
+    The segments that start before earlier ones have ended, with how
+    long each overlaps for in an `overlap` column.
+    """
+    df = df.sort_values("start")
+
+    overlapping = []
+    latest_stop = float("-inf")
+    for row in df.to_dict("records"):
+        if row["start"] < latest_stop:
+            # this segment starts before an earlier one has ended, and
+            # overlaps from its start until it or the earlier one ends
+            overlap = min(row["stop"], latest_stop) - row["start"]
+            if overlap > OVERLAP_TOLERANCE:
+                overlapping.append({**row, "overlap": overlap})
+        latest_stop = max(latest_stop, row["stop"])
+    return pd.DataFrame(overlapping, columns=[*df.columns, "overlap"])
+
+
 def _fill_gaps(df: pd.DataFrame) -> pd.DataFrame:
     rows = df.to_dict("records")
     filled = []
diff --git a/projects/online/online/utils/gdb.py b/projects/online/online/utils/gdb.py
index c0427e6ce..8c1177002 100644
--- a/projects/online/online/utils/gdb.py
+++ b/projects/online/online/utils/gdb.py
@@ -112,6 +112,7 @@ def submit(self, event: Event):
         event_dir = self.write_dir / event.event_dir
         filename = event_dir / event.filename
         self.logger.info("Creating event in GraceDB")
+        submission_start = gps_now()
         response = self.create_event(
             group="CBC",
             pipeline="aframe",
@@ -136,11 +137,12 @@ def submit(self, event: Event):
         with open(filename, "w") as f:
             f.write(url)
 
+        submission_end = gps_now()
+
         # record latencies for this event;
         # TODO: determine underlying issue here
         # Handle issue where sometimes the pipeline lags,
         # and the frame file has already left the buffer
-        submission_time = gps_now()
         try:
             t_write = event.get_frame_write_time()
         except FileNotFoundError:
@@ -157,13 +159,20 @@ def submit(self, event: Event):
             t_write = int(event.gpstime)
 
         # time to submit since event occured and since the file was written
-        total_latency = submission_time - event.gpstime
+        total_latency = submission_start - event.gpstime
         write_latency = t_write - event.gpstime
-        aframe_latency = submission_time - t_write
+        aframe_latency = submission_start - t_write
+        submission_duration = submission_end - submission_start
 
         latency_fname = event_dir / "latency.log"
-        latency = "Total Latency (s),Write Latency (s),Aframe Latency (s)\n"
-        latency += f"{total_latency},{write_latency},{aframe_latency}"
+        latency = (
+            "Total Latency (s),Write Latency (s),Aframe Latency (s),"
+            "Submission Duration (s)\n"
+        )
+        latency += (
+            f"{total_latency},{write_latency},"
+            f"{aframe_latency},{submission_duration}"
+        )
         with open(latency_fname, "w") as f:
             f.write(latency)
 
diff --git a/projects/online/online/utils/segments.py b/projects/online/online/utils/segments.py
index 8049ebccd..113d56d4a 100644
--- a/projects/online/online/utils/segments.py
+++ b/projects/online/online/utils/segments.py
@@ -99,12 +99,23 @@ def __init__(
     def _flush_current_file(self) -> None:
         with open(self.current_file, "r") as f:
             current_data = json.load(f)
-        self._append(
-            current_data["state"],
-            current_data["start"],
-            current_data["stop"],
-            current_data["ifos_ready"],
+
+        # The last process may have been interrupted on its way out,
+        # after recording this segment but before removing the file
+        with open(self.segments_file, "r") as f:
+            last_state, last_start = f.read().splitlines()[-1].split(",")[:2]
+        recorded = (
+            last_state == current_data["state"]
+            and last_start == f"{current_data['start']:.5f}"
         )
+
+        if not recorded:
+            self._append(
+                current_data["state"],
+                current_data["start"],
+                current_data["stop"],
+                current_data["ifos_ready"],
+            )
         self.current_file.unlink()
 
     def _append(