diff --git a/controllers/reviews.go b/controllers/reviews.go
index a400b0f..88d8df4 100644
--- a/controllers/reviews.go
+++ b/controllers/reviews.go
@@ -8,6 +8,7 @@ import (
 	"strconv"
 	"strings"
 	"time"
+	"unicode/utf8"
 
 	"git.golem15.com/golem15/sm-quizzes-plugin/classes"
 	"git.golem15.com/golem15/sm-quizzes-plugin/graders"
@@ -421,6 +422,61 @@ type reviewContextView struct {
 	ReviewComment   string
 	Reviewed        bool
 	ReviewedAt      string
+	// TextCut and Cut tell that the student's text, or another long text,
+	// was shortened to fit cabana's partial size cap (see fitEscaped).
+	TextCut bool
+	Cut     bool
+}
+
+// Escaped-byte budgets of the partial's long texts. cabana refuses a
+// partial that renders to more than 64 KiB, and html/template turns one
+// character into up to 5 bytes, so each text is cut to its budget measured
+// as escaped (fitEscaped). Their sum leaves room for the markup and the
+// translated labels; a long but ordinary answer (max_length up to 20000)
+// fits whole, only pathological texts are cut.
+const (
+	budgetText         = 32 << 10
+	budgetQuestionText = 4 << 10
+	budgetModelAnswer  = 6 << 10
+	budgetRubric       = 6 << 10
+	budgetFeedback     = 4 << 10
+	budgetComment      = 4 << 10
+	budgetError        = 1 << 10
+	budgetShort        = 512
+)
+
+// fitEscaped cuts s so that it takes at most budget bytes once escaped by
+// html/template, ending it with "…" when cut, and reports whether it cut.
+func fitEscaped(s string, budget int) (string, bool) {
+	total := 0
+	for _, r := range s {
+		total += escapedLen(r)
+	}
+	if total <= budget {
+		return s, false
+	}
+	used := 0
+	for i, r := range s {
+		used += escapedLen(r)
+		if used > budget-len("…") {
+			return s[:i] + "…", true
+		}
+	}
+	return s, false
+}
+
+// escapedLen is the most bytes html/template writes for r in text.
+func escapedLen(r rune) int {
+	switch r {
+	case '"', '&', '\'', '+', '<', '>':
+		return 5
+	case 0:
+		return len("\uFFFD")
+	}
+	if n := utf8.RuneLen(r); n > 0 {
+		return n
+	}
+	return 3
 }
 
 // PartialData builds the review context of an answer from the review
@@ -445,7 +501,8 @@ func (reviewsAdmin) PartialData(ctx context.Context, name string, record any) (a
 }
 
 // newReviewContextView copies a review detail into the partial's view
-// model.
+// model, each long text cut to its escaped-byte budget so the partial
+// always fits cabana's size cap.
 func newReviewContextView(item *classes.ReviewDetail, reg *graders.Registry) reviewContextView {
 	v := reviewContextView{
 		QuizTitle:       item.QuizTitle,
@@ -473,6 +530,25 @@ func newReviewContextView(item *classes.ReviewDetail, reg *graders.Registry) rev
 		v.HasGraderPoints, v.GraderPoints = true, *item.GraderPoints
 	}
 	v.GraderLabel = graderLabel(reg, v.Grader)
+	v.Text, v.TextCut = fitEscaped(v.Text, budgetText)
+	for _, f := range []struct {
+		s      *string
+		budget int
+	}{
+		{&v.QuizTitle, budgetShort},
+		{&v.StudentEmail, budgetShort},
+		{&v.StudentName, budgetShort},
+		{&v.QuestionText, budgetQuestionText},
+		{&v.ModelAnswer, budgetModelAnswer},
+		{&v.Rubric, budgetRubric},
+		{&v.GraderFeedback, budgetFeedback},
+		{&v.GradingError, budgetError},
+		{&v.ReviewComment, budgetComment},
+	} {
+		var cut bool
+		*f.s, cut = fitEscaped(*f.s, f.budget)
+		v.Cut = v.Cut || cut
+	}
 	return v
 }
 
diff --git a/controllers/reviews/_review_context.htm b/controllers/reviews/_review_context.htm
index 8cd13c2..00d4e9d 100644
--- a/controllers/reviews/_review_context.htm
+++ b/controllers/reviews/_review_context.htm
@@ -1,5 +1,8 @@
 <div class="quizzes-review">
 {{with .Data}}
+{{if .Cut}}
+    <p class="help-block">{{trans "golem15.quizzes::lang.reviews.cut"}}</p>
+{{end}}
     <dl>
         <dt>{{trans "golem15.quizzes::lang.reviews.quiz"}}</dt>
         <dd>{{.QuizTitle}}</dd>
@@ -16,7 +19,7 @@
         <dd><p class="whitespace-pre-wrap">{{.Rubric}}</p></dd>
 {{end}}
         <dt>{{trans "golem15.quizzes::lang.reviews.text"}} <span class="help-block">{{trans "golem15.quizzes::lang.reviews.max_length"}}: {{.MaxLength}}</span></dt>
-        <dd><blockquote class="whitespace-pre-wrap">{{.Text}}</blockquote></dd>
+        <dd><blockquote class="whitespace-pre-wrap">{{.Text}}</blockquote>{{if .TextCut}}<p class="help-block">{{trans "golem15.quizzes::lang.reviews.text_cut"}}</p>{{end}}</dd>
         <dt>{{trans "golem15.quizzes::lang.reviews.status"}}</dt>
         <dd><strong>{{trans .StatusLabel}}</strong></dd>
 {{if .Grader}}
diff --git a/controllers/reviews_test.go b/controllers/reviews_test.go
index d0fc518..e00e9c2 100644
--- a/controllers/reviews_test.go
+++ b/controllers/reviews_test.go
@@ -209,3 +209,58 @@ func TestReviewContextEscapes(t *testing.T) {
 		t.Errorf("student text not escaped: %s", html)
 	}
 }
+
+func TestFitEscaped(t *testing.T) {
+	if s, cut := fitEscaped("zażółć", 12); s != "zażółć" || cut {
+		t.Errorf("a text within budget = %q, %v", s, cut)
+	}
+	s, cut := fitEscaped(strings.Repeat(`"`, 10), 23)
+	if !cut || s != `""""`+"…" {
+		t.Errorf("quotes cut = %q, %v; want four quotes and an ellipsis", s, cut)
+	}
+	if s, cut := fitEscaped("abc", 0); !cut || s != "…" {
+		t.Errorf("zero budget = %q, %v", s, cut)
+	}
+}
+
+// TestReviewContextFitsPartialCap renders the partial with every text at
+// its worst case (characters html/template turns into 5 bytes) and checks
+// the output stays under cabana's 64 KiB partial cap.
+func TestReviewContextFitsPartialCap(t *testing.T) {
+	src, err := os.ReadFile("reviews/_review_context.htm")
+	if err != nil {
+		t.Fatal(err)
+	}
+	// A translation longer than any real label.
+	long := strings.Repeat("ą", 100)
+	tpl, err := template.New("review_context").Funcs(template.FuncMap{"trans": func(string) string { return long }}).Parse(string(src))
+	if err != nil {
+		t.Fatalf("parse: %v", err)
+	}
+	worst := func(n int) *string { s := strings.Repeat(`"&'`, n); return &s }
+	grader := "ai"
+	gp := 1.5
+	v := newReviewContextView(&classes.ReviewDetail{
+		ReviewRow: classes.ReviewRow{
+			Status: "reviewed", QuizTitle: *worst(300), Email: *worst(300), Name: worst(300), Surname: worst(300),
+			QuestionText: *worst(10000), Text: *worst(20000), GraderPoints: &gp,
+		},
+		AnswerText: worst(10000), Rubric: worst(10000), Grader: &grader,
+		GraderFeedback: worst(2000), GradingError: worst(500), ReviewComment: worst(2000),
+	}, nil)
+	if !v.TextCut || !v.Cut {
+		t.Fatalf("worst-case texts not cut: TextCut=%v Cut=%v", v.TextCut, v.Cut)
+	}
+	var out bytes.Buffer
+	if err := tpl.Execute(&out, map[string]any{"Data": v}); err != nil {
+		t.Fatalf("execute: %v", err)
+	}
+	if n := out.Len(); n > 64<<10 {
+		t.Errorf("partial renders to %d bytes, over the 64 KiB cap", n)
+	}
+
+	plain := strings.Repeat("Długa, zwykła odpowiedź studenta. ", 580)
+	if v := newReviewContextView(&classes.ReviewDetail{ReviewRow: classes.ReviewRow{Text: plain}}, nil); v.TextCut || v.Text != plain {
+		t.Error("an ordinary answer near 20000 characters was cut")
+	}
+}
diff --git a/lang/en/lang.yaml b/lang/en/lang.yaml
index f03319a..599265a 100644
--- a/lang/en/lang.yaml
+++ b/lang/en/lang.yaml
@@ -129,5 +129,7 @@ reviews:
     already_reviewed: This answer is already verified.
     grader_manual: This quiz is graded manually; nothing can grade the answer again.
     attempt_open: The test is still in progress; review this answer once it ends.
+    text_cut: The answer is too long to show here in full; the quizzes_review_show tool shows all of it.
+    cut: Some long texts are shortened on this screen; the quizzes_review_show tool shows them in full.
     not_found: This answer was not found.
     not_created: Open answers come from students; they cannot be added or deleted here.
diff --git a/lang/pl/lang.yaml b/lang/pl/lang.yaml
index a467153..c56695f 100644
--- a/lang/pl/lang.yaml
+++ b/lang/pl/lang.yaml
@@ -129,5 +129,7 @@ reviews:
     already_reviewed: Ta odpowiedź jest już zweryfikowana.
     grader_manual: Ten quiz jest oceniany ręcznie; nie ma kto ocenić odpowiedzi ponownie.
     attempt_open: Test wciąż trwa; oceń tę odpowiedź po jego zakończeniu.
+    text_cut: Odpowiedź jest zbyt długa, by pokazać ją tu w całości; całą pokazuje narzędzie quizzes_review_show.
+    cut: Część długich tekstów jest tu skrócona; w całości pokazuje je narzędzie quizzes_review_show.
     not_found: Nie znaleziono tej odpowiedzi.
     not_created: Odpowiedzi otwarte pochodzą od studentów; tutaj nie można ich dodawać ani usuwać.