Normalize line endings with .gitattributes and update IB-Tool references in docs

.gitattributes

@@ -0,0 +1,32 @@
+# Zeilenenden repo-weit normalisieren: git speichert LF, die Arbeitskopie darf
+# plattformueblich sein. Ohne diese Datei meldete git unter Windows den halben
+# Baum als geaendert - ein reiner CRLF/LF-Unterschied ohne Inhaltsaenderung.
+* text=auto
+
+# Dateien, die im Linux-Container ausgefuehrt werden, brauchen LF auch in der
+# Arbeitskopie: CRLF laesst bash mit "\r: command not found" scheitern.
+Dockerfile      text eol=lf
+*.sh            text eol=lf
+*.yml           text eol=lf
+*.yaml          text eol=lf
+
+# Windows-Skripte umgekehrt.
+*.bat           text eol=crlf
+*.ps1           text eol=crlf
+
+# Binaerdateien nie konvertieren. `text=auto` erkennt die meisten davon
+# ohnehin, die Liste macht es fuer die Geodatenformate explizit.
+*.png           binary
+*.jpg           binary
+*.ico           binary
+*.qm            binary
+*.zip           binary
+*.gpkg          binary
+*.shp           binary
+*.shx           binary
+*.dbf           binary
+*.qix           binary
+*.tif           binary
+
+# Nicht als binaer markieren: *.asc (ESRI-ASCII-Grid), *.prj, *.cpg und *.qml
+# sind Textformate und werden von `text=auto` korrekt behandelt.

README.md

@@ -14,7 +14,7 @@ and
 [`docs/quickstart.md`](https://github.com/IB-Tool/IB-Tool-3/blob/master/docs/quickstart.md)
 for how it fits into IB-Tool 3's own workflow.
 
-IBTool needs four more input layers. Its sibling plugin
+IB-Tool 3 needs four more input layers. Its sibling plugin
 **[Data Wizard](https://github.com/IB-Tool/data_wizard)** produces the
 `HU` / `RN` / `Aux` layers from raw ATKIS Basis-DLM data — the building
 footprints it outputs can be fed straight into this plugin as input.

docs/test-strategy.md

@@ -6,7 +6,7 @@ This document is the single authoritative reference for **why** the test suite i
 - A tutorial on pytest syntax — see the pytest documentation.
 - A list of tactical rules for geometry checks or test structure — see [`ai/core/testing-rules.md`](../ai/core/testing-rules.md).
 
-This plugin is a companion to **[IBTool](https://github.com/IB-Tool/IB-Tool-3)** and mirrors its testing conventions at a scale appropriate for a single-module plugin. See [IBTool's own `docs/test-strategy.md`](https://github.com/IB-Tool/IB-Tool-3/blob/master/docs/test-strategy.md) for the full-size version of this document.
+This plugin is a companion to **[IB-Tool 3](https://github.com/IB-Tool/IB-Tool-3)** and mirrors its testing conventions at a scale appropriate for a single-module plugin. See [IB-Tool 3's own `docs/test-strategy.md`](https://github.com/IB-Tool/IB-Tool-3/blob/master/docs/test-strategy.md) for the full-size version of this document.
 
 ---
 
@@ -209,6 +209,8 @@ def test_single_building(self, plugin, tmp_path):
 |---|---|
 | Full `run()` success path untested | Add a unit test that configures the mock dialog with valid paths/`cell_size`, stubs `siedgr()`, and asserts the success `pushMessage` call |
 | No `@pytest.mark.performance` test for `siedgr()` at scale | Add a `performance` + `slow` test with 50+ buildings, asserting it completes within a stated time budget |
+| A missing `qgis:heatmapkerneldensityestimation` skips every `siedgr()` integration test without failing | The collection-time `skipif` in `test_siedgr_integration.py` keeps the suite green even when the algorithm is absent from the Processing build — CI then reports success while `siedgr()` has zero real coverage. In the Docker image the algorithm *must* be present: make its absence a hard error there (assert it is registered when `QGIS_PREFIX_PATH=/usr`) and keep the skip for local runs only |
+| `siedgr()` is a single ~120-line method with 12 `processing.run()` calls | Its steps cannot be tested individually — any failing step surfaces as "the whole pipeline is wrong". Splitting it into named steps (centroids → heatmap → voronoi → cleanup → polygonize) would let the integration tests pinpoint failures. It also still writes to stdout via `print()` instead of the QGIS logger (`IbToolPartion.py`, `run()`) |
 
 ### Priority 2 — Larger effort, lower urgency
 
@@ -262,4 +264,3 @@ pytest test/test_siedgr_integration.py -v
 |------|---------|
 | [`docs/contributing.md`](contributing.md) | CI/CD pipeline, Docker environment, code linting |
 | [`ai/core/testing-rules.md`](../ai/core/testing-rules.md) | Tactical rules: geometry checks, test structure, framework conventions |
-| [`docs/Testplan-data_wizard-ibtoolpartion.md`](Testplan-data_wizard-ibtoolpartion.md) | Implementation plan this document and the current test suite were built from |

test/test_plugin_structure.py

@@ -1,11 +1,10 @@
 # -*- coding: utf-8 -*-
 """Structural tests that don't require QGIS to be available.
 
-Taken over from the former test/test_test.py (dissolved — see
-docs/Testplan-data_wizard-ibtoolpartion.md Phase 3.3). The other tests that
-used to live there duplicated coverage found elsewhere or tested the Python
-standard library / MagicMock itself and carried no assertion value, so they
-were dropped rather than moved.
+Taken over from the former test/test_test.py, which was dissolved. The other
+tests that used to live there duplicated coverage found elsewhere or tested
+the Python standard library / MagicMock itself and carried no assertion
+value, so they were dropped rather than moved.
 """
 import sys
 from pathlib import Path

test/test_siedgr_integration.py

@@ -7,14 +7,13 @@
 test/layer_factories.py and written to tmp_path — no test data checked into
 the repository is needed.
 
-Requires a full QGIS Processing environment (Docker, see
-docs/Testplan-data_wizard-ibtoolpartion.md Phase 3.2). All tests are skipped
+Requires a full QGIS Processing environment (Docker). All tests are skipped
 automatically when QGIS is unavailable, and again when the
 qgis:heatmapkerneldensityestimation algorithm siedgr() depends on is not
-registered in the running QGIS/Processing build (this is checked at
-collection time instead of via the one-off Docker experiment described in
-the plan, so the gap is self-documenting no matter which QGIS image runs
-these tests).
+registered in the running QGIS/Processing build (checked at collection time,
+so the gap is self-documenting no matter which QGIS image runs these tests -
+see docs/test-strategy.md, Gap Analysis, on why a silent skip is not enough
+in CI).
 """
 # pylint: disable=possibly-used-before-assignment
 # QgsWkbTypes/QgsVectorLayer/IbToolPartition/the layer_factories imports are