Add implementation plan for compass UI improvements

1 month ago · 6914b2284a
parent c5b39ced5f
commit 6914b2284a
1 changed files with 643 additions and 0 deletions
--- a/docs/superpowers/plans/2026-03-28-compass-ui-improvements.md
+++ b/docs/superpowers/plans/2026-03-28-compass-ui-improvements.md
@ -0,0 +1,643 @@
+# Compass UI Improvements Implementation Plan
+
+> **For agentic workers:** REQUIRED SUB-SKILL: Use superpowers:subagent-driven-development (recommended) or superpowers:executing-plans to implement this plan task-by-task. Steps use checkbox (`- [ ]`) syntax for tracking.
+
+**Goal:** Fix three independent UI issues in the political compass: (1) update stale axis 3/4/5 descriptions in SVD_THEMES, (2) fix broken Y-axis direction arrows, (3) add voting discipline section below compass.
+
+**Architecture:** All changes are confined to `explorer.py`. No new files. No schema changes. The discipline helper reads `mp_votes` read-only via DuckDB. Tests for the discipline function live in `tests/test_political_compass.py`.
+
+**Tech Stack:** Python, Streamlit, Plotly Express/Graph Objects, DuckDB, pytest (run via `uv run pytest`)
+
+---
+
+## File Map
+
+| File | Change |
+|------|--------|
+| `explorer.py` | Update SVD_THEMES axes 3–5; fix Y-axis labels in two px.scatter calls and one update_layout; add `compute_party_discipline`; add discipline rendering in `build_compass_tab` |
+| `tests/test_political_compass.py` | Add tests for `compute_party_discipline` |
+
+---
+
+## Task 1: Update SVD_THEMES axes 3, 4, 5
+
+**Files:**
+- Modify: `explorer.py:1156–1204`
+
+These descriptions were written for an earlier dataset. The new text reflects stable multi-year patterns (not a single year's snapshot). The `flip` booleans are unchanged.
+
+**New text for axis 3** (flip=True — currently "Sociaal-economisch links versus marktliberaal en landelijk rechts"):
+
+The positive pole consistently shows SP, PvdD, GL-PvdA on social welfare motions; the negative pole shows VVD on market-oriented motions. But PVV also appears prominently positive (anti-establishment spending motions), meaning this is not a clean left-right economic axis — it's more accurately described as *state intervention versus market liberalism*, with populist anti-establishment motions on the same side as the socialist left.
+
+**New text for axis 4** (flip=True — currently "Christelijk-sociaal centrum versus populistisch-soevereinistisch"):
+
+NSC, SGP, CU and CDA consistently dominate the positive pole; VVD and GL-PvdA the negative. FVD scores near zero across years. The axis captures *religious-conservative institutionalism vs secular liberalism*, not populism vs mainstream.
+
+**New text for axis 5** (flip=False — currently "Christelijk-conservatief en ruraal sociaal versus seculier-progressief"):
+
+CDA, CU, SGP score positive; SP, PvdD, DENK score negative. D66 tends positive (not strongly negative), and NSC tends negative. The stable pattern is *established-institutional vs protest-populist* rather than a religious/secular split.
+
+- [ ] **Step 1: Replace axis 3 entry in `explorer.py`**
+
+In `explorer.py`, replace lines 1156–1171:
+
+```python
+        3: {
+            "label": "Staatsingrijpen en publieke sector versus marktliberalisme",
+            "explanation": (
+                "Deze as weerspiegelt de spanning tussen staatsingrijpen en marktliberalisme. "
+                "Aan de positieve kant staan SP-moties die bezuinigingen op zorg en gemeentefonds "
+                "willen terugdraaien, winstuitkeringen in de zorg verbieden en publieke controle "
+                "over fusies eisen. Ook PVV stemt positief — niet vanuit sociaal ideaal maar vanuit "
+                "anti-establishment populisme dat neigt naar overheidsinterventie voor de eigen achterban. "
+                "Aan de negatieve kant staan VVD-moties over marktwerking en deregulering, en NSC- en "
+                "BBB-moties met een agrarisch-marktgericht karakter. "
+                "Samengevat: de as scheidt voorstanders van staatsingrijpen (links én populistisch-rechts) "
+                "van marktliberalen en agrarisch-rechts."
+            ),
+            "positive_pole": "Staatsingrijpen: SP, PvdD, GL-PvdA, PVV (populistisch)",
+            "negative_pole": "Marktliberaal en agrarisch-rechts: VVD, NSC, BBB",
+            "flip": True,
+        },
+```
+
+- [ ] **Step 2: Replace axis 4 entry in `explorer.py`**
+
+Replace lines 1172–1188:
+
+```python
+        4: {
+            "label": "Christelijk-conservatief institutionalisme versus seculier-liberalisme",
+            "explanation": (
+                "Deze as scheidt christelijk-conservatieve partijen die hechten aan traditionele "
+                "instituties en religieuze waarden (NSC, SGP, CU, CDA) van seculier-liberale partijen "
+                "die nadruk leggen op individuele rechten en internationale openheid (VVD, GL-PvdA). "
+                "CU-moties over vaderbetrokkenheid, huwelijksrecht en internationale samenwerking staan "
+                "aan de positieve kant; VVD-moties over marktregulering en D66-moties over internationale "
+                "verdragen aan de negatieve kant. FVD scoort dicht bij nul — het past noch in het "
+                "christelijk-conservatieve noch in het seculier-liberale kamp op deze as. "
+                "Dit is geen populisme-as maar een religieus-institutionele breuklijn."
+            ),
+            "positive_pole": "Christelijk-conservatief institutioneel: NSC, SGP, CU, CDA",
+            "negative_pole": "Seculier-liberaal: VVD, GL-PvdA, D66",
+            "flip": True,
+        },
+```
+
+- [ ] **Step 3: Replace axis 5 entry in `explorer.py`**
+
+Replace lines 1189–1204:
+
+```python
+        5: {
+            "label": "Gevestigd-institutioneel versus protest en populistisch",
+            "explanation": (
+                "Deze as scheidt gevestigde centrumpartijen die vertrouwen op bestaande instituties "
+                "(CDA, CU, SGP, D66) van protest- en populistische partijen die dat vertrouwen "
+                "afwijzen (SP, PvdD, DENK, NSC). CDA-moties over vrijwilligers in schuldhulp, "
+                "maatschappelijke diensttijd en WW-hervorming staan aan de positieve kant. "
+                "SP- en PvdD-moties over meerouderschap, abortusrecht en buitenlandse beïnvloeding "
+                "staan aan de negatieve kant. NSC scoort negatief — ondanks zijn christelijk-conservatieve "
+                "karakter op andere assen is het hier een protestpartij die systeemkritiek uitdraagt. "
+                "D66 scoort licht positief, consistent met zijn institutionele en pro-EU profiel."
+            ),
+            "positive_pole": "Gevestigd-institutioneel: CDA, CU, SGP, D66",
+            "negative_pole": "Protest en populistisch: SP, PvdD, DENK, NSC",
+            "flip": False,
+        },
+```
+
+- [ ] **Step 4: Run tests to confirm nothing broken**
+
+```bash
+uv run pytest tests/test_political_compass.py -v
+```
+
+Expected: all 3 tests PASS (these tests don't touch SVD_THEMES).
+
+- [ ] **Step 5: Commit**
+
+```bash
+git add explorer.py
+git commit -m "fix: update SVD_THEMES axes 3-5 descriptions to reflect stable multi-year patterns"
+```
+
+---
+
+## Task 2: Fix Y-axis direction indicators in compass and trajectories
+
+**Files:**
+- Modify: `explorer.py:810–812` (partijen scatter labels)
+- Modify: `explorer.py:830` (kamerleden scatter labels)
+- Modify: `explorer.py:833–838` (compass update_layout — add helper call)
+- Modify: `explorer.py:921–927` (trajectories update_layout)
+
+Plotly rotates Y-axis titles 90° counter-clockwise, so `↑` and `↓` in the title string point sideways. Fix: strip arrows from the axis title; add two `fig.add_annotation` calls to place `▲ Progressief` at the top and `▼ Conservatief` at the bottom of the chart area using `xref="paper", yref="paper"`.
+
+- [ ] **Step 1: Add `_add_y_direction_annotations` helper near top of `build_compass_tab`**
+
+Add this function just before `build_compass_tab` (after line 692, before line 694):
+
+```python
+def _add_y_direction_annotations(fig: go.Figure) -> None:
+    """Add ▲ Progressief / ▼ Conservatief labels above and below the Y axis."""
+    common = dict(
+        xref="paper",
+        yref="paper",
+        x=-0.07,
+        showarrow=False,
+        font=dict(size=11, color="#666666"),
+    )
+    fig.add_annotation(**common, y=1.02, text="▲ Progressief", xanchor="center")
+    fig.add_annotation(**common, y=-0.06, text="▼ Conservatief", xanchor="center")
+```
+
+- [ ] **Step 2: Update labels in the "Partijen" scatter (line 810–814)**
+
+Change:
+```python
+            labels={
+                "x": "Links ← → Rechts",
+                "y": "Progressief ↑ / Conservatief ↓",
+                "n": "Kamerleden",
+            },
+```
+
+To:
+```python
+            labels={
+                "x": "Links ← → Rechts",
+                "y": "Progressief / Conservatief",
+                "n": "Kamerleden",
+            },
+```
+
+- [ ] **Step 3: Update labels in the "Kamerleden" scatter (line 830)**
+
+Change:
+```python
+            labels={"x": "Links ← → Rechts", "y": "Progressief ↑ / Conservatief ↓"},
+```
+
+To:
+```python
+            labels={"x": "Links ← → Rechts", "y": "Progressief / Conservatief"},
+```
+
+- [ ] **Step 4: Call the annotation helper after `fig.update_layout` in `build_compass_tab` (after line 838)**
+
+Change:
+```python
+    fig.update_layout(
+        height=600,
+        legend_title_text="Partij",
+        xaxis={"range": [-1, 1]},
+        yaxis={"range": [-0.6, 0.6]},
+    )
+
+    with col1:
+        st.plotly_chart(fig, use_container_width=True)
+```
+
+To:
+```python
+    fig.update_layout(
+        height=600,
+        legend_title_text="Partij",
+        xaxis={"range": [-1, 1]},
+        yaxis={"range": [-0.6, 0.6]},
+    )
+    _add_y_direction_annotations(fig)
+
+    with col1:
+        st.plotly_chart(fig, use_container_width=True)
+```
+
+- [ ] **Step 5: Fix trajectories Y-axis title (line 924) and add annotation**
+
+Change `build_trajectories_tab` update_layout block:
+```python
+    fig.update_layout(
+        title="Partij trajectories",
+        xaxis_title="Links ← → Rechts",
+        yaxis_title="Progressief ↑ / Conservatief ↓",
+        height=600,
+        legend_title_text="Partij",
+    )
+    st.plotly_chart(fig, use_container_width=True)
+```
+
+To:
+```python
+    fig.update_layout(
+        title="Partij trajectories",
+        xaxis_title="Links ← → Rechts",
+        yaxis_title="Progressief / Conservatief",
+        height=600,
+        legend_title_text="Partij",
+    )
+    _add_y_direction_annotations(fig)
+    st.plotly_chart(fig, use_container_width=True)
+```
+
+- [ ] **Step 6: Run tests**
+
+```bash
+uv run pytest tests/test_political_compass.py -v
+```
+
+Expected: all 3 tests PASS.
+
+- [ ] **Step 7: Commit**
+
+```bash
+git add explorer.py
+git commit -m "fix: replace sideways Y-axis arrows with proper top/bottom annotations"
+```
+
+---
+
+## Task 3: Add voting discipline section below compass
+
+**Files:**
+- Modify: `explorer.py` — add `compute_party_discipline` function; add rendering block in `build_compass_tab`
+- Modify: `tests/test_political_compass.py` — add two tests
+
+### 3a: Write the failing tests first
+
+- [ ] **Step 1: Add tests to `tests/test_political_compass.py`**
+
+Append at the end of `tests/test_political_compass.py`:
+
+```python
+# ---------------------------------------------------------------------------
+# Tests for compute_party_discipline
+# ---------------------------------------------------------------------------
+
+
+def _make_mp_votes_db():
+    """Create an in-memory DuckDB with mp_votes fixture data.
+
+    6 motions, 2 parties (SP, VVD), each with 4 MPs.
+    SP is perfectly disciplined (all 4 vote the same each time).
+    VVD has 1 dissident on 2 of 6 motions → Rice index = (4+4+4+4+3+3)/6/4 ≈ 0.917.
+    Dates span 2023-01-01 to 2023-12-31.
+    """
+    import duckdb
+
+    conn = duckdb.connect(":memory:")
+    conn.execute("""
+        CREATE TABLE mp_votes (
+            id INTEGER,
+            motion_id VARCHAR,
+            mp_name VARCHAR,
+            party VARCHAR,
+            vote VARCHAR,
+            date DATE,
+            created_at TIMESTAMP
+        )
+    """)
+    rows = []
+    # motions 1-6, dates in 2023
+    dates = [
+        "2023-01-10",
+        "2023-03-15",
+        "2023-05-20",
+        "2023-07-25",
+        "2023-09-30",
+        "2023-11-05",
+    ]
+    sp_mps = ["Janssen, A.", "Pietersen, B.", "Willemsen, C.", "Hendriksen, D."]
+    vvd_mps = ["Adams, E.", "Bakker, F.", "Claassen, G.", "Dekker, H."]
+    for i, date in enumerate(dates, start=1):
+        m_id = f"M{i:03d}"
+        # SP: all vote 'voor' every motion (perfectly disciplined)
+        for mp in sp_mps:
+            rows.append((i * 10 + 1, m_id, mp, "SP", "voor", date, "2023-01-01"))
+        # VVD: motions 5 and 6 have one dissident (votes 'tegen' while others vote 'voor')
+        if i <= 4:
+            for mp in vvd_mps:
+                rows.append((i * 10 + 2, m_id, mp, "VVD", "voor", date, "2023-01-01"))
+        else:
+            for mp in vvd_mps[:3]:
+                rows.append((i * 10 + 2, m_id, mp, "VVD", "voor", date, "2023-01-01"))
+            rows.append((i * 10 + 3, m_id, vvd_mps[3], "VVD", "tegen", date, "2023-01-01"))
+    conn.executemany(
+        "INSERT INTO mp_votes VALUES (?, ?, ?, ?, ?, ?, ?)", rows
+    )
+    return conn
+
+
+def test_compute_party_discipline_basic(monkeypatch):
+    """compute_party_discipline returns correct Rice index for fixture data."""
+    import duckdb as _duckdb
+
+    fixture_conn = _make_mp_votes_db()
+
+    monkeypatch.setattr(
+        _duckdb, "connect", lambda path, **kw: fixture_conn
+    )
+
+    # Import after monkeypatch so explorer can be imported without Streamlit crashing
+    import importlib
+    import sys
+
+    # explorer imports streamlit — provide a minimal stub if not already stubbed
+    if "streamlit" not in sys.modules:
+        import types
+        st_stub = types.ModuleType("streamlit")
+        st_stub.cache_data = lambda **kw: (lambda f: f)
+        sys.modules["streamlit"] = st_stub
+
+    # We need the function directly; import the module
+    import explorer as _explorer
+    importlib.reload(_explorer)
+
+    df = _explorer.compute_party_discipline(
+        db_path="dummy",
+        start_date="2023-01-01",
+        end_date="2023-12-31",
+    )
+
+    assert not df.empty
+    assert set(df.columns) >= {"party", "n_motions", "discipline"}
+
+    sp_row = df[df["party"] == "SP"].iloc[0]
+    vvd_row = df[df["party"] == "VVD"].iloc[0]
+
+    assert sp_row["n_motions"] == 6
+    assert sp_row["discipline"] == pytest.approx(1.0, abs=1e-6)
+
+    assert vvd_row["n_motions"] == 6
+    # 4 motions fully disciplined (4/4=1.0), 2 motions with one dissident (3/4=0.75)
+    expected_vvd = (4 * 1.0 + 2 * 0.75) / 6
+    assert vvd_row["discipline"] == pytest.approx(expected_vvd, abs=1e-4)
+
+    # All values in [0, 1]
+    assert (df["discipline"] >= 0).all() and (df["discipline"] <= 1).all()
+
+
+def test_compute_party_discipline_empty_range(monkeypatch):
+    """Returns empty DataFrame when no motions fall in the date range."""
+    import duckdb as _duckdb
+
+    fixture_conn = _make_mp_votes_db()
+    monkeypatch.setattr(_duckdb, "connect", lambda path, **kw: fixture_conn)
+
+    import importlib, sys
+
+    if "streamlit" not in sys.modules:
+        import types
+        st_stub = types.ModuleType("streamlit")
+        st_stub.cache_data = lambda **kw: (lambda f: f)
+        sys.modules["streamlit"] = st_stub
+
+    import explorer as _explorer
+    importlib.reload(_explorer)
+
+    df = _explorer.compute_party_discipline(
+        db_path="dummy",
+        start_date="2000-01-01",
+        end_date="2000-12-31",
+    )
+
+    assert df.empty
+```
+
+- [ ] **Step 2: Run the failing tests**
+
+```bash
+uv run pytest tests/test_political_compass.py::test_compute_party_discipline_basic tests/test_political_compass.py::test_compute_party_discipline_empty_range -v
+```
+
+Expected: FAIL with `AttributeError: module 'explorer' has no attribute 'compute_party_discipline'`
+
+### 3b: Implement `compute_party_discipline`
+
+- [ ] **Step 3: Add `compute_party_discipline` to `explorer.py`**
+
+Add this function after the `load_active_mps` function (find it, then place `compute_party_discipline` immediately after). The function must be a plain function (not decorated with `@st.cache_data`) so tests can monkeypatch `duckdb.connect`.
+
+```python
+def compute_party_discipline(
+    db_path: str,
+    start_date: str,
+    end_date: str,
+) -> pd.DataFrame:
+    """Compute per-party voting discipline (Rice index) for roll-call votes in a date range.
+
+    Only individual MP vote rows are used (mp_name LIKE '%,%').
+    Returns a DataFrame with columns [party, n_motions, discipline] sorted by discipline ascending.
+    Returns an empty DataFrame if fewer than 1 qualifying motion exists or on any DB error.
+
+    Rice index per motion per party = fraction of party MPs voting with the party majority.
+    The per-party score is the average Rice index across all motions in the date range.
+    """
+    try:
+        conn = duckdb.connect(db_path, read_only=True)
+        result = conn.execute(
+            """
+            WITH individual_votes AS (
+                -- Only individual MP rows (mp_name contains a comma, e.g. "Janssen, A.")
+                SELECT
+                    motion_id,
+                    party,
+                    LOWER(vote) AS vote
+                FROM mp_votes
+                WHERE mp_name LIKE '%,%'
+                  AND date >= CAST(? AS DATE)
+                  AND date <= CAST(? AS DATE)
+                  AND vote IN ('voor', 'tegen', 'afwezig', 'onthouden')
+            ),
+            vote_counts AS (
+                -- Count each vote token per (motion, party)
+                SELECT
+                    motion_id,
+                    party,
+                    vote,
+                    COUNT(*) AS cnt
+                FROM individual_votes
+                GROUP BY motion_id, party, vote
+            ),
+            majority_vote AS (
+                -- Determine the majority vote token per (motion, party)
+                SELECT
+                    motion_id,
+                    party,
+                    FIRST(vote ORDER BY cnt DESC, vote ASC) AS maj_vote,
+                    SUM(cnt) AS total_mp_votes
+                FROM vote_counts
+                GROUP BY motion_id, party
+            ),
+            rice_per_motion AS (
+                -- Rice index: fraction voting with majority
+                SELECT
+                    mv.motion_id,
+                    mv.party,
+                    SUM(CASE WHEN vc.vote = mv.maj_vote THEN vc.cnt ELSE 0 END)
+                        * 1.0 / mv.total_mp_votes AS rice
+                FROM majority_vote mv
+                JOIN vote_counts vc
+                  ON mv.motion_id = vc.motion_id AND mv.party = vc.party
+                GROUP BY mv.motion_id, mv.party, mv.total_mp_votes
+            )
+            SELECT
+                party,
+                COUNT(DISTINCT motion_id) AS n_motions,
+                AVG(rice) AS discipline
+            FROM rice_per_motion
+            GROUP BY party
+            ORDER BY discipline ASC
+            """,
+            [start_date, end_date],
+        ).fetchdf()
+        conn.close()
+        return result
+    except Exception as exc:
+        logger.warning("compute_party_discipline failed: %s", exc)
+        return pd.DataFrame(columns=["party", "n_motions", "discipline"])
+```
+
+- [ ] **Step 4: Run the tests to confirm they pass**
+
+```bash
+uv run pytest tests/test_political_compass.py::test_compute_party_discipline_basic tests/test_political_compass.py::test_compute_party_discipline_empty_range -v
+```
+
+Expected: both PASS.
+
+- [ ] **Step 5: Run all compass tests**
+
+```bash
+uv run pytest tests/test_political_compass.py -v
+```
+
+Expected: all 5 tests PASS.
+
+### 3c: Render discipline section in `build_compass_tab`
+
+- [ ] **Step 6: Add `_window_to_dates` helper just before `build_compass_tab`**
+
+Add this function just before `build_compass_tab` (around line 692, after the `_add_y_direction_annotations` helper added in Task 2):
+
+```python
+def _window_to_dates(window_id: str) -> tuple[str, str]:
+    """Return (start_date, end_date) ISO strings for a given window_id.
+
+    Annual windows like '2024' → ('2024-01-01', '2024-12-31').
+    'current_parliament' → ('2023-11-22', '2099-12-31') (2023 formation date, open end).
+    Unknown formats → ('2000-01-01', '2099-12-31') (effectively all time).
+    """
+    if window_id == "current_parliament":
+        return ("2023-11-22", "2099-12-31")
+    if re.fullmatch(r"\d{4}", window_id):
+        return (f"{window_id}-01-01", f"{window_id}-12-31")
+    # Quarterly e.g. '2020-Q3' → 2020-07-01 to 2020-09-30
+    m = re.fullmatch(r"(\d{4})-Q([1-4])", window_id)
+    if m:
+        year, q = int(m.group(1)), int(m.group(2))
+        starts = {1: "01-01", 2: "04-01", 3: "07-01", 4: "10-01"}
+        ends = {1: "03-31", 2: "06-30", 3: "09-30", 4: "12-31"}
+        return (f"{year}-{starts[q]}", f"{year}-{ends[q]}")
+    return ("2000-01-01", "2099-12-31")
+```
+
+- [ ] **Step 7: Add discipline rendering after `st.plotly_chart` in `build_compass_tab`**
+
+The current end of `build_compass_tab` is (around line 840–841):
+
+```python
+    with col1:
+        st.plotly_chart(fig, use_container_width=True)
+```
+
+Add the discipline section immediately after (still inside the function, but outside the `with col1:` block):
+
+```python
+    # --- Voting discipline section ---
+    _MIN_MOTIONS_FOR_DISCIPLINE = 5
+    start_date, end_date = _window_to_dates(window_idx)
+    disc_df = compute_party_discipline(db_path, start_date, end_date)
+
+    st.subheader("Stemgedrag cohesie")
+    if disc_df.empty or disc_df["n_motions"].max() < _MIN_MOTIONS_FOR_DISCIPLINE:
+        st.caption(
+            "Te weinig hoofdelijke stemmingen in dit venster voor een cohesieanalyse."
+        )
+    else:
+        # Filter to parties that appear in the compass
+        compass_parties = set(df_pos["party"].unique())
+        disc_df = disc_df[disc_df["party"].isin(compass_parties)].copy()
+
+        if disc_df.empty:
+            st.caption("Geen overlappende partijen tussen kompas en stemmingsdata.")
+        else:
+            disc_df["discipline_pct"] = (disc_df["discipline"] * 100).round(1)
+            disc_df["party_label"] = disc_df.apply(
+                lambda r: f"{r['party']} ({int(r['n_motions'])} moties)", axis=1
+            )
+
+            bar_fig = px.bar(
+                disc_df.sort_values("discipline"),
+                x="discipline_pct",
+                y="party_label",
+                orientation="h",
+                color="discipline_pct",
+                color_continuous_scale="RdYlGn",
+                range_color=[80, 100],
+                labels={"discipline_pct": "Cohesie (%)", "party_label": "Partij"},
+                title="Cohesie bij hoofdelijke stemmingen",
+            )
+            bar_fig.update_layout(
+                height=max(300, len(disc_df) * 35 + 80),
+                showlegend=False,
+                coloraxis_showscale=False,
+                yaxis_title="",
+            )
+            st.plotly_chart(bar_fig, use_container_width=True)
+
+            # Extremes table
+            top3 = disc_df.nlargest(3, "discipline")[["party", "discipline_pct", "n_motions"]]
+            bot3 = disc_df.nsmallest(3, "discipline")[["party", "discipline_pct", "n_motions"]]
+            col_a, col_b = st.columns(2)
+            with col_a:
+                st.markdown("**Meest eensgezind**")
+                st.dataframe(
+                    top3.rename(columns={"party": "Partij", "discipline_pct": "Cohesie (%)", "n_motions": "Moties"}),
+                    hide_index=True,
+                    use_container_width=True,
+                )
+            with col_b:
+                st.markdown("**Meest verdeeld**")
+                st.dataframe(
+                    bot3.rename(columns={"party": "Partij", "discipline_pct": "Cohesie (%)", "n_motions": "Moties"}),
+                    hide_index=True,
+                    use_container_width=True,
+                )
+```
+
+- [ ] **Step 8: Run all tests**
+
+```bash
+uv run pytest tests/test_political_compass.py -v
+```
+
+Expected: all 5 tests PASS.
+
+- [ ] **Step 9: Commit**
+
+```bash
+git add explorer.py tests/test_political_compass.py
+git commit -m "feat: add voting discipline section below political compass"
+```
+
+---
+
+## Self-Review Checklist
+
+- [x] **Spec coverage:** Task 1 → SVD_THEMES axes 3–5. Task 2 → Y-axis arrows. Task 3 → discipline function + rendering. All three design requirements covered.
+- [x] **Placeholder scan:** No TBD or TODO. All code blocks are complete.
+- [x] **Type consistency:** `compute_party_discipline` returns `pd.DataFrame` with columns `["party", "n_motions", "discipline"]` — referenced consistently in tests and rendering code. `_window_to_dates` returns `tuple[str, str]` — used as `start_date, end_date` in rendering.
+- [x] **Test isolation:** Tests monkeypatch `duckdb.connect` to return in-memory DB; tests add a minimal `streamlit` stub to avoid import errors. Both patterns match existing test style in the file.
+- [x] **Edge case:** `_MIN_MOTIONS_FOR_DISCIPLINE = 5` guard ensures the section degrades gracefully for sparse windows. Empty DataFrame from `compute_party_discipline` is also handled.