From 3fc69b6dbb63822ddeaeec4c3209c19ea1304cc3 Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1s=20Link?= <tomas.link@globalfishingwatch.org>
Date: Fri, 2 Oct 2026 16:09:29 -0300
Subject: [PATCH 1/2] Fix positions_hours_before undercounted for open gaps at
 window overlaps

Boundaries.last_boundary() picked whichever boundary starts latest, not
whichever has the fullest lookback for the vessel's last message. When that
message falls in the overlap of two sliding windows, the later-starting
window's own history is shorter, truncating positions_hours_before(_sat) for
the resulting open gap.
---
 .../raw_gaps/fns/process_boundaries.py        | 24 +++++++++++++++-
 tests/conftest.py                             | 28 +++++++++++++++++++
 2 files changed, 51 insertions(+), 1 deletion(-)

diff --git a/src/pipe_gaps/pipelines/raw_gaps/fns/process_boundaries.py b/src/pipe_gaps/pipelines/raw_gaps/fns/process_boundaries.py
index b94d48f6..e89fc2b8 100644
--- a/src/pipe_gaps/pipelines/raw_gaps/fns/process_boundaries.py
+++ b/src/pipe_gaps/pipelines/raw_gaps/fns/process_boundaries.py
@@ -52,6 +52,28 @@ def first_boundary(self):
     def last_boundary(self):
         return self._boundaries[-1]
 
+    def boundary_for_last_message(self):
+        """Returns the boundary with the fullest lookback for the vessel's last message.
+
+        A message can fall inside two overlapping sliding windows at once, so more than
+        one boundary can share the same last message. The earlier-starting of those
+        windows reaches further back in time, so its boundary's ``end`` list is never
+        shorter -- and is often fuller -- than a later-starting window sharing the same
+        last message, whose own range begins later and leaves that earlier history out
+        entirely.
+
+        Returns the earliest-starting boundary (in ``self._boundaries``' own sort order)
+        whose own last message matches the overall last message across all boundaries --
+        unlike :meth:`last_boundary`, which simply picks whichever boundary starts latest,
+        regardless of whether it has the fullest history for that message.
+        """
+        overall_last_key = timestamp_msgid_key()(self.last_message())
+
+        return next(
+            b for b in self._boundaries
+            if timestamp_msgid_key()(b.last_message()) == overall_last_key
+        )
+
     def first_message(self):
         return self.first_boundary().first_message()
 
@@ -147,7 +169,7 @@ def process(
         # Step three:
         # Create open gap if last message of last group met condition.
         if self._eval_last:
-            last_boundary = boundaries.last_boundary()
+            last_boundary = boundaries.boundary_for_last_message()
             last_message = last_boundary.last_message()
 
             last_message_dt = datetime_from_timestamp(last_message[self.KEY_TIMESTAMP])
diff --git a/tests/conftest.py b/tests/conftest.py
index 8d68f42d..58e9a887 100644
--- a/tests/conftest.py
+++ b/tests/conftest.py
@@ -298,6 +298,34 @@ class TestCases:
             ],
             "id": "one_ssvid_with_open_gaps"
         },
+        {
+            # The vessel's last message (the open gap's OFF message) falls in the 12h
+            # overlap shared by two consecutive sliding windows, both in range here
+            # (date_range spans 3 days so window B's own responsibility day, Jan 3,
+            # isn't filtered out by FilterWindowsByDateRange). The earlier satellite
+            # message, 4 hours before the OFF message, is only inside the
+            # earlier-starting window's own range -- the later-starting window's
+            # range begins after it, so its own boundary never includes it.
+            #
+            # eval_last must use whichever window has the fuller history for the
+            # vessel's last message, not simply whichever window starts latest.
+            "messages": [
+                create_message(time=datetime(2024, 1, 2, 9), receiver_type="satellite"),
+                create_message(time=datetime(2024, 1, 2, 13)),  # open gap (last message).
+            ],
+            "open_gaps": [],
+            "threshold": 6,
+            "date_range": ("2024-01-02", "2024-01-04"),
+            "expected_gaps": [
+                {
+                    "positions_hours_before": 2,
+                    "positions_hours_before_ter": 1,
+                    "positions_hours_before_sat": 1,
+                    "positions_hours_before_dyn": 0
+                },
+            ],
+            "id": "one_ssvid_with_open_gap_at_window_overlap_boundary"
+        },
     ]
 
     GAP_BETWEEN_ARBITRARY_PERIODS = [

From 409934b38722bd261a6a940594d5d59d040862fd Mon Sep 17 00:00:00 2001
From: =?UTF-8?q?Tom=C3=A1s=20Link?= <tomas.link@globalfishingwatch.org>
Date: Fri, 2 Oct 2026 16:36:53 -0300
Subject: [PATCH 2/2] Bump version to 0.14.3

---
 pyproject.toml | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/pyproject.toml b/pyproject.toml
index dfab6fb8..8c9cd654 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -12,7 +12,7 @@ namespaces = false
 
 [project]
 name = "pipe-gaps"
-version = "0.14.2"
+version = "0.14.3"
 description = "Tools for detecting interruptions in vessel position reporting systems (e.g., AIS, VMS)."
 readme = "README.md"
 license = "Apache-2.0"