Add pytest suite + fix two regex bugs uncovered by it (#46)

Erste Tests für die Codebase. 77 Tests, 0.08s Laufzeit, decken die
drei Bug-Klassen aus der April-2026-Adapter-Session ab plus haben
schon zwei weitere Bugs in Production-Code aufgedeckt.

## Setup

- requirements-dev.txt mit pytest + pytest-asyncio
- pytest.ini mit asyncio_mode=auto
- tests/conftest.py stubbt fitz/bs4/openai/pydantic_settings, damit
  die Suite ohne den vollen prod-requirements-Satz läuft (pure unit
  tests, kein PDF-Parsing, kein HTTP)

## Tests

- tests/test_parlamente.py (33 Tests)
  * PortalaAdapter._parse_hit_list_cards: doctype/doctype_full
    NameError-Regression aus 1cb030a, plus Title/Drucksache/Fraktion-
    /Datum/PDF-Extraktion gegen ein BE-Card-Fixture
  * PortalaAdapter._parse_hit_list_dump: gegen ein LSA-Perl-Dump-
    Fixture inkl. Hex-Escape-Decoding (\x{fc} → ü)
  * PortalaAdapter._parse_hit_list_html: Auto-Detection zwischen
    Card- und Dump-Format
  * PortalaAdapter._normalize_fraktion: kanonische Fraktion-Codes
    inkl. F.D.P.-mit-Punkten, BÜNDNIS 90, DIE LINKE, BSW
  * ParLDokAdapter._hit_to_drucksache: JSON-Hit → Drucksache
    Mapping inkl. /navpanes-Stripping, MdL-mit-Partei-in-Klammern,
    Landesregierung-Detection
  * ParLDokAdapter._fulltext_id: bundle.js-mirroring (deferred,
    aber dokumentiert)
  * ADAPTERS-Registry-Sanity

- tests/test_embeddings.py (11 Tests)
  * _chunk_source_label: Programm-Name + Seite (Halluzinations-
    Bug-Regression aus 1b5fd96)
  * format_quotes_for_prompt: jeder Chunk muss Programm-Name
    enthalten, strict-citation-Hinweis muss im Output sein,
    keine NRW-Halluzinationen für MV/BE-Chunk-Sets

- tests/test_wahlprogramme.py (14 Tests)
  * Registry-Struktur (jahr int, seiten int, .pdf-Endung)
  * File-Existenz: jede registrierte PDF muss in
    static/referenzen/ liegen — würde Tippfehler in den 22
    indexierten Programmen sofort fangen
  * embeddings.PROGRAMME-Konsistenz-Cross-Check

- tests/test_bundeslaender.py (15 Tests)
  * Sanity über 16-State-Registry
  * #48-Klassifikations-Regression: TH=ParlDok, HB=StarWeb,
    SN=Eigensystem
  * Wahltermine plausibel (zwischen 2026 und 2035)

- tests/test_analyzer.py (4 Tests)
  * Markdown-Codeblock-Stripping aus dem JSON-Retry-Loop

## Bug-Funde während der Test-Schreibphase

Zwei Production-Bugs in den _normalize_fraktion-Helfern wurden
durch die neuen Tests sofort aufgedeckt und im selben Commit gefixt:

1. PortalaAdapter._normalize_fraktion matched "F.D.P." (mit Punkten,
   wie historische SH/HB-Drucksachen) nicht — Regex \bFDP\b ist zu
   strikt. Fix: \bF\.?\s*D\.?\s*P\.?\b analog zu ParLDokAdapter.

2. ParLDokAdapter._normalize_fraktion (auch PortalaAdapter) matched
   "Ministerium der Finanzen" nicht als Landesregierung, weil
   \bMINISTER\b die Wortgrenze auch nach MINISTER verlangt — bei
   MINISTERIUM steht aber IUM danach, keine Wortgrenze. Fix:
   \bMINISTER ohne abschließendes \b.

Beide Bugs hätten Fraktion-Felder bei Drucksachen der Bremischen
Bürgerschaft (FDP-Listen) und bei Landesregierungs-Drucksachen
in MV/LSA fälschlich leer gelassen — exakt der "fraktionen=[]"-
Befund aus dem MV-Smoke-Test in #4.

Phase 0 aus Roadmap-Issue #49.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

app/parlamente.py

@@ -443,7 +443,9 @@ class PortalaAdapter(ParlamentAdapter):
             out.append("CDU")
         if has(r"\bSPD\b"):
             out.append("SPD")
-        if has(r"\bFDP\b"):
+        # F.D.P. (with dots, historical SH/HB-style) and FDP (modern) — same
+        # flexible pattern as ParLDokAdapter so the test suite stays consistent.
+        if has(r"\bF\.?\s*D\.?\s*P\.?\b"):
             out.append("FDP")
         if has(r"\bAFD\b"):
             out.append("AfD")
@@ -451,7 +453,9 @@ class PortalaAdapter(ParlamentAdapter):
             out.append("LINKE")
         if has(r"\bBSW\b"):
             out.append("BSW")
-        if has(r"LANDESREGIERUNG|SENAT VON BERLIN|REGIERENDE[RN]?\s+BÜRGERMEISTER|MINISTER\b|STAATSKANZLEI"):
+        # MINISTERIUM/MINISTER beide treffen — \bMINISTER ohne abschließende
+        # Wortgrenze, damit "Ministerium der Finanzen" mit erfasst wird.
+        if has(r"LANDESREGIERUNG|SENAT VON BERLIN|REGIERENDE[RN]?\s+BÜRGERMEISTER|\bMINISTER|STAATSKANZLEI|MINISTERPRÄSIDENT"):
             out.append("Landesregierung")
         return out
 
@@ -944,7 +948,10 @@ class ParLDokAdapter(ParlamentAdapter):
             out.append("LINKE")
         if re.search(r"\bBSW\b", u):
             out.append("BSW")
-        if re.search(r"LANDESREGIERUNG|MINISTER\b|STAATSKANZLEI|MINISTERPRÄSIDENT", u):
+        # \bMINISTER ohne abschließende Wortgrenze, damit MINISTERIUM
+        # auch trifft (z.B. "Ministerium der Finanzen" als Urheber von
+        # Landesregierungs-Drucksachen).
+        if re.search(r"LANDESREGIERUNG|\bMINISTER|STAATSKANZLEI|MINISTERPRÄSIDENT", u):
             out.append("Landesregierung")
         return out
 

pytest.ini

@@ -0,0 +1,5 @@
+[pytest]
+testpaths = tests
+asyncio_mode = auto
+filterwarnings =
+    ignore::DeprecationWarning

requirements-dev.txt

@@ -0,0 +1,13 @@
+# Test- und Entwicklungs-Abhängigkeiten — getrennt von requirements.txt,
+# damit der prod-Container sie nicht installieren muss.
+#
+# Installation lokal:
+#   pip install -r requirements.txt -r requirements-dev.txt
+#
+# Tests laufen lassen:
+#   pytest -v tests/
+
+-r requirements.txt
+
+pytest>=8.0.0
+pytest-asyncio>=0.24.0

tests/conftest.py

@@ -0,0 +1,51 @@
+"""Shared pytest fixtures and path setup.
+
+Stubs heavy optional dependencies (``fitz``/PyMuPDF, ``bs4``/BeautifulSoup,
+``openai``) so the test suite can run without the full prod requirements
+installed. The tests in this directory are pure unit tests over parser
+logic and prompt formatters — they neither parse PDFs nor make HTTP
+calls, so the stubs are inert placeholders that satisfy the import
+machinery but never get exercised.
+
+If a test ever does need real PyMuPDF or httpx integration, give it a
+fixture marked with ``@pytest.mark.integration`` and skip it by default.
+"""
+import sys
+import types
+from pathlib import Path
+
+# Make the `app` package importable when pytest is run from the webapp/ root.
+ROOT = Path(__file__).resolve().parent.parent
+sys.path.insert(0, str(ROOT))
+
+
+def _stub(name: str, **attrs) -> None:
+    if name in sys.modules:
+        return
+    mod = types.ModuleType(name)
+    for k, v in attrs.items():
+        setattr(mod, k, v)
+    sys.modules[name] = mod
+
+
+_stub("fitz")  # PyMuPDF — used for PDF parsing, not in unit tests
+_stub("bs4", BeautifulSoup=lambda *a, **kw: None)  # only needed by NRWAdapter live calls
+_stub("openai", OpenAI=lambda **kw: None)  # only needed by embeddings live calls
+
+
+# pydantic_settings is a small but external dep that's not in the test
+# environment. Stub it with a minimal BaseSettings shim so app.config can
+# import without crashing — the tests don't actually read settings values.
+class _BaseSettingsShim:
+    model_config: dict = {}
+
+    def __init__(self, **kwargs):
+        for k, v in kwargs.items():
+            setattr(self, k, v)
+
+
+def _settings_config_dict(**kwargs):
+    return kwargs
+
+
+_stub("pydantic_settings", BaseSettings=_BaseSettingsShim, SettingsConfigDict=_settings_config_dict)

tests/test_analyzer.py

@@ -0,0 +1,62 @@
+"""Tests for analyzer.py JSON-stripping logic.
+
+Reproduces the markdown-codeblock-stripping in the LLM retry loop. Real
+Qwen responses sometimes wrap their JSON in ```json …``` fences (despite
+the prompt asking for raw JSON), and the analyzer must tolerate that
+without resorting to retries.
+"""
+import json
+import sys
+import types
+
+# Stub openai before importing analyzer
+if "openai" not in sys.modules:
+    openai_stub = types.ModuleType("openai")
+    openai_stub.OpenAI = lambda **kw: None
+    sys.modules["openai"] = openai_stub
+
+
+def _strip_markdown_fences(content: str) -> str:
+    """Mirror the analyzer's markdown-stripping snippet so we can unit-test
+    the parsing rules without actually invoking the LLM.
+
+    Keep this in sync with analyzer.py around the `if content.startswith("```")`
+    branch — if the analyzer changes, this helper changes too. The point of
+    the duplication is that the analyzer's stripping is buried in an async
+    LLM call that we cannot easily unit-test directly.
+    """
+    content = content.strip()
+    if content.startswith("```"):
+        content = content.split("\n", 1)[1]
+    if content.endswith("```"):
+        content = content.rsplit("```", 1)[0]
+    if content.startswith("```json"):
+        content = content[7:]
+    return content.strip()
+
+
+SAMPLE_JSON = '{"gwoeScore": 7.0, "title": "Test"}'
+
+
+class TestMarkdownStripping:
+    def test_plain_json_unchanged(self):
+        assert _strip_markdown_fences(SAMPLE_JSON) == SAMPLE_JSON
+
+    def test_json_in_markdown_fence(self):
+        wrapped = f"```json\n{SAMPLE_JSON}\n```"
+        cleaned = _strip_markdown_fences(wrapped)
+        assert json.loads(cleaned)["gwoeScore"] == 7.0
+
+    def test_json_in_plain_fence(self):
+        wrapped = f"```\n{SAMPLE_JSON}\n```"
+        cleaned = _strip_markdown_fences(wrapped)
+        assert json.loads(cleaned)["gwoeScore"] == 7.0
+
+    def test_leading_whitespace_stripped(self):
+        wrapped = f"   \n  {SAMPLE_JSON}  \n  "
+        assert json.loads(_strip_markdown_fences(wrapped))["gwoeScore"] == 7.0
+
+    def test_trailing_fence_stripped(self):
+        wrapped = f"{SAMPLE_JSON}\n```"
+        cleaned = _strip_markdown_fences(wrapped)
+        assert json.loads(cleaned)["gwoeScore"] == 7.0

tests/test_bundeslaender.py

@@ -0,0 +1,81 @@
+"""Tests for bundeslaender.py — sanity over 16-state registry.
+
+Includes the #48 classification regression: TH must be ParlDok, HB must
+be StarWeb, SN must be Eigensystem (not ParlDok).
+"""
+from app.bundeslaender import BUNDESLAENDER, get, aktive_bundeslaender, alle_bundeslaender
+
+
+class TestRegistryStructure:
+    def test_sixteen_bundeslaender(self):
+        assert len(BUNDESLAENDER) == 16
+
+    def test_codes_are_uppercase(self):
+        for code in BUNDESLAENDER:
+            assert code.isupper(), f"{code} is not uppercase"
+
+    def test_each_entry_has_naechste_wahl_or_none(self):
+        for code, bl in BUNDESLAENDER.items():
+            assert bl.naechste_wahl is None or len(bl.naechste_wahl) == 10
+
+    def test_wahlperiode_is_positive_integer(self):
+        for bl in BUNDESLAENDER.values():
+            assert isinstance(bl.wahlperiode, int) and bl.wahlperiode > 0
+
+
+class TestActiveBundeslaender:
+    def test_four_active_bundeslaender(self):
+        active = aktive_bundeslaender()
+        codes = {bl.code for bl in active}
+        assert codes == {"NRW", "LSA", "MV", "BE"}
+
+    def test_alle_bundeslaender_returns_all_sixteen(self):
+        assert len(alle_bundeslaender()) == 16
+
+    def test_alle_bundeslaender_active_first(self):
+        out = alle_bundeslaender()
+        active_codes = {bl.code for bl in aktive_bundeslaender()}
+        # The first len(active) entries must all be active
+        for bl in out[: len(active_codes)]:
+            assert bl.code in active_codes
+
+
+class TestGetHelper:
+    def test_returns_bundesland_for_known_code(self):
+        bl = get("NRW")
+        assert bl is not None
+        assert bl.name == "Nordrhein-Westfalen"
+
+    def test_returns_none_for_unknown_code(self):
+        assert get("XX") is None
+
+
+class TestClassificationFix48:
+    """Regression: #48 corrected three doku_system entries that the
+    follow-up adapter issues depend on."""
+
+    def test_th_is_parldok_not_starweb(self):
+        assert BUNDESLAENDER["TH"].doku_system == "ParlDok"
+
+    def test_hb_is_starweb_not_paris(self):
+        """PARiS is just a StarWeb skin — must be classified as StarWeb."""
+        assert BUNDESLAENDER["HB"].doku_system == "StarWeb"
+
+    def test_sn_is_eigensystem_not_parldok(self):
+        """EDAS is ASP.NET-Webforms, NOT ParlDok-compatible with MV."""
+        assert BUNDESLAENDER["SN"].doku_system == "Eigensystem"
+
+
+class TestWahltermineSane:
+    """All upcoming elections must be in chronological order and in the
+    near future (sanity check that someone has not pasted a 1990 date)."""
+
+    def test_no_election_before_2026(self):
+        for bl in BUNDESLAENDER.values():
+            if bl.naechste_wahl:
+                assert bl.naechste_wahl >= "2026-01-01"
+
+    def test_no_election_after_2035(self):
+        for bl in BUNDESLAENDER.values():
+            if bl.naechste_wahl:
+                assert bl.naechste_wahl < "2035-01-01"

tests/test_embeddings.py

@@ -0,0 +1,151 @@
+"""Tests for embeddings.py prompt formatting.
+
+Reproduces the LLM-Halluzinations-Bug from the 2026-04-08 session
+(commits 1b5fd96 + bc7f4a6): the original ``format_quotes_for_prompt``
+rendered each chunk as ``- S. X: "text"`` without any reference to the
+programme name. As a result the LLM hallucinated familiar source labels
+("FDP NRW Wahlprogramm 2022") for chunks that actually came from MV/BE,
+because that was the strongest training-set prior for budget-policy
+citations.
+
+Fix: prepend the fully-qualified PROGRAMME[programm_id]["name"] to each
+quote.
+"""
+import sys
+import types
+
+# Stub openai before importing embeddings, since the test environment may
+# not have it installed and we don't actually need to make API calls.
+if "openai" not in sys.modules:
+    openai_stub = types.ModuleType("openai")
+    openai_stub.OpenAI = lambda **kw: None
+    sys.modules["openai"] = openai_stub
+
+from app.embeddings import _chunk_source_label, format_quotes_for_prompt
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# _chunk_source_label — fully-qualified programme name + page
+# ─────────────────────────────────────────────────────────────────────────────
+
+class TestChunkSourceLabel:
+    def test_known_programme_id(self):
+        chunk = {"programm_id": "fdp-mv-2021", "seite": 73, "text": "..."}
+        label = _chunk_source_label(chunk)
+        assert "FDP Mecklenburg-Vorpommern" in label
+        assert "S. 73" in label
+
+    def test_known_programme_id_for_be(self):
+        chunk = {"programm_id": "spd-be-2023", "seite": 24, "text": "..."}
+        label = _chunk_source_label(chunk)
+        assert "SPD Berlin" in label
+        assert "2021" in label  # the BE-2023.pdf files contain 2021er programmes
+        assert "S. 24" in label
+
+    def test_unknown_programme_id_falls_back_to_id(self):
+        chunk = {"programm_id": "fake-xx-9999", "seite": 1, "text": "..."}
+        label = _chunk_source_label(chunk)
+        # Should not crash, should at least include the id and the page
+        assert "fake-xx-9999" in label
+        assert "S. 1" in label
+
+    def test_missing_seite_uses_questionmark(self):
+        chunk = {"programm_id": "cdu-mv-2021", "text": "..."}
+        label = _chunk_source_label(chunk)
+        assert "?" in label
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# format_quotes_for_prompt — every chunk must carry programme identification
+# ─────────────────────────────────────────────────────────────────────────────
+
+EXAMPLE_QUOTES = {
+    "FDP": {
+        "wahlprogramm": [
+            {
+                "programm_id": "fdp-mv-2021",
+                "partei": "FDP",
+                "typ": "wahlprogramm",
+                "seite": 73,
+                "text": "Die Grundsätze von Wirtschaftlichkeit und Sparsamkeit",
+                "similarity": 0.63,
+            },
+        ],
+        "parteiprogramm": [
+            {
+                "programm_id": "fdp-grundsatz",
+                "partei": "FDP",
+                "typ": "parteiprogramm",
+                "seite": 93,
+                "text": "Liberale Marktwirtschaft erfordert solide Haushalte",
+                "similarity": 0.60,
+            },
+        ],
+    },
+    "SPD": {
+        "wahlprogramm": [
+            {
+                "programm_id": "spd-mv-2021",
+                "partei": "SPD",
+                "typ": "wahlprogramm",
+                "seite": 22,
+                "text": "Verkehrswende weg vom motorisierten Individualverkehr",
+                "similarity": 0.58,
+            },
+        ],
+    },
+}
+
+
+class TestFormatQuotesForPrompt:
+    def test_empty_input_returns_empty_string(self):
+        assert format_quotes_for_prompt({}) == ""
+
+    def test_renders_party_headings(self):
+        out = format_quotes_for_prompt(EXAMPLE_QUOTES)
+        assert "### FDP" in out
+        assert "### SPD" in out
+
+    def test_every_chunk_has_programme_name(self):
+        """Regression: pre-fix this used "S. X:" only, no programme name —
+        the LLM then hallucinated NRW-2022 sources from training data."""
+        out = format_quotes_for_prompt(EXAMPLE_QUOTES)
+        # Each of the three chunks must reference its source programme
+        assert "FDP Mecklenburg-Vorpommern" in out
+        assert "FDP Grundsatzprogramm" in out
+        assert "SPD Mecklenburg-Vorpommern" in out
+
+    def test_contains_strict_citation_instruction(self):
+        """The prompt header must explicitly forbid hallucinated sources."""
+        out = format_quotes_for_prompt(EXAMPLE_QUOTES)
+        assert "ausschließlich" in out.lower() or "verbatim" in out.lower() or "wörtlich" in out.lower()
+
+    def test_no_nrw_2022_appears_unless_chunks_are_actually_nrw(self):
+        """Sanity: a pure MV+SPD chunk set must not mention NRW anywhere."""
+        out = format_quotes_for_prompt(EXAMPLE_QUOTES)
+        assert "NRW" not in out
+        assert "Nordrhein-Westfalen" not in out
+
+    def test_renders_separate_blocks_for_wahl_and_parteiprogramm(self):
+        out = format_quotes_for_prompt(EXAMPLE_QUOTES)
+        assert "**Wahlprogramm:**" in out
+        assert "**Grundsatzprogramm:**" in out
+
+    def test_text_truncated_at_500_chars(self):
+        long_chunk = {
+            "FDP": {
+                "wahlprogramm": [
+                    {
+                        "programm_id": "fdp-mv-2021",
+                        "seite": 1,
+                        "text": "A" * 1000,  # 1000 chars → should be truncated
+                        "similarity": 0.7,
+                    }
+                ],
+            }
+        }
+        out = format_quotes_for_prompt(long_chunk)
+        # Truncation marker
+        assert "..." in out
+        # Original chunk text 1000 chars not present in full
+        assert "A" * 1000 not in out

tests/test_parlamente.py

@@ -0,0 +1,328 @@
+"""Tests for parlamente.py adapter parsers — pure functions over fixture HTML.
+
+Reproduces the three regression scenarios from the 2026-04-08 adapter session:
+
+1. PortalaAdapter `_parse_hit_list_cards` had a `doctype` vs. `doctype_full`
+   NameError that was hot-fixed live on the prod server (commit 1cb030a).
+2. ParLDokAdapter `_hit_to_drucksache` needs to map ParlDok 8.x JSON hit
+   dicts to Drucksache objects without losing fraction or date info.
+3. PortalaAdapter `_normalize_fraktion` and ParLDokAdapter same-named method
+   must yield canonical fraction codes for both comma-lists and embedded
+   "MdL (Partei)" patterns.
+"""
+from app.parlamente import ParLDokAdapter, PortalaAdapter, NRWAdapter, Drucksache
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# PortalaAdapter — Berlin-style HTML cards
+# ─────────────────────────────────────────────────────────────────────────────
+
+BE_CARD_FIXTURE = """
+<div class="other-prelude">ignored</div>
+<div class="record-card efxRecordRepeater">
+  <h3 class="h5"><span>Schwimmstatistik für die dritten Klassen der Berliner Schulen</span></h3>
+  <span class="h6">Antrag (Eilantrag)&nbsp;<a href="/files/drs19-3104.pdf">Drucksache 19/3104</a> S. 1 bis 24 vom 31.03.2026</span>
+</div>
+<div class="record-card efxRecordRepeater">
+  <h3 class="h5"><span>Klimaneutrales Bauen im Bestand</span></h3>
+  <span class="h6">Antrag CDU, SPD&nbsp;<a href="/files/drs19-3107.pdf">Drucksache 19/3107</a> vom 02.04.2026</span>
+</div>
+"""
+
+
+def _make_be_adapter():
+    return PortalaAdapter(
+        bundesland="BE",
+        name="test BE",
+        base_url="https://pardok.parlament-berlin.de",
+        db_id="lah.lissh",
+        wahlperiode=19,
+        portala_path="/portala",
+        document_type=None,
+    )
+
+
+class TestPortalaAdapterCardParser:
+    """Issue: doctype/doctype_full NameError (hot-fix 1cb030a)."""
+
+    def test_parses_two_cards_without_nameerror(self):
+        """Smoke test — must NOT raise NameError or any other exception.
+
+        Pre-fix this method referenced an undefined ``doctype`` variable
+        in the query-filter branch when computing the haystack. The fix
+        renamed it to ``doctype_full``. A direct call covers both branches.
+        """
+        adapter = _make_be_adapter()
+        result = adapter._parse_hit_list_cards(BE_CARD_FIXTURE, query_filter="")
+        assert len(result) == 2
+
+    def test_first_card_extracts_drucksache_and_title(self):
+        adapter = _make_be_adapter()
+        result = adapter._parse_hit_list_cards(BE_CARD_FIXTURE, query_filter="")
+        d = result[0]
+        assert d.drucksache == "19/3104"
+        assert "Schwimmstatistik" in d.title
+        assert d.datum == "2026-03-31"
+        assert d.bundesland == "BE"
+
+    def test_second_card_extracts_fraktionen_from_h6(self):
+        """Card 2 packs CDU+SPD into the type line — must split out cleanly."""
+        adapter = _make_be_adapter()
+        result = adapter._parse_hit_list_cards(BE_CARD_FIXTURE, query_filter="")
+        d = result[1]
+        assert d.drucksache == "19/3107"
+        assert d.fraktionen == ["CDU", "SPD"]
+        # The typ string should have the parties stripped back out
+        assert d.typ.strip() == "Antrag"
+
+    def test_pdf_link_is_absolute_url(self):
+        adapter = _make_be_adapter()
+        result = adapter._parse_hit_list_cards(BE_CARD_FIXTURE, query_filter="")
+        assert result[0].link.startswith("https://pardok.parlament-berlin.de/")
+        assert result[0].link.endswith(".pdf")
+
+    def test_query_filter_uses_doctype_full_not_doctype(self):
+        """Regression: the filter branch references doctype_full, not doctype.
+
+        Pre-fix this raised NameError as soon as a query was passed.
+        """
+        adapter = _make_be_adapter()
+        # Schwimmstatistik matches card 1, Klimaneutral matches card 2
+        result = adapter._parse_hit_list_cards(BE_CARD_FIXTURE, query_filter="Schwimm")
+        assert len(result) == 1
+        assert result[0].drucksache == "19/3104"
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# PortalaAdapter — LSA-style Perl-Dump records
+# ─────────────────────────────────────────────────────────────────────────────
+
+LSA_DUMP_FIXTURE = """
+<pre>$VAR1 = {
+  'WEV06' => [{ 'main' => 'Demokratie beginnt im Klassenzimmer' }],
+  'WEV32' => [{
+    'main' => 'Antrag B\\x{fc}ndnis 90/Die Gr\\x{fc}nen 06.03.2026 Drucksache <b>8/6726</b> ...',
+    '5' => 'drs/wp8/drs/d6726lan.pdf'
+  }]
+}</pre>
+<pre>$VAR1 = {
+  'WEV06' => [{ 'main' => 'Andere Drucksache ohne Schul-Bezug' }],
+  'WEV32' => [{
+    'main' => 'Antrag CDU, SPD 14.01.2026 Drucksache <b>8/6171</b> ...',
+    '5' => 'drs/wp8/drs/d6171lan.pdf'
+  }]
+}</pre>
+"""
+
+
+def _make_lsa_adapter():
+    return PortalaAdapter(
+        bundesland="LSA",
+        name="test LSA",
+        base_url="https://padoka.landtag.sachsen-anhalt.de",
+        db_id="lsa.lissh",
+        wahlperiode=8,
+        portala_path="/portal",
+        document_type="Antrag",
+    )
+
+
+class TestPortalaAdapterDumpParser:
+    def test_parses_two_dump_records(self):
+        adapter = _make_lsa_adapter()
+        result = adapter._parse_hit_list_dump(LSA_DUMP_FIXTURE, query_filter="")
+        assert len(result) == 2
+
+    def test_extracts_drucksache_from_perl_dump(self):
+        adapter = _make_lsa_adapter()
+        result = adapter._parse_hit_list_dump(LSA_DUMP_FIXTURE, query_filter="")
+        assert result[0].drucksache == "8/6726"
+        assert result[1].drucksache == "8/6171"
+
+    def test_decodes_perl_hex_escapes_in_urheber(self):
+        """The first record has \\x{fc} (ü) and \\x{e4} (ä) in WEV32."""
+        adapter = _make_lsa_adapter()
+        result = adapter._parse_hit_list_dump(LSA_DUMP_FIXTURE, query_filter="")
+        # GRÜNE should be detected from "Bündnis 90/Die Grünen"
+        assert "GRÜNE" in result[0].fraktionen
+
+    def test_extracts_date_iso(self):
+        adapter = _make_lsa_adapter()
+        result = adapter._parse_hit_list_dump(LSA_DUMP_FIXTURE, query_filter="")
+        assert result[0].datum == "2026-03-06"
+        assert result[1].datum == "2026-01-14"
+
+    def test_pdf_url_uses_pdf_url_prefix(self):
+        adapter = _make_lsa_adapter()
+        result = adapter._parse_hit_list_dump(LSA_DUMP_FIXTURE, query_filter="")
+        assert result[0].link == (
+            "https://padoka.landtag.sachsen-anhalt.de/files/drs/wp8/drs/d6726lan.pdf"
+        )
+
+    def test_client_side_query_filter(self):
+        adapter = _make_lsa_adapter()
+        result = adapter._parse_hit_list_dump(LSA_DUMP_FIXTURE, query_filter="Demokratie")
+        assert len(result) == 1
+        assert result[0].drucksache == "8/6726"
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# PortalaAdapter — Auto-detection between dump and card formats
+# ─────────────────────────────────────────────────────────────────────────────
+
+class TestPortalaAdapterAutoDetect:
+    def test_dump_html_routes_to_dump_parser(self):
+        adapter = _make_lsa_adapter()
+        result = adapter._parse_hit_list_html(LSA_DUMP_FIXTURE, query_filter="")
+        assert len(result) == 2
+
+    def test_card_html_routes_to_card_parser(self):
+        adapter = _make_be_adapter()
+        result = adapter._parse_hit_list_html(BE_CARD_FIXTURE, query_filter="")
+        assert len(result) == 2
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# PortalaAdapter._normalize_fraktion — canonical fraction codes
+# ─────────────────────────────────────────────────────────────────────────────
+
+class TestPortalaAdapterNormalizeFraktion:
+    def test_comma_separated_list(self):
+        out = PortalaAdapter._normalize_fraktion("CDU, SPD, F.D.P.")
+        assert "CDU" in out and "SPD" in out and "FDP" in out
+
+    def test_buendnis_90_die_gruenen(self):
+        out = PortalaAdapter._normalize_fraktion("BÜNDNIS 90/DIE GRÜNEN")
+        assert out == ["GRÜNE"]
+
+    def test_die_linke(self):
+        out = PortalaAdapter._normalize_fraktion("DIE LINKE")
+        assert out == ["LINKE"]
+
+    def test_bsw(self):
+        out = PortalaAdapter._normalize_fraktion("BSW")
+        assert out == ["BSW"]
+
+    def test_landesregierung_keywords(self):
+        out = PortalaAdapter._normalize_fraktion("Senat von Berlin")
+        assert "Landesregierung" in out
+
+    def test_empty_string(self):
+        assert PortalaAdapter._normalize_fraktion("") == []
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# ParLDokAdapter — JSON hit dict → Drucksache mapping
+# ─────────────────────────────────────────────────────────────────────────────
+
+def _make_mv_adapter():
+    return ParLDokAdapter(
+        bundesland="MV",
+        name="test MV",
+        base_url="https://www.dokumentation.landtag-mv.de",
+        wahlperiode=8,
+        prefix="/parldok",
+        document_typ="Antrag",
+    )
+
+
+SAMPLE_PARLDOK_HIT = {
+    "id": 70748,
+    "title": "Zweckentfremdung von Sondervermögen des Bundes beenden",
+    "date": "18.03.2026",
+    "prelink": "/dokument/70748",
+    "link": "/dokument/70748#navpanes=0",
+    "authorhtml": "FDP",
+    "kind": "Drucksache",
+    "type": "Antrag",
+    "lp": 8,
+    "number": "6409",
+}
+
+
+class TestParLDokAdapterHitMapping:
+    def test_hit_to_drucksache_basic(self):
+        adapter = _make_mv_adapter()
+        d = adapter._hit_to_drucksache(SAMPLE_PARLDOK_HIT)
+        assert d is not None
+        assert d.drucksache == "8/6409"
+        assert d.title == "Zweckentfremdung von Sondervermögen des Bundes beenden"
+        assert d.datum == "2026-03-18"
+        assert d.fraktionen == ["FDP"]
+        assert d.typ == "Antrag"
+        assert d.bundesland == "MV"
+
+    def test_pdf_link_strips_navpanes_fragment_and_prepends_prefix(self):
+        adapter = _make_mv_adapter()
+        d = adapter._hit_to_drucksache(SAMPLE_PARLDOK_HIT)
+        assert d.link == "https://www.dokumentation.landtag-mv.de/parldok/dokument/70748"
+        assert "#navpanes" not in d.link
+
+    def test_missing_lp_returns_none(self):
+        adapter = _make_mv_adapter()
+        hit = dict(SAMPLE_PARLDOK_HIT)
+        del hit["lp"]
+        assert adapter._hit_to_drucksache(hit) is None
+
+    def test_mdl_with_party_in_parens(self):
+        """MV often packs the MdL into authorhtml: 'Thomas X (AfD)'."""
+        adapter = _make_mv_adapter()
+        hit = dict(SAMPLE_PARLDOK_HIT, authorhtml="Thomas de Jesus Fernandes (AfD)")
+        d = adapter._hit_to_drucksache(hit)
+        assert "AfD" in d.fraktionen
+
+    def test_landesregierung_detection(self):
+        adapter = _make_mv_adapter()
+        hit = dict(SAMPLE_PARLDOK_HIT, authorhtml="Ministerium der Finanzen")
+        d = adapter._hit_to_drucksache(hit)
+        assert "Landesregierung" in d.fraktionen
+
+
+class TestParLDokFulltextIdSanitization:
+    """Reverse-engineered from bundle.js pd.getFulltextId — must mirror exactly.
+
+    Even though server-side fulltext is currently disabled (#18), the helper
+    is kept around in code as documentation. If it ever gets re-activated,
+    the sanitization must still match the SPA's behavior 1:1.
+    """
+
+    def test_simple_word_unchanged(self):
+        assert ParLDokAdapter._fulltext_id("Schule") == "Schule"
+
+    def test_whitespace_becomes_dash(self):
+        assert ParLDokAdapter._fulltext_id("Klima Schutz") == "Klima-Schutz"
+
+    def test_umlauts_become_dashes(self):
+        # The JS regex is /[^a-zA-z0-9]/ — note the lowercase z, deliberate.
+        # Umlauts are non-ASCII so they get replaced.
+        assert ParLDokAdapter._fulltext_id("Bürger") == "B-rger"
+
+    def test_punctuation_becomes_dashes(self):
+        assert ParLDokAdapter._fulltext_id("CO2-Emission") == "CO2-Emission"
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Adapter registry sanity
+# ─────────────────────────────────────────────────────────────────────────────
+
+class TestAdapterRegistry:
+    def test_active_adapters_present(self):
+        from app.parlamente import ADAPTERS
+        for code in ["NRW", "MV", "BE", "LSA"]:
+            assert code in ADAPTERS, f"missing adapter for {code}"
+
+    def test_get_adapter_returns_none_for_unknown(self):
+        from app.parlamente import get_adapter
+        assert get_adapter("XX") is None
+
+    def test_mv_adapter_is_parldok_instance(self):
+        from app.parlamente import ADAPTERS
+        assert isinstance(ADAPTERS["MV"], ParLDokAdapter)
+
+    def test_be_adapter_is_portala_instance(self):
+        from app.parlamente import ADAPTERS
+        assert isinstance(ADAPTERS["BE"], PortalaAdapter)
+
+    def test_lsa_adapter_is_portala_instance(self):
+        from app.parlamente import ADAPTERS
+        assert isinstance(ADAPTERS["LSA"], PortalaAdapter)

tests/test_wahlprogramme.py

@@ -0,0 +1,118 @@
+"""Tests for wahlprogramme.py — registry consistency + file existence."""
+from app.wahlprogramme import (
+    WAHLPROGRAMME,
+    REFERENZEN_PATH,
+    get_wahlprogramm,
+    parteien_mit_wahlprogramm,
+)
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Registry consistency
+# ─────────────────────────────────────────────────────────────────────────────
+
+class TestRegistryStructure:
+    def test_active_bundeslaender_present(self):
+        for code in ["NRW", "LSA", "MV", "BE"]:
+            assert code in WAHLPROGRAMME, f"missing wahlprogramme entry for {code}"
+
+    def test_each_entry_has_required_keys(self):
+        required = {"file", "titel", "partei", "jahr", "seiten"}
+        for bl, parteien in WAHLPROGRAMME.items():
+            for partei, info in parteien.items():
+                missing = required - set(info.keys())
+                assert not missing, f"{bl}/{partei} missing keys: {missing}"
+
+    def test_jahr_is_integer(self):
+        for bl, parteien in WAHLPROGRAMME.items():
+            for partei, info in parteien.items():
+                assert isinstance(info["jahr"], int), f"{bl}/{partei} jahr not int"
+
+    def test_seiten_is_positive_integer(self):
+        for bl, parteien in WAHLPROGRAMME.items():
+            for partei, info in parteien.items():
+                assert isinstance(info["seiten"], int)
+                assert info["seiten"] > 0
+
+    def test_file_extension_is_pdf(self):
+        for bl, parteien in WAHLPROGRAMME.items():
+            for partei, info in parteien.items():
+                assert info["file"].endswith(".pdf")
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# File existence — every registered file must exist on disk
+# ─────────────────────────────────────────────────────────────────────────────
+
+class TestFileExistence:
+    """Catches typos in the file field that would silently break embedding
+    indexing or PDF download links."""
+
+    def test_every_registered_pdf_exists(self):
+        missing = []
+        for bl, parteien in WAHLPROGRAMME.items():
+            for partei, info in parteien.items():
+                path = REFERENZEN_PATH / info["file"]
+                if not path.exists():
+                    missing.append(f"{bl}/{partei}: {info['file']}")
+        assert not missing, "missing PDFs:\n  " + "\n  ".join(missing)
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# Lookup helpers
+# ─────────────────────────────────────────────────────────────────────────────
+
+class TestGetWahlprogramm:
+    def test_returns_dict_for_known_combination(self):
+        info = get_wahlprogramm("MV", "CDU")
+        assert info is not None
+        assert info["partei"] == "CDU Mecklenburg-Vorpommern"
+
+    def test_returns_none_for_unknown_bundesland(self):
+        assert get_wahlprogramm("XX", "CDU") is None
+
+    def test_returns_none_for_unknown_partei(self):
+        assert get_wahlprogramm("NRW", "BSW") is None
+
+
+class TestParteienMitWahlprogramm:
+    def test_nrw_has_five_parteien(self):
+        parteien = parteien_mit_wahlprogramm("NRW")
+        assert len(parteien) == 5
+        assert set(parteien) == {"CDU", "SPD", "GRÜNE", "FDP", "AfD"}
+
+    def test_mv_has_six_parteien(self):
+        parteien = parteien_mit_wahlprogramm("MV")
+        assert set(parteien) == {"CDU", "SPD", "GRÜNE", "FDP", "AfD", "LINKE"}
+
+    def test_be_has_five_parteien(self):
+        parteien = parteien_mit_wahlprogramm("BE")
+        assert set(parteien) == {"CDU", "SPD", "GRÜNE", "LINKE", "AfD"}
+
+    def test_unknown_bundesland_empty_list(self):
+        assert parteien_mit_wahlprogramm("XX") == []
+
+
+# ─────────────────────────────────────────────────────────────────────────────
+# embeddings.PROGRAMME consistency cross-check
+# ─────────────────────────────────────────────────────────────────────────────
+
+class TestEmbeddingsRegistryConsistency:
+    """Every entry in WAHLPROGRAMME must also exist in embeddings.PROGRAMME
+    so the indexer can find it. Mismatch is the kind of bug a manual smoke
+    misses but would show up during indexing."""
+
+    def test_every_wahlprogramm_has_embeddings_entry(self):
+        from app.embeddings import PROGRAMME
+
+        # Build expected programm_id from filename: "cdu-mv-2021.pdf" → "cdu-mv-2021"
+        missing = []
+        for bl, parteien in WAHLPROGRAMME.items():
+            for partei, info in parteien.items():
+                pid = info["file"].rsplit(".", 1)[0]
+                if pid not in PROGRAMME:
+                    missing.append(f"{bl}/{partei} → {pid}")
+        assert not missing, (
+            "WAHLPROGRAMME entries missing in embeddings.PROGRAMME:\n  "
+            + "\n  ".join(missing)
+        )