diff --git a/shelfmark/release_sources/direct_download/annas_archive.py b/shelfmark/release_sources/direct_download/annas_archive.py
index c3855c2..6978663 100644
--- a/shelfmark/release_sources/direct_download/annas_archive.py
+++ b/shelfmark/release_sources/direct_download/annas_archive.py
@@ -768,6 +768,9 @@ def search_books(query: str, filters: SearchFilters) -> list[BrowseRecord]:
         provider_id="annas_archive",
         item_selector="tr",
         extract_item=_extract_aa_search_result,
+        # AA already applied &lang= server-side; only the path-language pass below
+        # (which skips &lang=) needs a local language filter.
+        filter_languages=False,
     )
 
     if path_language_enabled and requested_langs:
diff --git a/shelfmark/release_sources/direct_download/common.py b/shelfmark/release_sources/direct_download/common.py
index e134894..f68d77e 100644
--- a/shelfmark/release_sources/direct_download/common.py
+++ b/shelfmark/release_sources/direct_download/common.py
@@ -188,13 +188,19 @@ def parse_search_items(
     *,
     provider_id: str,
     extract_item: Callable[[Tag], ParsedSearchResult | None],
+    filter_languages: bool = True,
 ) -> list[BrowseRecord]:
     """Normalize provider-specific HTML elements into Direct Download records.
 
     Providers only describe how fields are extracted from their DOM. Language and
     format filtering, stable IDs, and BrowseRecord construction stay shared.
+
+    Pass ``filter_languages=False`` when the site already filtered by language: its
+    language cells are free text, and re-matching them locally drops rows it matched.
     """
-    requested_languages = normalize_requested_languages(filters.lang) if filters else set()
+    requested_languages = (
+        normalize_requested_languages(filters.lang) if filters and filter_languages else set()
+    )
     requested_formats = (
         {value.casefold() for value in (filters.format or get_supported_formats())}
         if filters
@@ -253,6 +259,7 @@ def parse_search_page(
     provider_id: str,
     item_selector: str,
     extract_item: Callable[[Tag], ParsedSearchResult | None],
+    filter_languages: bool = True,
 ) -> list[BrowseRecord]:
     """Parse a result page using provider-specific selectors and extraction."""
     root = BeautifulSoup(page, "html.parser") if isinstance(page, str) else page
@@ -261,4 +268,5 @@ def parse_search_page(
         filters,
         provider_id=provider_id,
         extract_item=extract_item,
+        filter_languages=filter_languages,
     )
diff --git a/shelfmark/release_sources/direct_download/source.py b/shelfmark/release_sources/direct_download/source.py
index 9a23bfd..022cd14 100644
--- a/shelfmark/release_sources/direct_download/source.py
+++ b/shelfmark/release_sources/direct_download/source.py
@@ -173,7 +173,7 @@ class DirectDownloadSource(ReleaseSource):
             raise DirectDownloadUnavailableError(unavailable_reason)
 
         releases: list[Release] = []
-        unavailable_errors: list[SourceUnavailableError] = []
+        failures: list[Exception] = []
         for provider in registry.enabled_providers(self._providers):
             try:
                 records = provider.search(
@@ -183,7 +183,7 @@ class DirectDownloadSource(ReleaseSource):
                     content_type=content_type,
                 )
             except SourceUnavailableError as exc:
-                unavailable_errors.append(exc)
+                failures.append(exc)
                 continue
             except (
                 RuntimeError,
@@ -192,11 +192,14 @@ class DirectDownloadSource(ReleaseSource):
                 requests.exceptions.RequestException,
             ) as exc:
                 logger.warning("%s search failed: %s", provider.display_name, exc)
+                failures.append(exc)
                 continue
             releases.extend(_browse_record_to_release(record) for record in records)
 
-        if unavailable_errors and not releases:
-            raise unavailable_errors[0]
+        # A provider failure is only quiet when another provider answered. Otherwise the
+        # caller has to see it, or a failed search reads as a search with no hits.
+        if failures and not releases:
+            raise failures[0]
         return releases
 
     def is_available(self) -> bool:
diff --git a/tests/direct_download/test_search_queries.py b/tests/direct_download/test_search_queries.py
index fef3fcd..0a7ad3c 100644
--- a/tests/direct_download/test_search_queries.py
+++ b/tests/direct_download/test_search_queries.py
@@ -350,3 +350,38 @@ def test_book_matches_requested_languages_logic():
     assert aa._book_matches_requested_languages(None, set()) is True
     assert aa._book_matches_requested_languages("en", {"fr"}) is False
     assert aa._book_matches_requested_languages("fr", {"fr"}) is True
+
+
+def test_search_books_keeps_server_language_matches_when_path_language_disabled(monkeypatch):
+    monkeypatch.setattr(aa, "_is_language_from_path_enabled", lambda: False)
+    monkeypatch.setattr(aa.network, "get_aa_base_url", lambda: "https://mirror.example")
+    monkeypatch.setattr(aa.network, "AAMirrorSelector", lambda: object())
+
+    captured_url: dict[str, str] = {}
+
+    def _row(record_id: str, language: str) -> str:
+        return f"""
+            <tr>
+                <td><a href="/md5/{record_id}"><img src="c.jpg"></a></td>
+                <td><span>Book {record_id}</span></td><td><span>Author</span></td>
+                <td><span>Publisher</span></td><td><span>2025</span></td>
+                <td><span>-</span></td><td><span>-</span></td>
+                <td><span>{language}</span></td>
+                <td><span>fiction</span></td><td><span>pdf</span></td>
+                <td><span>2 mb</span></td>
+            </tr>
+        """
+
+    def _fake_html_get_page(url: str, selector, **_kwargs):
+        del selector
+        captured_url["url"] = url
+        rows = _row("rec-multi", "English, French") + _row("rec-unknown", "unknown")
+        return f"<table>{rows}</table>"
+
+    monkeypatch.setattr(aa.downloader, "html_get_page", _fake_html_get_page)
+
+    records = aa.search_books("demo", SearchFilters(lang=["en"], format=["pdf"]))
+
+    # AA already narrowed by &lang=; its free-text language cells must not be re-matched.
+    assert "&lang=en" in captured_url["url"]
+    assert [record.id for record in records] == ["rec-multi", "rec-unknown"]
diff --git a/tests/direct_download/test_web_providers.py b/tests/direct_download/test_web_providers.py
index 93a7754..c2e16af 100644
--- a/tests/direct_download/test_web_providers.py
+++ b/tests/direct_download/test_web_providers.py
@@ -183,3 +183,28 @@ def test_provider_failure_is_suppressed_when_another_provider_succeeds(monkeypat
     releases = source.search(SimpleNamespace(title="Example"), SimpleNamespace())
 
     assert [release.source_id for release in releases] == ["working:1"]
+
+
+def test_provider_search_error_surfaces_when_no_provider_answers(monkeypatch):
+    from shelfmark.release_sources.direct_download.source import DirectDownloadSource
+
+    class FailingProvider:
+        id = "failing"
+        display_name = "Failing"
+
+        def is_enabled(self):
+            return True
+
+        def search(self, *_args, **_kwargs):
+            raise RuntimeError("No books found. Please try another query.")
+
+    monkeypatch.setattr(
+        registry.config,
+        "get",
+        lambda key, default=None: True if key == "DIRECT_DOWNLOAD_ENABLED" else default,
+    )
+    source = DirectDownloadSource()
+    source._providers = (FailingProvider(),)
+
+    with pytest.raises(RuntimeError, match="No books found"):
+        source.search(SimpleNamespace(title="Example"), SimpleNamespace())