The full snapshot is still saved (this does not strip added/changed/removed lines), only limits the triggers of the change detection.

changedetectionio/blueprint/ui/templates/edit.html

@@ -328,8 +328,9 @@ Math: {{ 1 + 1 }}") }}
                         {{ render_checkbox_field(form.filter_text_replaced) }}
                         {{ render_checkbox_field(form.filter_text_removed) }}
                     <span class="pure-form-message-inline">Note: Depending on the length and similarity of the text on each line, the algorithm may consider an <strong>addition</strong> instead of <strong>replacement</strong> for example.</span><br>
-                    <span class="pure-form-message-inline">&nbsp;So it's always better to select <strong>Added</strong>+<strong>Replaced</strong> when you're interested in new content.</span><br>
-                    <span class="pure-form-message-inline">&nbsp;When content is merely moved in a list, it will also trigger an <strong>addition</strong>, consider enabling <code><strong>Only trigger when unique lines appear</strong></code></span>
+                    <span class="pure-form-message-inline">So it's always better to select <strong>Added</strong>+<strong>Replaced</strong> when you're interested in new content.</span><br>
+                    <span class="pure-form-message-inline">When content is merely moved in a list, it will also trigger an <strong>addition</strong>, consider enabling <code><strong>Only trigger when unique lines appear</strong></code>.</span><br>
+                    <span class="pure-form-message-inline">The full snapshot is still saved (this does not strip added/changed/removed lines), only limits the triggers of the change detection.</span>
                 </fieldset>
                 <fieldset class="pure-control-group">
                     {{ render_checkbox_field(form.check_unique_lines) }}

changedetectionio/processors/text_json_diff/processor.py

@@ -459,15 +459,27 @@ class perform_site_check(difference_detection_processor):
         # Save text before ignore filters (for diff calculation)
         text_content_before_ignored_filter = stripped_text
 
+        # Save full content before diff filtering for consistent MD5 calculation
+        full_content_for_md5 = None
+
         # === DIFF FILTERING ===
         # If user wants specific diff types (added/removed/replaced only)
         if watch.has_special_diff_filter_options_set() and len(watch.history.keys()):
-            stripped_text = self._apply_diff_filtering(watch, stripped_text, text_content_before_ignored_filter)
-            if stripped_text is None:
-                # No differences found, but content exists
+            # Save full content BEFORE applying diff filtering
+            # This ensures MD5 is always calculated from full content, not the filtered diff
+            full_content_for_md5 = stripped_text
+
+            filtered_diff = self._apply_diff_filtering(watch, stripped_text, text_content_before_ignored_filter)
+            if filtered_diff is None:
+                # No matching differences found (e.g., only removed lines when user wants added/replaced)
+                # Calculate MD5 of full content and return early
                 c = ChecksumCalculator.calculate(text_content_before_ignored_filter, ignore_whitespace=True)
                 return False, {'previous_md5': c}, text_content_before_ignored_filter.encode('utf-8')
 
+            # Has matching changes - use filtered diff for trigger_text evaluation and display,
+            # but full_content_for_md5 will be used later for MD5 calculation
+            stripped_text = filtered_diff
+
         # === EMPTY PAGE CHECK ===
         empty_pages_are_a_change = self.datastore.data['settings']['application'].get('empty_pages_are_a_change', False)
         if not stream_content_type.is_json and not empty_pages_are_a_change and len(stripped_text.strip()) == 0:
@@ -495,17 +507,23 @@ class perform_site_check(difference_detection_processor):
             stripped_text = transformer.sort_alphabetically(stripped_text)
 
         # === CHECKSUM CALCULATION ===
-        text_for_checksuming = stripped_text
+        # When diff filtering is active, use full content for MD5, not the filtered diff
+        # This ensures consistent MD5 calculation regardless of what changes occurred
+        if full_content_for_md5 is not None:
+            text_for_checksuming = full_content_for_md5
+        else:
+            text_for_checksuming = stripped_text
 
         # Apply ignore_text for checksum calculation
         if filter_config.ignore_text:
-            text_for_checksuming = html_tools.strip_ignore_text(stripped_text, filter_config.ignore_text)
+            text_for_checksuming = html_tools.strip_ignore_text(text_for_checksuming, filter_config.ignore_text)
 
             # Optionally remove ignored lines from output
+            # Note: Only apply to stripped_text if we're not using full_content_for_md5
             strip_ignored_lines = watch.get('strip_ignored_lines')
             if strip_ignored_lines is None:
                 strip_ignored_lines = self.datastore.data['settings']['application'].get('strip_ignored_lines')
-            if strip_ignored_lines:
+            if strip_ignored_lines and full_content_for_md5 is None:
                 stripped_text = text_for_checksuming
 
         # Calculate checksum
@@ -571,7 +589,12 @@ class perform_site_check(difference_detection_processor):
         if 'text_for_checksuming' in locals() and text_for_checksuming is not stripped_text:
             del text_for_checksuming
 
-        return changed_detected, update_obj, stripped_text
+        # When diff filtering is active, return full content for history snapshots
+        # stripped_text contains only the filtered diff, which would create confusing/broken snapshots
+        if full_content_for_md5 is not None:
+            return changed_detected, update_obj, full_content_for_md5
+        else:
+            return changed_detected, update_obj, stripped_text
 
     def _apply_diff_filtering(self, watch, stripped_text, text_before_filter):
         """Apply user's diff filtering preferences (show only added/removed/replaced lines)."""

changedetectionio/tests/test_add_replace_remove_filter.py

@@ -183,3 +183,104 @@ def test_check_add_line_contains_trigger(client, live_server, measure_memory_usa
         assert '网站监测 内容更新了'.encode('utf-8') in response
 
     delete_all_watches(client)
+
+
+def test_consistent_md5_with_diff_filtering(client, live_server, measure_memory_usage):
+    """
+    Test that MD5 checksums are calculated consistently when diff filtering is active.
+
+    This test ensures that after a change is detected with diff filtering enabled,
+    subsequent checks with identical content don't trigger false positives.
+
+    Bug: Previously, MD5 was calculated from the filtered diff (partial content)
+    when changes were found, but from full content when no changes were found.
+    This caused false positives on the next check with identical content.
+
+    Fix: Always calculate MD5 from full content, regardless of diff filtering.
+    """
+
+    delete_all_watches(client)
+    time.sleep(1)
+
+    # Setup initial content
+    set_original()
+    test_url = url_for('test_endpoint', _external=True)
+    uuid = client.application.config.get('DATASTORE').add_watch(url=test_url)
+
+    # Configure: Only track ADDED and REPLACED lines, ignore REMOVED lines
+    res = client.post(
+        url_for("ui.ui_edit.edit_page", uuid="first"),
+        data={
+            "url": test_url,
+            'processor': 'text_json_diff',
+            'fetch_backend': "html_requests",
+            'filter_text_added': 'y',       # Track added lines
+            'filter_text_replaced': 'y',    # Track replaced lines
+            'filter_text_removed': '',      # Don't track removed lines
+            "time_between_check_use_default": "y"
+        },
+        follow_redirects=True
+    )
+    assert b"Updated watch." in res.data
+
+    # CHECK 1: Initial baseline
+    client.get(url_for("ui.form_watch_checknow"), follow_redirects=True)
+    wait_for_all_checks(client)
+    res = client.get(url_for("watchlist.index"))
+    assert b'has-unread-changes' not in res.data  # First check, no change
+
+    # Mark as viewed to start fresh
+    client.get(url_for("ui.mark_all_viewed"), follow_redirects=True)
+
+    # CHECK 2: Remove a line (should NOT trigger - removed lines are filtered out)
+    set_original(excluding='Something irrelevant')
+    client.get(url_for("ui.form_watch_checknow"), follow_redirects=True)
+    wait_for_all_checks(client)
+    res = client.get(url_for("watchlist.index"))
+    assert b'has-unread-changes' not in res.data  # No change (removed line filtered)
+
+    # CHECK 3: Add a line (should trigger - added lines are tracked)
+    set_original(excluding='Something irrelevant', add_line='<p>New exciting feature!</p>')
+    client.get(url_for("ui.form_watch_checknow"), follow_redirects=True)
+    wait_for_all_checks(client)
+    res = client.get(url_for("watchlist.index"))
+    assert b'has-unread-changes' in res.data  # Change detected (added line)
+
+    # Mark as viewed
+    client.get(url_for("ui.mark_all_viewed"), follow_redirects=True)
+
+    # CHECK 4: Same content as CHECK 3 (THE CRITICAL TEST - should NOT trigger)
+    # This is where the bug would manifest: false positive change detection
+    # because previous MD5 was from filtered diff, current MD5 is from full content
+    set_original(excluding='Something irrelevant', add_line='<p>New exciting feature!</p>')
+    client.get(url_for("ui.form_watch_checknow"), follow_redirects=True)
+    wait_for_all_checks(client)
+    res = client.get(url_for("watchlist.index"))
+
+    # CRITICAL ASSERTION: Should NOT detect change (content is identical to CHECK 3)
+    assert b'has-unread-changes' not in res.data, \
+        "False positive! Content identical to previous check but change was detected. " \
+        "MD5 calculation is inconsistent with diff filtering."
+
+    # CHECK 5: Verify system still detects real changes (replace a line)
+    # Change "Some initial text" to "Some modified text"
+    modified_content = """<html>
+     <body>
+     <p>Some modified text</p>
+     <p>So let's see what happens.</p>
+     <p>and a new line!</p>
+     <p>The golden line</p>
+     <p>New exciting feature!</p>
+     <p>A BREAK TO MAKE THE TOP LINE STAY AS "REMOVED" OR IT WILL GET COUNTED AS "CHANGED INTO"</p>
+     </body>
+     </html>
+    """
+    with open("test-datastore/endpoint-content.txt", "w") as f:
+        f.write(modified_content)
+
+    client.get(url_for("ui.form_watch_checknow"), follow_redirects=True)
+    wait_for_all_checks(client)
+    res = client.get(url_for("watchlist.index"))
+    assert b'has-unread-changes' in res.data  # Change detected (replaced line)
+
+    delete_all_watches(client)