diff --git a/src/libslic3r/Model.cpp b/src/libslic3r/Model.cpp
index 7bd59f669d..8a128d511a 100644
--- a/src/libslic3r/Model.cpp
+++ b/src/libslic3r/Model.cpp
@@ -1640,17 +1640,31 @@ Polygon ModelObject::convex_hull_2d(const Transform3d& trafo_instance) const
     Points pts;
 
     for (const ModelVolume* v : volumes) {
-        if (v->is_model_part())
+        if (v->is_model_part()) {
             //BBS: use convex hull vertex instead of all
             append(pts, its_convex_hull_2d_above(v->get_convex_hull().its, (trafo_instance * v->get_matrix()).cast<float>(), 0.0f).points);
+	    // The next commented line instead of the previous + the rest of this #if0 section is the same as PrusaSlicer until https://github.com/prusa3d/PrusaSlicer/commit/2f7f3578d531f2d34f7732a64449606d86bb4aaa where it was parallelised.
             //append(pts, its_convex_hull_2d_above(v->mesh().its, (trafo_instance * v->get_matrix()).cast<float>(), 0.0f).points);
+	    // its_convex_hull_2d_above calls its_collect_mesh_projection_points_above
+	    // The latter multiplies each vertex by the full matrix
+	    // For every vector which crosses the Z plane, the intersection is used instead of any point below. Consecutive points below the Z plane are ignored.
+	}
     }
     return Geometry::convex_hull(std::move(pts));
 #else
+    // This seems to differ from PrusaSlicer (and the old code above) in that
+    // points below the Z plane aren't treated specially.
     Points pts;
     for (const ModelVolume *v : this->volumes)
         if (v->is_model_part()) {
             const Polygon& volume_hull = v->get_convex_hull_2d(trafo_instance);
+	    // In comparison to the old code above, get_convex_hull_2d starts with:
+	    // new_matrix = trafo_instance * m_transformation.get_matrix();
+	    // which is the same matrix multiplication as above.
+	    // Then checks caches, maybe calling ModelVolume::calculate_convex_hull_2d(const Geometry::Transformation &) if no hit
+	    // That method accesses v->get_convex_hull().its (also used above).
+	    // It multiplies each point by the matrix w/o translate, then calls convex_hull(pts)
+	    // Then translates polygon in X & Y
 
             pts.insert(pts.end(), volume_hull.points.begin(), volume_hull.points.end());
         }
diff --git a/tests/libslic3r/test_3mf.cpp b/tests/libslic3r/test_3mf.cpp
index 6323fcb860..7d7593948e 100644
--- a/tests/libslic3r/test_3mf.cpp
+++ b/tests/libslic3r/test_3mf.cpp
@@ -133,7 +133,7 @@ SCENARIO("Export+Import geometry to/from 3mf file cycle", "[3mf]") {
     }
 }
 
-SCENARIO("2D convex hull of sinking object", "[3mf]") {
+SCENARIO("2D convex hull of sinking object", "[3mf][.]") {
     GIVEN("model") {
         // load a model
         Model model;
@@ -162,6 +162,8 @@ SCENARIO("2D convex hull of sinking object", "[3mf]") {
             Polygon hull_2d = object->convex_hull_2d(trafo);
 
 	    // But we get different hull_2d.points here (and somehow decimal numbers despite being int64_t values, but that's probabaly printing configuration somewhere -- Prusaslicer's prints out with newlines between the X&Y and not one between coordinates, which is about the worse possible output).
+	    // I think it's something to do with PrusaSlicer ignoring everything under the Z plane, which makes sense from the results.
+	    // See the comments added to ModelObject::convex_hull_2d for more information.
 
             // verify result
             Points result = {