diff --git a/pytest.ini b/pytest.ini
index e93a5da..09177d9 100644
--- a/pytest.ini
+++ b/pytest.ini
@@ -12,6 +12,7 @@ markers =
     e2e: end-to-end CLI / integration tests
     install: import / dependency sanity tests
     fixture_sweep: parametrized sweep over the test-cases/ folder
+    gui: Streamlit AppTest-driven tests (live in tests/gui/)
 
 # Warnings discipline: fail on unexpected DeprecationWarning from our own
 # code, but tolerate third-party deprecations that we can't fix.
diff --git a/tests/gui/__init__.py b/tests/gui/__init__.py
new file mode 100644
index 0000000..52c3274
--- /dev/null
+++ b/tests/gui/__init__.py
@@ -0,0 +1,6 @@
+"""GUI tests — Streamlit AppTest-driven coverage of pages/, components/, and i18n.
+
+Marked with ``@pytest.mark.gui`` so a quick core-only run can skip them
+via ``pytest -m 'not gui'``. The default ``pytest`` invocation still runs
+everything; the marker is opt-out, not opt-in.
+"""
diff --git a/tests/gui/_findings_panel_harness.py b/tests/gui/_findings_panel_harness.py
new file mode 100644
index 0000000..0513b8c
--- /dev/null
+++ b/tests/gui/_findings_panel_harness.py
@@ -0,0 +1,44 @@
+"""Test harness page for ``render_findings_panel``.
+
+A standalone Streamlit page module the AppTest layer can drive
+directly. Renders the findings panel with whatever findings live in
+``st.session_state["test_findings"]`` so test code can inject a list
+and inspect what's rendered, without having to fake a file_uploader
+widget.
+
+Lives next to its test file so it ships with the GUI test layer and
+never gets confused with a real page.
+"""
+
+from __future__ import annotations
+
+import sys
+from pathlib import Path
+
+import streamlit as st
+
+# Same sys.path bootstrap as the real pages so ``src.*`` imports work
+# regardless of how AppTest invokes the script.
+_project_root = Path(__file__).resolve().parent.parent.parent
+if str(_project_root) not in sys.path:
+    sys.path.insert(0, str(_project_root))
+
+
+# ``st.page_link`` requires a multipage-app context (Streamlit looks up
+# the target page's URL from the app's PagesManager). AppTest doesn't
+# wire that up for a standalone page, so any ``page_link`` call raises
+# ``KeyError: 'url_pathname'`` here. We swap it for a markdown stub
+# that renders the label inline — same observable text, no nav, no
+# crash.
+def _page_link_stub(page: str, *, label: str, **_kwargs) -> None:
+    st.markdown(f"[{label}]")
+
+st.page_link = _page_link_stub  # type: ignore[assignment]
+
+from src.gui.components import hide_streamlit_chrome, render_findings_panel
+
+st.set_page_config(page_title="findings test", page_icon="🧪", layout="wide")
+hide_streamlit_chrome()
+
+findings = st.session_state.get("test_findings", [])
+render_findings_panel(findings)
diff --git a/tests/gui/conftest.py b/tests/gui/conftest.py
new file mode 100644
index 0000000..5e98ef9
--- /dev/null
+++ b/tests/gui/conftest.py
@@ -0,0 +1,204 @@
+"""Shared fixtures for the GUI test layer.
+
+Streamlit's ``AppTest.from_file`` runs a page module in-process inside
+its own ScriptRunContext. Each fixture here returns either bytes (for
+the upload-session-state path) or a configured ``AppTest`` ready to
+``.run()``. Tests should NOT call ``AppTest.from_file`` directly so the
+project-root path and the default session-state stash live in one place.
+
+Page paths are resolved relative to the repo root so the test suite
+works from any cwd (e.g., ``pytest`` from a subdir, IDE runners).
+"""
+
+from __future__ import annotations
+
+import hashlib
+from pathlib import Path
+from typing import Iterable
+
+import pandas as pd
+import pytest
+
+from streamlit.testing.v1 import AppTest
+
+
+# ---------------------------------------------------------------------------
+# Paths
+# ---------------------------------------------------------------------------
+
+PROJECT_ROOT: Path = Path(__file__).resolve().parent.parent.parent
+PAGES_DIR: Path = PROJECT_ROOT / "src" / "gui" / "pages"
+APP_PATH: Path = PROJECT_ROOT / "src" / "gui" / "app.py"
+SAMPLES_DIR: Path = PROJECT_ROOT / "samples"
+
+
+# All GUI tests get the marker automatically so a single
+# ``pytest -m 'not gui'`` skips the whole subdir.
+def pytest_collection_modifyitems(config, items):
+    for item in items:
+        # Only mark items collected from this subdir.
+        if "tests/gui/" in str(item.fspath).replace("\\", "/"):
+            item.add_marker(pytest.mark.gui)
+
+
+# ---------------------------------------------------------------------------
+# Sample data
+# ---------------------------------------------------------------------------
+
+@pytest.fixture
+def messy_sales_bytes() -> bytes:
+    """Raw bytes of the ``messy_sales.csv`` sample (mixed dup + text noise)."""
+    return (SAMPLES_DIR / "messy_sales.csv").read_bytes()
+
+
+@pytest.fixture
+def small_csv_bytes() -> bytes:
+    """A tiny CSV the tool pages can chew through fast — three rows, three
+    columns, one obvious duplicate. Suitable for happy-path workflow tests
+    where wall-clock matters."""
+    return (
+        b"name,email,phone\n"
+        b"Alice,alice@gmail.com,5551234567\n"
+        b"Alice,Alice@Gmail.com,(555) 123-4567\n"
+        b"Bob,bob@example.com,5559876543\n"
+    )
+
+
+@pytest.fixture
+def small_csv_df(small_csv_bytes) -> pd.DataFrame:
+    """The ``small_csv_bytes`` parsed — for tests that need the DataFrame
+    form (e.g., direct ``deduplicate()`` calls within a test)."""
+    import io
+    return pd.read_csv(io.BytesIO(small_csv_bytes), dtype=str, keep_default_na=False)
+
+
+# ---------------------------------------------------------------------------
+# AppTest builders
+# ---------------------------------------------------------------------------
+
+def _new_app(page_path: Path) -> AppTest:
+    """Build an AppTest pointed at *page_path*. Keeps the project-root /
+    page-resolution boilerplate out of the test bodies."""
+    if not page_path.exists():
+        raise FileNotFoundError(f"GUI test target missing: {page_path}")
+    return AppTest.from_file(str(page_path))
+
+
+@pytest.fixture
+def home_app() -> AppTest:
+    """A fresh AppTest pointed at the home page (``src/gui/app.py``)."""
+    return _new_app(APP_PATH)
+
+
+@pytest.fixture
+def app_factory():
+    """Callable returning a fresh AppTest for any page slug.
+
+    Usage::
+
+        app = app_factory("1_Deduplicator")
+        app.run()
+    """
+    def _make(slug: str) -> AppTest:
+        return _new_app(PAGES_DIR / f"{slug}.py")
+    return _make
+
+
+# ---------------------------------------------------------------------------
+# Upload-session helpers
+# ---------------------------------------------------------------------------
+
+def stash_upload(app: AppTest, *, name: str, data: bytes) -> str:
+    """Pre-populate the home-screen upload stash + the gate's normalisation
+    result so a tool page renders past ``require_normalization_gate()``.
+
+    Returns the SHA-256 hex of *data* (used as the gate key) in case the
+    test wants to assert against it.
+
+    The gate checks::
+
+        - ``home_uploaded_bytes`` is set
+        - ``normalization_for == sha256(home_uploaded_bytes)``
+        - ``normalization_result.passed is True``
+
+    We synthesise a passing result via a tiny stub object that satisfies
+    the gate's only attribute access (``.passed``). Tests that want to
+    exercise gate-blocking behaviour should NOT call this helper — they
+    should stash bytes without the normalisation result.
+    """
+    sha = hashlib.sha256(data).hexdigest()
+    app.session_state["home_uploaded_bytes"] = data
+    app.session_state["home_uploaded_name"] = name
+    app.session_state["home_uploaded_size"] = len(data)
+    app.session_state["normalization_for"] = sha
+    app.session_state["normalization_result"] = _PassedGateResult()
+    return sha
+
+
+class _PassedGateResult:
+    """Minimal stand-in for the real NormalizationResult shape — the gate
+    only reads ``.passed``. Using a real NormalizationResult here would
+    pull in core.normalize and tie GUI tests to its constructor surface.
+    """
+
+    passed: bool = True
+
+
+def stash_upload_without_gate(app: AppTest, *, name: str, data: bytes) -> None:
+    """Stash the upload bytes but do NOT pre-pass the gate. Used by gate
+    tests that want the warning + Go-to-Review button to appear."""
+    app.session_state["home_uploaded_bytes"] = data
+    app.session_state["home_uploaded_name"] = name
+    app.session_state["home_uploaded_size"] = len(data)
+
+
+# ---------------------------------------------------------------------------
+# i18n helpers
+# ---------------------------------------------------------------------------
+
+def with_language(app: AppTest, lang: str) -> None:
+    """Set the active GUI language *before* ``app.run()``. The selector
+    widget reads ``_ui_lang_select`` on first render; we set the canonical
+    ``ui_lang`` key (what ``current_language()`` reads) directly so the
+    first render is already in the chosen language."""
+    app.session_state["ui_lang"] = lang
+
+
+def collected_text(app: AppTest) -> str:
+    """Flatten every text-bearing element on the page into one big string
+    so a test can assert ``"Maestría" in collected_text(app)`` without
+    juggling individual widget collections.
+
+    Covers: title, header, subheader, caption, markdown, info, warning,
+    error, success, code, metric labels/values, button labels, expander
+    labels, page_link labels. Doesn't try to be exhaustive — if a widget
+    type isn't here, add it (cheap, mechanical).
+    """
+    chunks: list[str] = []
+
+    def _extend(seq: Iterable, attrs: tuple[str, ...] = ("value", "label", "body")) -> None:
+        for el in seq:
+            for a in attrs:
+                v = getattr(el, a, None)
+                if isinstance(v, str) and v:
+                    chunks.append(v)
+                    break  # one value per element is enough
+
+    _extend(app.title)
+    _extend(app.header)
+    _extend(app.subheader)
+    _extend(app.caption)
+    _extend(app.markdown)
+    _extend(app.info)
+    _extend(app.warning)
+    _extend(app.error)
+    _extend(app.success)
+    _extend(app.button)
+    # Sidebar caption / markdown / button (Streamlit exposes the sidebar
+    # as a sub-tree with the same widget collections).
+    if hasattr(app, "sidebar"):
+        sb = app.sidebar
+        _extend(sb.markdown)
+        _extend(sb.caption)
+        _extend(sb.button)
+    return "\n".join(chunks)
diff --git a/tests/gui/test_advanced_panels.py b/tests/gui/test_advanced_panels.py
new file mode 100644
index 0000000..ae0b200
--- /dev/null
+++ b/tests/gui/test_advanced_panels.py
@@ -0,0 +1,194 @@
+"""Advanced-options panel tests.
+
+``config_panel`` (in ``src.gui.components``) is the dedup-page's
+expander that houses every per-column / per-strategy knob. It's the
+densest single widget surface in the GUI, so a session-state key drift
+in there cascades into every dedup session.
+
+We exercise it via the Deduplicator page (rendering ``config_panel``
+in isolation requires a fake Streamlit context). The page provides
+the surrounding state; we poke widgets and verify their effects.
+"""
+
+from __future__ import annotations
+
+import pandas as pd
+import pytest
+
+from .conftest import stash_upload
+
+
+GATED_PAGE = "1_Deduplicator"
+
+
+def _render_page(app_factory, small_csv_bytes):
+    app = app_factory(GATED_PAGE)
+    stash_upload(app, name="messy.csv", data=small_csv_bytes)
+    app.run()
+    return app
+
+
+# ---------------------------------------------------------------------------
+# Expander presence + collapsed state
+# ---------------------------------------------------------------------------
+
+class TestAdvancedExpander:
+    def test_advanced_options_expander_renders(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        labels = [e.label for e in app.expander]
+        assert any("Advanced Options" in lbl for lbl in labels), (
+            f"Advanced Options expander missing; expanders: {labels}"
+        )
+
+
+# ---------------------------------------------------------------------------
+# Algorithm selector
+# ---------------------------------------------------------------------------
+
+class TestAlgorithmSelector:
+    """The fuzzy-algorithm dropdown drives ``Algorithm.{LEVENSHTEIN,
+    JARO_WINKLER, TOKEN_SET_RATIO}`` on every column. Default value
+    must be jaro_winkler — the strong-key build_default_strategies
+    assumes it."""
+
+    def test_default_algorithm_is_jaro_winkler(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        # Find the selectbox by label.
+        algo_boxes = [
+            sb for sb in app.selectbox
+            if sb.label == "Fuzzy algorithm"
+        ]
+        assert len(algo_boxes) == 1
+        assert algo_boxes[0].value == "jaro_winkler"
+
+    def test_algorithm_options_complete(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        algo = next(sb for sb in app.selectbox if sb.label == "Fuzzy algorithm")
+        assert set(algo.options) == {
+            "jaro_winkler", "levenshtein", "token_set_ratio",
+        }
+
+
+# ---------------------------------------------------------------------------
+# Threshold slider
+# ---------------------------------------------------------------------------
+
+class TestThresholdSlider:
+    def test_default_threshold_is_85(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        thresholds = [s for s in app.slider if "Similarity" in (s.label or "")]
+        assert len(thresholds) == 1
+        assert thresholds[0].value == 85
+
+    def test_threshold_bounds(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        thr = next(s for s in app.slider if "Similarity" in (s.label or ""))
+        assert thr.min == 50
+        assert thr.max == 100
+
+
+# ---------------------------------------------------------------------------
+# Survivor rule selector
+# ---------------------------------------------------------------------------
+
+class TestSurvivorSelector:
+    def test_default_is_first(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        surv = next(
+            sb for sb in app.selectbox
+            if sb.label == "Survivor rule"
+        )
+        assert surv.value == "first"
+
+    def test_all_four_rules_offered(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        surv = next(sb for sb in app.selectbox if sb.label == "Survivor rule")
+        assert set(surv.options) == {
+            "first", "last", "most-complete", "most-recent",
+        }
+
+    def test_selecting_most_recent_does_not_crash(
+        self, app_factory, small_csv_bytes,
+    ):
+        """When ``most-recent`` is chosen the page should reveal a
+        Date column dropdown. Pin the no-crash invariant."""
+        app = _render_page(app_factory, small_csv_bytes)
+        surv = next(sb for sb in app.selectbox if sb.label == "Survivor rule")
+        surv.select("most-recent").run()
+        assert not app.exception
+
+
+# ---------------------------------------------------------------------------
+# Merge checkbox
+# ---------------------------------------------------------------------------
+
+class TestMergeCheckbox:
+    def test_merge_default_off(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        merge_boxes = [c for c in app.checkbox if c.label == "Merge mode"]
+        assert len(merge_boxes) == 1
+        assert merge_boxes[0].value is False
+
+    def test_toggling_merge_does_not_crash(
+        self, app_factory, small_csv_bytes,
+    ):
+        app = _render_page(app_factory, small_csv_bytes)
+        merge = next(c for c in app.checkbox if c.label == "Merge mode")
+        merge.check().run()
+        assert not app.exception
+        # After checking, the value persists in session_state via the
+        # widget's own key.
+
+
+# ---------------------------------------------------------------------------
+# Column multiselects
+# ---------------------------------------------------------------------------
+
+class TestColumnMultiselects:
+    """Match-on / Strong-keys / Fuzzy multiselects use ``st.multiselect``
+    on every column. Empty default = auto-detect."""
+
+    def test_three_multiselects_present(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        labels = {m.label for m in app.multiselect}
+        assert {"Match on columns", "Strong keys", "Fuzzy columns"} <= labels
+
+    def test_defaults_are_empty(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        for ms in app.multiselect:
+            if ms.label in {
+                "Match on columns", "Strong keys", "Fuzzy columns",
+            }:
+                assert ms.value == [], (
+                    f"{ms.label!r} default should be []; got {ms.value}"
+                )
+
+    def test_options_match_dataframe_columns(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        df_cols = list(app.session_state["df"].columns)
+        for ms in app.multiselect:
+            if ms.label in {
+                "Match on columns", "Strong keys", "Fuzzy columns",
+            }:
+                assert list(ms.options) == df_cols
+
+
+# ---------------------------------------------------------------------------
+# Save / Load config
+# ---------------------------------------------------------------------------
+
+class TestConfigSaveLoadButtons:
+    def test_save_settings_button_present(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        labels = [b.label for b in app.button]
+        assert any("Save current settings" in lbl for lbl in labels)
+
+    def test_config_file_uploader_present(self, app_factory, small_csv_bytes):
+        app = _render_page(app_factory, small_csv_bytes)
+        # AppTest exposes uploaders via ``app.file_uploader``. There
+        # are two: the main file (pickup-or-upload) and the config
+        # JSON. Make sure the config one is there.
+        labels = [u.label for u in app.file_uploader]
+        assert any("Load config profile" in lbl for lbl in labels), (
+            f"config uploader missing; uploaders: {labels}"
+        )
diff --git a/tests/gui/test_chrome.py b/tests/gui/test_chrome.py
new file mode 100644
index 0000000..537f3ea
--- /dev/null
+++ b/tests/gui/test_chrome.py
@@ -0,0 +1,181 @@
+"""Chrome tests — language selector, hide_streamlit_chrome, quit flow.
+
+These verify the GUI plumbing that every page depends on. Failures here
+cascade into every other page, so they run cheap and run first
+(alphabetical name ordering after smoke).
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from .conftest import collected_text, with_language
+
+
+# ---------------------------------------------------------------------------
+# hide_streamlit_chrome mounts the selector
+# ---------------------------------------------------------------------------
+
+class TestHideChromeMountsSelector:
+    """``hide_streamlit_chrome()`` is the one place the language selector
+    is mounted. Every page that hides chrome (= every page) must get
+    exactly one sidebar selectbox with the i18n label."""
+
+    def test_home_has_one_sidebar_selectbox(self, home_app):
+        home_app.run()
+        # Only one selectbox in the sidebar today; if a page adds
+        # another, this becomes a weaker bound.
+        assert len(home_app.sidebar.selectbox) == 1, (
+            "expected exactly one sidebar selectbox (the language picker); "
+            f"got {len(home_app.sidebar.selectbox)}"
+        )
+
+    def test_selector_label_is_localized(self, home_app):
+        with_language(home_app, "es")
+        home_app.run()
+        labels = [sb.label for sb in home_app.sidebar.selectbox]
+        assert "Idioma" in labels, (
+            f"Spanish selector should be labelled 'Idioma'; got {labels}"
+        )
+
+    def test_selector_label_english_default(self, home_app):
+        home_app.run()  # no with_language → default = en
+        labels = [sb.label for sb in home_app.sidebar.selectbox]
+        assert "Language" in labels
+
+
+# ---------------------------------------------------------------------------
+# Language selector switches session state
+# ---------------------------------------------------------------------------
+
+class TestLanguageSwitch:
+    """Picking 'es' in the selector flips ``st.session_state['ui_lang']``
+    and re-renders the page with Spanish strings on the next run."""
+
+    def test_default_language_is_english(self, home_app):
+        home_app.run()
+        # AppTest's session_state proxy doesn't implement .get(); use
+        # membership check + attribute access. Absence == default ("en").
+        lang = home_app.session_state["ui_lang"] if "ui_lang" in home_app.session_state else "en"
+        assert lang == "en"
+        text = collected_text(home_app)
+        assert "Data Cleaning Mastery" in text
+
+    def test_selecting_spanish_persists_in_session(self, home_app):
+        home_app.run()
+        selector = home_app.sidebar.selectbox[0]
+        selector.select("es").run()
+        assert home_app.session_state["ui_lang"] == "es"
+
+    def test_selecting_spanish_re_renders_in_spanish(self, home_app):
+        home_app.run()
+        selector = home_app.sidebar.selectbox[0]
+        selector.select("es").run()
+        text = collected_text(home_app)
+        assert "Maestría" in text, (
+            "after selecting Spanish, the home title should switch to "
+            f"'🧹 DataTools — Maestría…'; got:\n{text[:300]}"
+        )
+
+    def test_selecting_back_to_english_reverts(self, home_app):
+        # Start in Spanish, then flip back.
+        with_language(home_app, "es")
+        home_app.run()
+        assert "Maestría" in collected_text(home_app)
+
+        selector = home_app.sidebar.selectbox[0]
+        selector.select("en").run()
+        text = collected_text(home_app)
+        assert "Data Cleaning Mastery" in text
+        assert "Maestría" not in text
+
+
+# ---------------------------------------------------------------------------
+# Footer + page_title localization
+# ---------------------------------------------------------------------------
+
+class TestLocalizedChrome:
+    """A spot-check on the parts of the chrome that aren't the selector:
+    the bottom footer caption and the home-page hero text. Other strings
+    are pinned indirectly by ``TestEveryPageRenders.test_expected_*``."""
+
+    def test_footer_english(self, home_app):
+        home_app.run()
+        text = collected_text(home_app)
+        assert "Your data never leaves" in text
+
+    def test_footer_spanish(self, home_app):
+        with_language(home_app, "es")
+        home_app.run()
+        text = collected_text(home_app)
+        assert "Tus datos nunca salen" in text
+
+    def test_upload_section_heading_localizes(self, home_app):
+        with_language(home_app, "es")
+        home_app.run()
+        text = collected_text(home_app)
+        # ``📤 Sube un archivo para empezar`` from the es pack.
+        assert "Sube un archivo" in text
+
+
+# ---------------------------------------------------------------------------
+# Quit / Close page
+# ---------------------------------------------------------------------------
+
+class TestQuitButtonRenders:
+    """The Close page must show the localized title, body, and the
+    Close-the-app button. We don't actually click the button — that
+    would call ``os._exit(0)`` and kill the test process. We only
+    assert the button is present and its label is localized."""
+
+    def test_close_page_english(self, app_factory):
+        app = app_factory("99_Close")
+        app.run()
+        text = collected_text(app)
+        assert "Close DataTools" in text
+        labels = [b.label for b in app.button]
+        assert any("Close the app" in lbl for lbl in labels), (
+            f"Close-the-app button missing; buttons: {labels}"
+        )
+
+    def test_close_page_spanish(self, app_factory):
+        app = app_factory("99_Close")
+        with_language(app, "es")
+        app.run()
+        text = collected_text(app)
+        assert "Cerrar DataTools" in text
+        labels = [b.label for b in app.button]
+        assert any("Cerrar la app" in lbl for lbl in labels), (
+            f"Spanish Close button missing; buttons: {labels}"
+        )
+
+    def test_close_body_describes_unsaved_work_warning_es(self, app_factory):
+        app = app_factory("99_Close")
+        with_language(app, "es")
+        app.run()
+        text = collected_text(app)
+        assert "trabajo sin guardar" in text
+
+
+# ---------------------------------------------------------------------------
+# Tool cards use localized names on the home grid
+# ---------------------------------------------------------------------------
+
+class TestHomeToolGridLocalization:
+    """The home grid pulls tool display names through ``tool_name()`` in
+    ``tools_registry``. The Spanish pack provides translations for every
+    tool id; a regression in that wiring would make Spanish users see
+    English names. Pin a few representative ones."""
+
+    @pytest.mark.parametrize("needle", [
+        "Eliminador de duplicados",
+        "Limpiador de texto",
+        "Estandarizador de formatos",
+        "Gestor de valores faltantes",
+        "Mapeador de columnas",
+    ])
+    def test_es_tool_name_on_home_grid(self, home_app, needle):
+        with_language(home_app, "es")
+        home_app.run()
+        text = collected_text(home_app)
+        assert needle in text, f"missing localized tool name {needle!r}"
diff --git a/tests/gui/test_dedup_review.py b/tests/gui/test_dedup_review.py
new file mode 100644
index 0000000..bcb4ae0
--- /dev/null
+++ b/tests/gui/test_dedup_review.py
@@ -0,0 +1,205 @@
+"""Dedup review widget tests.
+
+``match_group_card`` from ``src.gui.components`` has two modes (decided
+/ undecided) and a Confirm/Undo flow keyed by session_state. We test
+each state by exercising the parent Deduplicator page end to end and
+then poking at ``review_decisions`` directly.
+
+Why not unit-test ``match_group_card`` in isolation? AppTest needs a
+real page module, not a function call, so we drive the page and verify
+the side effects on session_state. This catches integration bugs the
+unit test couldn't see (e.g., session-state key drift between the
+page and the component).
+"""
+
+from __future__ import annotations
+
+import pandas as pd
+import pytest
+
+from .conftest import collected_text, stash_upload
+
+
+# We need a frame that produces at least one match group. The 3-row
+# small_csv has two Alice rows that share an email (case-folded) → one
+# group of two members.
+def _run_with_results(app):
+    """Drive the page through to the post-Find-Duplicates state.
+
+    1. First ``run()`` — page picks up the stashed upload, reads it, and
+       renders the preview + Find Duplicates button. ``result`` is None.
+    2. Click Find Duplicates and ``run()`` again — page calls
+       ``deduplicate()`` and stashes the result. Match group cards
+       render on this pass.
+
+    Mirrors what a real user does instead of trying to short-circuit
+    the page by stashing ``result`` directly (the page resets it to
+    None on every new upload).
+    """
+    app.run()
+    target = next(b for b in app.button if "Find Duplicates" in b.label)
+    target.click().run()
+
+
+class TestMatchGroupCardUndecided:
+    """A freshly-found group has no decision → the card renders the
+    interactive editor + Confirm button."""
+
+    def test_card_expander_present(self, app_factory, small_csv_bytes):
+        app = app_factory("1_Deduplicator")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        _run_with_results(app)
+        # An expander per group. The dedup result should produce
+        # exactly one match group on this fixture.
+        result = app.session_state["result"]
+        assert len(result.match_groups) >= 1, (
+            "fixture should produce at least one match group"
+        )
+        # Match group cards use ``st.expander``. AppTest exposes them
+        # via ``app.expander``.
+        labels = [e.label for e in app.expander]
+        assert any("Group 1" in lbl for lbl in labels), (
+            f"undecided card expander missing; got: {labels}"
+        )
+
+    def test_confirm_button_renders_for_undecided_group(
+        self, app_factory, small_csv_bytes,
+    ):
+        app = app_factory("1_Deduplicator")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        _run_with_results(app)
+        # Confirm button is keyed ``confirm_<group_id>``.
+        result = app.session_state["result"]
+        gid = result.match_groups[0].group_id
+        labels = [b.label for b in app.button]
+        # Streamlit renders the button label as "Confirm".
+        assert any(lbl == "Confirm" for lbl in labels), (
+            f"undecided card missing Confirm button; buttons: {labels}"
+        )
+
+
+class TestBatchActions:
+    """Accept All / Reject All / Clear Decisions are the three batch
+    buttons that mutate ``review_decisions`` across all groups."""
+
+    def test_accept_all_populates_decisions(self, app_factory, small_csv_bytes):
+        app = app_factory("1_Deduplicator")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        _run_with_results(app)
+
+        target = next(b for b in app.button if b.label == "Accept All")
+        target.click().run()
+        decisions = app.session_state["review_decisions"]
+        result = app.session_state["result"]
+        assert len(decisions) == len(result.match_groups), (
+            "Accept All should record a decision per group; "
+            f"got {len(decisions)} decisions for "
+            f"{len(result.match_groups)} groups"
+        )
+        # Each Accept-All decision keeps exactly one row (the survivor).
+        for d in decisions.values():
+            assert len(d["keep_indices"]) == 1
+
+    def test_reject_all_keeps_every_member(self, app_factory, small_csv_bytes):
+        app = app_factory("1_Deduplicator")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        _run_with_results(app)
+
+        target = next(b for b in app.button if b.label == "Reject All")
+        target.click().run()
+        decisions = app.session_state["review_decisions"]
+        result = app.session_state["result"]
+        # Reject = keep every member → keep_indices == row_indices.
+        for g in result.match_groups:
+            assert set(decisions[g.group_id]["keep_indices"]) == set(g.row_indices)
+
+    def test_clear_decisions_wipes_state(self, app_factory, small_csv_bytes):
+        app = app_factory("1_Deduplicator")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        _run_with_results(app)
+
+        # Populate decisions via Accept All, then Clear, then verify.
+        accept = next(b for b in app.button if b.label == "Accept All")
+        accept.click().run()
+        assert app.session_state["review_decisions"], (
+            "precondition failed: Accept All didn't populate"
+        )
+
+        clear = next(b for b in app.button if "Clear Decisions" in b.label)
+        clear.click().run()
+        assert app.session_state["review_decisions"] == {}
+
+
+class TestApplyReviewDecisions:
+    """The component-layer ``apply_review_decisions`` function is the
+    actual semantic engine; unit-test it directly. The GUI just feeds
+    its output to a download button."""
+
+    def test_keep_all_means_no_rows_removed(
+        self, app_factory, small_csv_bytes,
+    ):
+        from src.gui.components import apply_review_decisions
+        from src.core import deduplicate
+        import io
+
+        df = pd.read_csv(io.BytesIO(small_csv_bytes), dtype=str, keep_default_na=False)
+        result = deduplicate(df, preview=True)
+        decisions = {
+            g.group_id: {
+                "keep_indices": list(g.row_indices),
+                "overrides": {},
+            }
+            for g in result.match_groups
+        }
+        deduped, removed = apply_review_decisions(df, result.match_groups, decisions)
+        assert len(deduped) == len(df), (
+            "Keep-All should preserve every row"
+        )
+        assert removed.empty
+
+    def test_merge_decision_drops_losers(
+        self, app_factory, small_csv_bytes,
+    ):
+        from src.gui.components import apply_review_decisions
+        from src.core import deduplicate
+        import io
+
+        df = pd.read_csv(io.BytesIO(small_csv_bytes), dtype=str, keep_default_na=False)
+        result = deduplicate(df, preview=True)
+        # Merge each group to its first member.
+        decisions = {
+            g.group_id: {
+                "keep_indices": [g.row_indices[0]],
+                "overrides": {},
+            }
+            for g in result.match_groups
+        }
+        deduped, removed = apply_review_decisions(df, result.match_groups, decisions)
+        expected_removed = sum(len(g.row_indices) - 1 for g in result.match_groups)
+        assert len(removed) == expected_removed
+        assert len(deduped) == len(df) - expected_removed
+
+    def test_column_override_applies_to_survivor(
+        self, app_factory, small_csv_bytes,
+    ):
+        from src.gui.components import apply_review_decisions
+        from src.core import deduplicate
+        import io
+
+        df = pd.read_csv(io.BytesIO(small_csv_bytes), dtype=str, keep_default_na=False)
+        result = deduplicate(df, preview=True)
+        group = result.match_groups[0]
+        survivor = group.row_indices[0]
+        decisions = {
+            group.group_id: {
+                "keep_indices": [survivor],
+                "overrides": {"phone": "OVERRIDE_VALUE"},
+            }
+        }
+        deduped, _ = apply_review_decisions(df, result.match_groups, decisions)
+        # The survivor row in ``deduped`` must carry the override. Find
+        # it via the original (non-loser) name.
+        match = deduped[deduped["phone"] == "OVERRIDE_VALUE"]
+        assert len(match) == 1, (
+            f"override didn't apply; deduped frame: {deduped.to_dict()}"
+        )
diff --git a/tests/gui/test_errors.py b/tests/gui/test_errors.py
new file mode 100644
index 0000000..22abde4
--- /dev/null
+++ b/tests/gui/test_errors.py
@@ -0,0 +1,103 @@
+"""Error-display tests.
+
+Tool pages catch core exceptions (via ``format_for_user``) and surface
+them through ``st.error``. We verify that the message structure makes
+it through the GUI layer, not just that it gets raised by core (the
+core tests already cover that).
+
+These tests deliberately feed garbage bytes / malformed content and
+check the rendered error, not just that the page didn't crash.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from .conftest import collected_text, stash_upload
+
+
+# ---------------------------------------------------------------------------
+# Malformed upload
+# ---------------------------------------------------------------------------
+
+class TestMalformedUploadErrors:
+    """Bytes that look like a CSV but aren't parseable. The Deduplicator
+    page wraps ``read_file`` failures in an ``st.error`` with the file
+    name and the structured ``format_for_user`` output."""
+
+    @pytest.fixture
+    def garbage_bytes(self) -> bytes:
+        """Binary garbage with embedded NULs and non-UTF-8 sequences —
+        triggers the gate's repair pipeline failures, ultimately
+        produces a parse error on the dedup page if it makes it that
+        far. We bypass the gate so the dedup page sees it raw."""
+        return b"\xff\xfe\x00\x01\x02garbage,without,structure\n\x00\xff" * 50
+
+    def test_garbage_bytes_do_not_crash_dedup(
+        self, app_factory, garbage_bytes,
+    ):
+        app = app_factory("1_Deduplicator")
+        stash_upload(app, name="garbage.csv", data=garbage_bytes)
+        app.run()
+        # The page should either render an error OR successfully parse
+        # the bytes as text (the gate has been pre-passed, so the
+        # pre-parse repair didn't run on this fixture). We just need
+        # no uncaught Python exception.
+        assert not app.exception
+
+
+# ---------------------------------------------------------------------------
+# Empty upload
+# ---------------------------------------------------------------------------
+
+class TestEmptyUpload:
+    """Zero-byte upload — must be handled gracefully."""
+
+    def test_empty_bytes_renders(self, app_factory):
+        app = app_factory("1_Deduplicator")
+        stash_upload(app, name="empty.csv", data=b"")
+        app.run()
+        # Either: (a) we render an error, or (b) we render the page
+        # with no preview. Either is acceptable — what's NOT is an
+        # uncaught Python exception bubbling up.
+        assert not app.exception
+
+
+# ---------------------------------------------------------------------------
+# Single-column file
+# ---------------------------------------------------------------------------
+
+class TestSingleColumnFile:
+    """A 1-column CSV is technically valid but produces no auto-detect
+    strategies. The page must explain this to the user rather than
+    silently producing zero match groups."""
+
+    def test_single_column_does_not_crash(self, app_factory):
+        app = app_factory("1_Deduplicator")
+        data = b"only_col\nvalue1\nvalue2\nvalue3\n"
+        stash_upload(app, name="single.csv", data=data)
+        app.run()
+        assert not app.exception
+
+
+# ---------------------------------------------------------------------------
+# Header collision in column_mapper
+# ---------------------------------------------------------------------------
+
+class TestColumnMapperDuplicateTarget:
+    """The column mapper rejects mappings where two source columns
+    point at the same target. This is surfaced as an error.
+
+    Test approach: ``map_columns`` validates upfront via core, and
+    raises ``InputValidationError`` — the GUI wraps it. We invoke the
+    core function directly to pin the validation contract."""
+
+    def test_duplicate_target_raises(self):
+        import pandas as pd
+        from src.core.column_mapper import map_columns, MapOptions
+        from src.core.errors import InputValidationError
+
+        df = pd.DataFrame({"a": [1, 2], "b": [3, 4]})
+        opts = MapOptions(mapping={"a": "name", "b": "name"})
+        with pytest.raises(InputValidationError):
+            map_columns(df, opts)
diff --git a/tests/gui/test_findings_panel.py b/tests/gui/test_findings_panel.py
new file mode 100644
index 0000000..7a82751
--- /dev/null
+++ b/tests/gui/test_findings_panel.py
@@ -0,0 +1,227 @@
+"""Findings panel rendering tests.
+
+``render_findings_panel`` is the central widget on the home page and
+the Review page; failures here cascade into the user's first
+impression. We drive it via a tiny test harness page
+(``_findings_panel_harness.py``) so the test can inject findings
+directly into session state — no file_uploader simulation needed.
+
+We verify:
+
+- Empty findings list → localized "no issues" success message.
+- Findings with tool ids → one expander per tool, labeled in the
+  active language.
+- Header + severity summary render at the top.
+- Untargeted findings land in the "Other / file-level" expander.
+
+Pack-key parity is already pinned by ``test_lang_packs.py``; this
+file pins the call sites instead.
+"""
+
+from __future__ import annotations
+
+from pathlib import Path
+
+import pandas as pd
+import pytest
+
+from streamlit.testing.v1 import AppTest
+
+from .conftest import PROJECT_ROOT, collected_text, with_language
+
+HARNESS_PATH = Path(__file__).resolve().parent / "_findings_panel_harness.py"
+
+
+def _harness(findings, lang: str = "en") -> AppTest:
+    """Build an AppTest of the harness page with ``findings`` pre-stashed."""
+    app = AppTest.from_file(str(HARNESS_PATH))
+    app.session_state["test_findings"] = findings
+    if lang != "en":
+        app.session_state["ui_lang"] = lang
+    return app
+
+
+def _make_finding(tool: str = "", **overrides):
+    """Build a minimal :class:`Finding` object. ``Finding`` is a frozen
+    dataclass; constructor signature is well-pinned by core tests, so
+    we use it directly here rather than building dicts."""
+    from src.core.analyze import Finding
+    kwargs = dict(
+        id="test_finding",
+        severity="warn",
+        tool=tool,
+        count=1,
+        description="A test finding.",
+        column=None,
+        samples=[],
+        confidence="medium",
+        fix_action="",
+    )
+    kwargs.update(overrides)
+    return Finding(**kwargs)
+
+
+# ---------------------------------------------------------------------------
+# Empty findings → success message
+# ---------------------------------------------------------------------------
+
+class TestEmptyFindings:
+    def test_empty_renders_no_issues_english(self):
+        app = _harness([])
+        app.run()
+        text = collected_text(app)
+        assert "No issues detected" in text
+
+    def test_empty_renders_no_issues_spanish(self):
+        app = _harness([], lang="es")
+        app.run()
+        text = collected_text(app)
+        assert "No se detectaron problemas" in text
+
+
+# ---------------------------------------------------------------------------
+# Header text
+# ---------------------------------------------------------------------------
+
+class TestHeader:
+    def test_header_english(self):
+        app = _harness([_make_finding(tool="02_text_cleaner")])
+        app.run()
+        text = collected_text(app)
+        assert "Detected issues" in text
+
+    def test_header_spanish(self):
+        app = _harness([_make_finding(tool="02_text_cleaner")], lang="es")
+        app.run()
+        text = collected_text(app)
+        assert "Problemas detectados" in text
+
+
+# ---------------------------------------------------------------------------
+# Per-tool grouping → one expander per tool id
+# ---------------------------------------------------------------------------
+
+class TestGrouping:
+    def test_findings_grouped_into_per_tool_expanders(self):
+        findings = [
+            _make_finding(tool="02_text_cleaner", id="whitespace_padding"),
+            _make_finding(tool="02_text_cleaner", id="nbsp_padding"),
+            _make_finding(tool="03_format_standardizer", id="mixed_case_email"),
+        ]
+        app = _harness(findings)
+        app.run()
+        labels = [e.label for e in app.expander]
+        # Two unique tools → two expanders. Each label carries the
+        # tool's display name + finding count.
+        text_cleaner_expanders = [lbl for lbl in labels if "Text Cleaner" in lbl]
+        format_expanders = [lbl for lbl in labels if "Format Standardizer" in lbl]
+        assert len(text_cleaner_expanders) == 1, (
+            f"expected one Text Cleaner expander; got: {labels}"
+        )
+        assert len(format_expanders) == 1, (
+            f"expected one Format Standardizer expander; got: {labels}"
+        )
+
+    def test_tool_names_localize_in_spanish(self):
+        findings = [_make_finding(tool="02_text_cleaner")]
+        app = _harness(findings, lang="es")
+        app.run()
+        labels = [e.label for e in app.expander]
+        assert any("Limpiador de texto" in lbl for lbl in labels), (
+            f"Spanish tool name missing; expanders: {labels}"
+        )
+
+    def test_finding_count_in_expander_label(self):
+        findings = [
+            _make_finding(tool="02_text_cleaner", id=f"f{i}")
+            for i in range(3)
+        ]
+        app = _harness(findings)
+        app.run()
+        labels = [e.label for e in app.expander]
+        # Pack template: "{tool} — {n} finding(s)"
+        text_cleaner_label = next(l for l in labels if "Text Cleaner" in l)
+        assert "3" in text_cleaner_label, (
+            f"expected count '3' in expander label; got {text_cleaner_label!r}"
+        )
+
+
+# ---------------------------------------------------------------------------
+# Open-tool button localizes
+# ---------------------------------------------------------------------------
+
+class TestOpenToolButton:
+    """Each tool section has an ``st.page_link`` to jump to that tool's
+    page. AppTest exposes page_links as ``app.button`` entries with
+    label ``"Open {tool} →"`` (English) / ``"Abrir {tool} →"`` (Spanish)."""
+
+    def test_open_tool_label_english(self):
+        findings = [_make_finding(tool="02_text_cleaner")]
+        app = _harness(findings)
+        app.run()
+        # ``st.page_link`` may show up under ``app.button`` or in the
+        # raw markdown. We probe both.
+        text = collected_text(app)
+        # Pack template: "Open {tool} →"
+        assert "Open Text Cleaner" in text
+
+    def test_open_tool_label_spanish(self):
+        findings = [_make_finding(tool="02_text_cleaner")]
+        app = _harness(findings, lang="es")
+        app.run()
+        text = collected_text(app)
+        # Pack template: "Abrir {tool} →"
+        assert "Abrir Limpiador de texto" in text
+
+
+# ---------------------------------------------------------------------------
+# Untargeted findings (file-level) go in the "Other" expander
+# ---------------------------------------------------------------------------
+
+class TestUntargetedFindings:
+    def test_untargeted_goes_to_other_expander_en(self):
+        findings = [
+            _make_finding(tool="", id="csv_bom_stripped"),
+            _make_finding(tool="02_text_cleaner", id="nbsp_padding"),
+        ]
+        app = _harness(findings)
+        app.run()
+        labels = [e.label for e in app.expander]
+        # Pack template: "Other / file-level — {n} finding(s)"
+        assert any("Other / file-level" in lbl for lbl in labels), (
+            f"untargeted expander missing; got: {labels}"
+        )
+
+    def test_untargeted_label_spanish(self):
+        findings = [_make_finding(tool="", id="csv_bom_stripped")]
+        app = _harness(findings, lang="es")
+        app.run()
+        labels = [e.label for e in app.expander]
+        # Spanish pack: "Otros / a nivel de archivo — {n} hallazgo(s)"
+        assert any("Otros / a nivel de archivo" in lbl for lbl in labels), (
+            f"Spanish 'Other' expander missing; got: {labels}"
+        )
+
+
+# ---------------------------------------------------------------------------
+# Severity summary
+# ---------------------------------------------------------------------------
+
+class TestSeveritySummary:
+    """The panel renders a per-severity summary caption like
+    ``⚠️ 2 warn · ℹ️ 1 info``. We pin the icon + count rendering."""
+
+    def test_severity_icons_render(self):
+        findings = [
+            _make_finding(tool="02_text_cleaner", severity="warn"),
+            _make_finding(tool="02_text_cleaner", severity="warn"),
+            _make_finding(tool="03_format_standardizer", severity="info"),
+        ]
+        app = _harness(findings)
+        app.run()
+        text = collected_text(app)
+        # Icons live in the per-language pack ("findings.severity_*").
+        # The summary template is shared between languages.
+        assert "⚠️" in text or "warn" in text
+        # Counts present.
+        assert "2 warn" in text or "2  warn" in text
diff --git a/tests/gui/test_gate.py b/tests/gui/test_gate.py
new file mode 100644
index 0000000..27cbe77
--- /dev/null
+++ b/tests/gui/test_gate.py
@@ -0,0 +1,157 @@
+"""Gate tests — ``require_normalization_gate()`` behaviour.
+
+The gate sits between every tool page and the user's data. Three states
+exist, each pinned here:
+
+1. **No upload** — gate is a no-op; the page proceeds and its own
+   uploader handles the file.
+2. **Upload but no normalization result** — gate shows a warning and a
+   "Go to Review & Normalize" button, then ``st.stop()`` short-circuits
+   the rest of the page.
+3. **Upload + matching passed normalization** — gate is a no-op; the
+   page proceeds.
+
+We exercise the gate via the Deduplicator page (any tool page would
+work; dedup is the smallest one that doesn't depend on heavy widgets).
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from .conftest import (
+    collected_text,
+    stash_upload,
+    stash_upload_without_gate,
+    with_language,
+)
+
+
+# Deduplicator is our canary — it calls ``require_normalization_gate``
+# on the second line of the module. If the gate blocks, the dedup-
+# specific title shouldn't even render.
+GATED_PAGE = "1_Deduplicator"
+
+
+class TestGateNoUpload:
+    """No upload → the gate exits early and the page renders normally,
+    showing its own file uploader. (This is the "user opened the dedup
+    page first instead of coming from home" path.)"""
+
+    def test_no_upload_lets_page_render(self, app_factory):
+        app = app_factory(GATED_PAGE)
+        app.run()
+        assert not app.exception
+        text = collected_text(app)
+        # The dedup page title is the unambiguous signal that the gate
+        # didn't short-circuit.
+        assert "Deduplicator" in text
+
+    def test_no_upload_no_gate_warning(self, app_factory):
+        app = app_factory(GATED_PAGE)
+        app.run()
+        # The gate's warning string starts with the upload filename. No
+        # warning should be present when there's no upload.
+        for w in app.warning:
+            assert "normalization gate" not in (w.body or "")
+
+
+class TestGateBlocksWithoutNormalization:
+    """Upload present but no passing normalization → gate fires:
+    warning + Go-to-Review button + page short-circuit."""
+
+    def test_gate_warning_renders(self, app_factory, small_csv_bytes):
+        app = app_factory(GATED_PAGE)
+        stash_upload_without_gate(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        warnings = [w.body for w in app.warning if w.body]
+        joined = " ".join(warnings)
+        assert "normalization gate" in joined, (
+            f"expected gate warning; got warnings: {warnings}"
+        )
+        assert "messy.csv" in joined, (
+            "gate warning should name the offending file"
+        )
+
+    def test_gate_renders_go_to_review_button(self, app_factory, small_csv_bytes):
+        app = app_factory(GATED_PAGE)
+        stash_upload_without_gate(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        labels = [b.label for b in app.button]
+        assert any("Review & Normalize" in lbl for lbl in labels), (
+            f"missing 'Go to Review & Normalize' button; got: {labels}"
+        )
+
+    def test_gate_short_circuits_page(self, app_factory, small_csv_bytes):
+        app = app_factory(GATED_PAGE)
+        stash_upload_without_gate(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        # When the gate fires it calls ``st.stop()`` after the warning.
+        # The page-body widgets (e.g., the advanced-options expander, the
+        # dedup-strategy widgets) must NOT be present.
+        labels = [b.label for b in app.button]
+        # The Run-Dedup primary action lives below the gate — make sure
+        # the gate killed the render before it.
+        assert not any("Run Deduplication" in lbl for lbl in labels), (
+            f"gate failed to short-circuit; saw button: {labels}"
+        )
+
+    def test_gate_warning_localizes_to_spanish(self, app_factory, small_csv_bytes):
+        app = app_factory(GATED_PAGE)
+        with_language(app, "es")
+        stash_upload_without_gate(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        warnings = " ".join(w.body for w in app.warning if w.body)
+        # Spanish pack: ``debe pasar la verificación de normalización CSV``.
+        assert "normalización" in warnings
+
+    def test_gate_button_localizes_to_spanish(self, app_factory, small_csv_bytes):
+        app = app_factory(GATED_PAGE)
+        with_language(app, "es")
+        stash_upload_without_gate(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        labels = [b.label for b in app.button]
+        assert any("Revisar y Normalizar" in lbl for lbl in labels), (
+            f"Spanish gate button missing; got: {labels}"
+        )
+
+
+class TestGateAllowsWithPassedNormalization:
+    """Upload + passed normalization → gate is a no-op and the page
+    renders past the gate."""
+
+    def test_passed_gate_lets_page_render(self, app_factory, small_csv_bytes):
+        app = app_factory(GATED_PAGE)
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        assert not app.exception, f"page raised past gate: {app.exception}"
+        # The pickup banner uses the upload name — that's our signal
+        # that the gate let us through AND the pickup helper engaged.
+        text = collected_text(app)
+        assert "messy.csv" in text
+
+
+class TestGateMismatchedHash:
+    """Upload changes (different bytes) but normalization_for still
+    points at the old hash → gate fires again because the result is
+    stale. Pins the security-relevant "stale fix doesn't carry over to
+    a new file" invariant."""
+
+    def test_stale_normalization_blocks_new_upload(self, app_factory, small_csv_bytes):
+        app = app_factory(GATED_PAGE)
+        # Stash bytes A but a normalization_for hash that points at B.
+        app.session_state["home_uploaded_bytes"] = small_csv_bytes
+        app.session_state["home_uploaded_name"] = "new.csv"
+        app.session_state["home_uploaded_size"] = len(small_csv_bytes)
+        app.session_state["normalization_for"] = "different-hash-from-an-old-upload"
+
+        # A passed-result object exists but is keyed to a different file.
+        class _Passed:
+            passed = True
+        app.session_state["normalization_result"] = _Passed()
+
+        app.run()
+        warnings = " ".join(w.body for w in app.warning if w.body)
+        assert "normalization gate" in warnings, (
+            "stale gate result should not unlock a new upload"
+        )
diff --git a/tests/gui/test_smoke.py b/tests/gui/test_smoke.py
new file mode 100644
index 0000000..0f6d79f
--- /dev/null
+++ b/tests/gui/test_smoke.py
@@ -0,0 +1,147 @@
+"""Smoke tests: every page renders without exception in EN and ES.
+
+The cheapest, highest-value GUI tests in the project. They catch:
+
+- Page-level Python errors (import failures, syntax errors that
+  ``ast.parse`` misses because they're runtime, e.g., a missing
+  attribute on a module).
+- i18n pack key drift (a string that used to render in EN now renders
+  literally as ``"chrome.language_label"`` because someone renamed the
+  key in en.json but forgot es.json or the call site).
+- Streamlit API churn that breaks ``set_page_config`` /
+  ``hide_streamlit_chrome`` on a single page.
+
+What they don't cover: user interactions. Those live in the workflow
+tests.
+"""
+
+from __future__ import annotations
+
+import pytest
+
+from .conftest import collected_text, with_language
+
+
+# Every page that ships in the sidebar nav. Slugs match the filenames
+# under ``src/gui/pages/`` so failures point at a real file.
+PAGE_SLUGS = [
+    "0_Review",
+    "1_Deduplicator",
+    "2_Text_Cleaner",
+    "3_Format_Standardizer",
+    "4_Missing_Values",
+    "5_Column_Mapper",
+    "6_Outlier_Detector",
+    "7_Multi_File_Merger",
+    "8_Validator_Reporter",
+    "9_Pipeline_Runner",
+    "99_Close",
+]
+
+
+# Substrings that must appear on each page for each language.
+#
+# v1.6 coverage reality (also documented in docs/USER-GUIDE.md §3.4):
+# only the home page, the Close page, and the shared chrome /
+# components ship Spanish strings. Per-tool page bodies are still
+# hard-coded English in both modes — translating them is tracked as a
+# follow-up. The substrings below reflect that reality: a page that
+# isn't translated yet asserts the same English substring under both
+# languages. The fact that the page *renders at all* in 'es' is still
+# the value of the smoke test.
+#
+# When a page gains real Spanish translation, flip its 'es' entry to
+# the localized substring — the test surface stays the same.
+EXPECTED_SUBSTRINGS: dict[str, dict[str, str]] = {
+    "0_Review":              {"en": "Review",         "es": "Review"},
+    "1_Deduplicator":        {"en": "Deduplicator",   "es": "Deduplicator"},
+    "2_Text_Cleaner":        {"en": "Text Cleaner",   "es": "Text Cleaner"},
+    "3_Format_Standardizer": {"en": "Format",         "es": "Format"},
+    "4_Missing_Values":      {"en": "Missing",        "es": "Missing"},
+    "5_Column_Mapper":       {"en": "Column",         "es": "Column"},
+    "6_Outlier_Detector":    {"en": "Outlier",        "es": "Outlier"},
+    "7_Multi_File_Merger":   {"en": "Merger",         "es": "Merger"},
+    "8_Validator_Reporter":  {"en": "Validator",      "es": "Validator"},
+    "9_Pipeline_Runner":     {"en": "Pipeline",       "es": "Pipeline"},
+    "99_Close":              {"en": "Close DataTools", "es": "Cerrar DataTools"},
+}
+
+
+class TestHomePageRenders:
+    """The home page is the only one with full EN/ES coverage in v1.6.
+    Pin it independently so its translation is non-regressable."""
+
+    @pytest.mark.parametrize("lang,expected", [
+        ("en", "DataTools — Data Cleaning Mastery"),
+        ("es", "DataTools — Maestría en limpieza de datos"),
+    ])
+    def test_home_renders_in_language(self, home_app, lang, expected):
+        with_language(home_app, lang)
+        home_app.run()
+        assert home_app.exception is None or home_app.exception == [], (
+            f"home page raised: {home_app.exception}"
+        )
+        assert expected in collected_text(home_app)
+
+    def test_home_renders_footer_in_es(self, home_app):
+        with_language(home_app, "es")
+        home_app.run()
+        text = collected_text(home_app)
+        assert "Tus datos nunca salen" in text or "Se ejecuta localmente" in text
+
+    def test_home_tool_card_uses_es_name(self, home_app):
+        """When the home grid renders in Spanish, the dedup card title
+        must use the Spanish display name, not the English fallback."""
+        with_language(home_app, "es")
+        home_app.run()
+        text = collected_text(home_app)
+        assert "Eliminador de duplicados" in text
+
+
+class TestEveryPageRenders:
+    """Parametrize over (page, language). Failure tells you exactly which
+    page + which language broke."""
+
+    @pytest.mark.parametrize("slug", PAGE_SLUGS)
+    @pytest.mark.parametrize("lang", ["en", "es"])
+    def test_renders_without_exception(self, app_factory, slug, lang):
+        app = app_factory(slug)
+        with_language(app, lang)
+        app.run()
+        # AppTest exposes ``exception`` as a list of element-wrapped
+        # exceptions (empty when no error fired).
+        assert not app.exception, (
+            f"page {slug!r} raised in language {lang!r}: {app.exception}"
+        )
+
+    @pytest.mark.parametrize("slug", PAGE_SLUGS)
+    @pytest.mark.parametrize("lang", ["en", "es"])
+    def test_expected_substring_present(self, app_factory, slug, lang):
+        app = app_factory(slug)
+        with_language(app, lang)
+        app.run()
+        needle = EXPECTED_SUBSTRINGS[slug][lang]
+        text = collected_text(app)
+        assert needle in text, (
+            f"page {slug!r} ({lang!r}) missing expected substring "
+            f"{needle!r}\nGot:\n{text[:500]}…"
+        )
+
+
+class TestPageHasLanguageSelector:
+    """Every page that calls ``hide_streamlit_chrome`` should mount the
+    sidebar language selector. This is the only place the picker is
+    rendered — if the chrome helper stops calling it, the test fails."""
+
+    @pytest.mark.parametrize("slug", PAGE_SLUGS)
+    def test_sidebar_selectbox_present(self, app_factory, slug):
+        app = app_factory(slug)
+        app.run()
+        # The selector is the only sidebar selectbox we ship today; if
+        # a page adds another the test should be loosened to "at least
+        # one selectbox with the language label."
+        assert len(app.sidebar.selectbox) >= 1, (
+            f"page {slug!r} has no sidebar selectbox — "
+            f"hide_streamlit_chrome() should have mounted the language "
+            f"selector."
+        )
diff --git a/tests/gui/test_workflows.py b/tests/gui/test_workflows.py
new file mode 100644
index 0000000..a67c41a
--- /dev/null
+++ b/tests/gui/test_workflows.py
@@ -0,0 +1,207 @@
+"""Happy-path workflow tests for each Ready tool page.
+
+These drive the GUI like a user would: pre-stash an upload + a passed
+gate, render the page, click the primary action, assert the result
+landed in session state. They catch wiring bugs that smoke tests
+can't see — e.g., a primary button mis-keyed, a result not stashed in
+session state, a page reading the wrong key.
+
+Slow-ish (~0.5–2s per workflow). Sits behind the ``gui`` marker so
+``pytest -m 'not gui'`` skips them.
+"""
+
+from __future__ import annotations
+
+import pandas as pd
+import pytest
+
+from .conftest import collected_text, stash_upload
+
+
+# ---------------------------------------------------------------------------
+# Deduplicator
+# ---------------------------------------------------------------------------
+
+class TestDeduplicatorWorkflow:
+    """Upload → click Find Duplicates → result lands in session_state."""
+
+    def _setup(self, app_factory, small_csv_bytes):
+        app = app_factory("1_Deduplicator")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        return app
+
+    def test_upload_renders_preview(self, app_factory, small_csv_bytes):
+        app = self._setup(app_factory, small_csv_bytes)
+        app.run()
+        text = collected_text(app)
+        assert "Preview: messy.csv" in text, (
+            f"upload preview header missing; got:\n{text[:500]}"
+        )
+
+    def test_find_duplicates_button_present(self, app_factory, small_csv_bytes):
+        app = self._setup(app_factory, small_csv_bytes)
+        app.run()
+        labels = [b.label for b in app.button]
+        assert any("Find Duplicates" in lbl for lbl in labels), (
+            f"primary action missing; got: {labels}"
+        )
+
+    def test_clicking_find_duplicates_stashes_result(
+        self, app_factory, small_csv_bytes,
+    ):
+        app = self._setup(app_factory, small_csv_bytes)
+        app.run()
+        # Find the Find-Duplicates button and click it. AppTest's
+        # button-by-key access is via ``.button(key=...)`` — we don't
+        # have the key here, so locate it by label.
+        target = next(b for b in app.button if "Find Duplicates" in b.label)
+        target.click().run()
+        # The page stores the result under ``result`` in session state.
+        result = app.session_state["result"]
+        assert result is not None, "Find Duplicates didn't stash a result"
+        # The sample has Alice twice → one match group.
+        assert len(result.match_groups) >= 1
+
+
+# ---------------------------------------------------------------------------
+# Text Cleaner
+# ---------------------------------------------------------------------------
+
+class TestTextCleanerWorkflow:
+    def _setup(self, app_factory, small_csv_bytes):
+        app = app_factory("2_Text_Cleaner")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        return app
+
+    def test_page_renders_with_upload(self, app_factory, small_csv_bytes):
+        app = self._setup(app_factory, small_csv_bytes)
+        app.run()
+        assert not app.exception
+        text = collected_text(app)
+        assert "Text Cleaner" in text
+
+    def test_preview_or_clean_button_present(self, app_factory, small_csv_bytes):
+        """The text cleaner ships a primary action (label varies by
+        version). We just assert at least one primary-looking button
+        exists past the upload."""
+        app = self._setup(app_factory, small_csv_bytes)
+        app.run()
+        # Filter out the gate-redirect button (which would only be
+        # present if the gate fired, which our setup prevents).
+        gate_buttons = {"Go to Review & Normalize", "Ir a Revisar y Normalizar"}
+        non_gate = [b for b in app.button if b.label not in gate_buttons]
+        assert non_gate, (
+            f"no primary buttons rendered; got: {[b.label for b in app.button]}"
+        )
+
+
+# ---------------------------------------------------------------------------
+# Format Standardizer
+# ---------------------------------------------------------------------------
+
+class TestFormatStandardizerWorkflow:
+    def test_page_renders_with_upload(self, app_factory, small_csv_bytes):
+        app = app_factory("3_Format_Standardizer")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        assert not app.exception
+        text = collected_text(app)
+        assert "Format Standardizer" in text
+
+
+# ---------------------------------------------------------------------------
+# Missing Value Handler
+# ---------------------------------------------------------------------------
+
+class TestMissingValuesWorkflow:
+    def test_page_renders_with_upload(self, app_factory, small_csv_bytes):
+        app = app_factory("4_Missing_Values")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        assert not app.exception
+        text = collected_text(app)
+        assert "Missing" in text
+
+
+# ---------------------------------------------------------------------------
+# Column Mapper
+# ---------------------------------------------------------------------------
+
+class TestColumnMapperWorkflow:
+    def test_page_renders_with_upload(self, app_factory, small_csv_bytes):
+        app = app_factory("5_Column_Mapper")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        assert not app.exception
+        text = collected_text(app)
+        assert "Column" in text
+
+
+# ---------------------------------------------------------------------------
+# Pipeline Runner
+# ---------------------------------------------------------------------------
+
+class TestPipelineRunnerWorkflow:
+    def test_page_renders_with_upload(self, app_factory, small_csv_bytes):
+        app = app_factory("9_Pipeline_Runner")
+        stash_upload(app, name="messy.csv", data=small_csv_bytes)
+        app.run()
+        assert not app.exception
+        text = collected_text(app)
+        assert "Pipeline" in text
+
+
+# ---------------------------------------------------------------------------
+# Review page — special: doesn't gate on upload, has its own analyzer flow
+# ---------------------------------------------------------------------------
+
+class TestReviewWorkflow:
+    """The Review page is the gate-fixer. Without an upload it shows a
+    'go back to home' message. With an upload it runs the analyzer and
+    shows findings."""
+
+    def test_no_upload_shows_back_to_home(self, app_factory):
+        app = app_factory("0_Review")
+        app.run()
+        text = collected_text(app)
+        # Page shows ``No file uploaded`` + ``Back to home``.
+        assert "No file uploaded" in text or "uploaded" in text.lower()
+
+    def test_with_upload_shows_review_content(
+        self, app_factory, small_csv_bytes,
+    ):
+        app = app_factory("0_Review")
+        # Review page only needs the upload bytes, not a pre-passed gate.
+        app.session_state["home_uploaded_bytes"] = small_csv_bytes
+        app.session_state["home_uploaded_name"] = "messy.csv"
+        app.session_state["home_uploaded_size"] = len(small_csv_bytes)
+        app.run()
+        assert not app.exception
+        text = collected_text(app)
+        # Page ran the analyzer — either we get findings or the
+        # "already clean" success message. Either way confirms the
+        # analyzer pipeline ran end-to-end with the stashed bytes.
+        clean_msg = "No findings to review" in text
+        encoding_section = "File encoding" in text
+        assert clean_msg or encoding_section, (
+            f"Review page didn't surface analyzer output; got:\n{text[:400]}"
+        )
+
+
+# ---------------------------------------------------------------------------
+# Coming-Soon pages still render (just a stub) — pinned so we know if a
+# Coming-Soon goes from "stub renders" to "import error".
+# ---------------------------------------------------------------------------
+
+@pytest.mark.parametrize("slug,name", [
+    ("6_Outlier_Detector", "Outlier"),
+    ("7_Multi_File_Merger", "Merger"),
+    ("8_Validator_Reporter", "Validator"),
+])
+class TestComingSoonStubs:
+    def test_stub_renders(self, app_factory, slug, name):
+        app = app_factory(slug)
+        app.run()
+        assert not app.exception
+        text = collected_text(app)
+        assert name in text