diff --git a/internal/i18n/untranslated_test.go b/internal/i18n/untranslated_test.go
new file mode 100644
index 0000000..c68d2a4
--- /dev/null
+++ b/internal/i18n/untranslated_test.go
@@ -0,0 +1,70 @@
+package i18n
+
+import (
+	"os"
+	"path/filepath"
+	"regexp"
+	"strings"
+	"testing"
+)
+
+// Danish prose left in a component is the failure the dictionaries cannot see.
+//
+// `i18n_test.go` checks that the two dictionaries agree. It cannot check the thing
+// that actually went wrong: a string that was never put in either, because nobody
+// noticed it. Three survived the sweep — "Ingen nyere version fundet.", "Ikke mere
+// i dag." and a token hint split across two lines — and they survived it for the
+// same reason. The sweep grepped for æøå, and none of them contain one.
+//
+// So this reads the templates instead and looks for the words that give Danish
+// away whatever letters it happens to use. A word list rather than a language
+// detector: this repository ships one static binary, and a list of thirty common
+// words catches the sentences a person would actually write.
+func TestNoDanishProseOutsideTheDictionaries(t *testing.T) {
+	// Text nodes of three characters or more, and the three attributes a person
+	// reads. Anything inside {…} is already an expression — usually a t() call —
+	// and is not what this is looking for.
+	text := regexp.MustCompile(`>\s*([^<>{}\n][^<>{}]{3,})\s*<`)
+	attrs := regexp.MustCompile(`(?:aria-label|title|placeholder)="([^"{][^"]{3,})"`)
+	// Words that are Danish and are not also English. "is", "for" and "to" would
+	// match English prose and are deliberately absent.
+	danish := regexp.MustCompile(`(?i)\b(og|er|ikke|det|den|der|til|kan|som|har|med|af|en|et|du|din|dine|vises|ingen|nyere|fundet|ude|valgt|slettet|gemt|opgave|projekt|indstillinger|fjern|opret|vælg|luk|tilføj|ryd|gem|slet|omdøb)\b`)
+	comments := regexp.MustCompile(`(?s)<!--.*?-->`)
+	styles := regexp.MustCompile(`(?s)<style>.*`)
+
+	var found []string
+	root := filepath.Join("..", "..", "web", "src")
+	err := filepath.Walk(root, func(path string, info os.FileInfo, err error) error {
+		if err != nil || info.IsDir() || !strings.HasSuffix(path, ".svelte") {
+			return err
+		}
+		body, err := os.ReadFile(path)
+		if err != nil {
+			return err
+		}
+		markup := string(body)
+		if at := strings.Index(markup, "</script>"); at > 0 {
+			markup = markup[at:]
+		}
+		markup = styles.ReplaceAllString(comments.ReplaceAllString(markup, ""), "")
+
+		rel, _ := filepath.Rel(root, path)
+		for _, m := range append(text.FindAllStringSubmatch(markup, -1),
+			attrs.FindAllStringSubmatch(markup, -1)...) {
+			phrase := strings.TrimSpace(m[1])
+			if danish.MatchString(phrase) {
+				found = append(found, rel+": "+strings.Join(strings.Fields(phrase), " "))
+			}
+		}
+		return nil
+	})
+	if err != nil {
+		t.Fatalf("walk %s: %v", root, err)
+	}
+
+	if len(found) > 0 {
+		t.Errorf("%d Danish string(s) are written into a component instead of going through "+
+			"t(). An English interface shows them in Danish, and neither dictionary knows "+
+			"they exist:\n  %s", len(found), strings.Join(found, "\n  "))
+	}
+}