diff --git a/classes/review.go b/classes/review.go
index 5f3e5ac..601ca04 100644
--- a/classes/review.go
+++ b/classes/review.go
@@ -29,7 +29,8 @@ import (
 // Review status filters besides the grading statuses themselves.
 const (
 	// ReviewStatusOpen is every answer the owner has not reviewed yet:
-	// pending, graded and failed. It is the default filter.
+	// pending, graded and failed, except those of a test attempt that is
+	// still running. It is the default filter.
 	ReviewStatusOpen = "open"
 	// ReviewStatusAll lists every open answer, reviewed ones included.
 	ReviewStatusAll = "all"
@@ -52,6 +53,10 @@ var (
 	// ErrGraderManual is a regrade on a quiz whose grader is manual or not
 	// registered by any activated plugin.
 	ErrGraderManual = errors.New("quizzes: the quiz has no grader that can grade again")
+	// ErrAttemptOpen is a review write on an answer of a test attempt that
+	// is still running: a test's results, the owner's review included, come
+	// only after it ends (Phase 7 D-07, D-08).
+	ErrAttemptOpen = errors.New("quizzes: the test attempt is still open")
 )
 
 // ReviewFilter narrows the review queue. Zero values do not filter; Status
@@ -287,6 +292,8 @@ func (r reviewScan) detail() *ReviewDetail {
 }
 
 // reviewWhere turns a filter into predicates appended to reviewFromSQL.
+// The default open filter leaves out the answers of a test attempt that is
+// still running: they cannot be reviewed until it ends (ErrAttemptOpen).
 func reviewWhere(f ReviewFilter, statuses []string) (string, []any) {
 	var b strings.Builder
 	var args []any
@@ -294,6 +301,10 @@ func reviewWhere(f ReviewFilter, statuses []string) (string, []any) {
 		b.WriteString(" AND aa.grading_status IN ?")
 		args = append(args, statuses)
 	}
+	if status := strings.TrimSpace(f.Status); status == "" || status == ReviewStatusOpen {
+		b.WriteString(" AND NOT (qz.mode = ? AND a.completed_at IS NULL)")
+		args = append(args, models.ModeTest)
+	}
 	if slug := strings.TrimSpace(f.QuizSlug); slug != "" {
 		b.WriteString(" AND qz.slug = ?")
 		args = append(args, slug)
@@ -369,7 +380,8 @@ type reviewTarget struct {
 // lockReviewTarget locks the attempt of an open answer and reads the
 // answer again under the lock, so a grading job or a re-answer cannot
 // change it in between. A missing answer or attempt, or an answer that is
-// not open, is ErrNotFound.
+// not open, is ErrNotFound; an answer of a test attempt that is still
+// running is ErrAttemptOpen.
 func lockReviewTarget(ctx context.Context, tx *gorm.DB, answerID uint) (*reviewTarget, error) {
 	tx = tx.WithContext(ctx)
 	var row models.AttemptAnswer
@@ -388,6 +400,9 @@ func lockReviewTarget(ctx context.Context, tx *gorm.DB, answerID uint) (*reviewT
 		return nil, notFound(err)
 	}
 	attempt.Quiz = &quiz
+	if Withheld(attempt) {
+		return nil, ErrAttemptOpen
+	}
 	var question models.Question
 	if err := tx.Where("id = ?", row.QuestionID).Take(&question).Error; err != nil {
 		return nil, notFound(err)
@@ -583,9 +598,10 @@ func (s *Service) SetScore(ctx context.Context, answerID uint, points float64, c
 // confirmed. Pending, failed and reviewed answers are left alone. The
 // affected attempts are locked in id order, recalculated once each, and
 // every changed answer and every completed attempt that is now fully
-// reviewed fire their events after the commit. A missing quiz or attempt
-// is ErrNotFound; a scope that is not exactly one of the two is a
-// *ValidationError.
+// reviewed fire their events after the commit. Answers of a test attempt
+// that is still running are skipped; a scope of one such attempt is
+// ErrAttemptOpen. A missing quiz or attempt is ErrNotFound; a scope that
+// is not exactly one of the two is a *ValidationError.
 func (s *Service) ConfirmBulk(ctx context.Context, scope ReviewScope, by Reviewer) (int, error) {
 	if problems := CheckReviewScope(scope, false); len(problems) > 0 {
 		return 0, &ValidationError{Problems: problems}
@@ -607,6 +623,12 @@ func (s *Service) ConfirmBulk(ctx context.Context, scope ReviewScope, by Reviewe
 		}
 		touched := map[uint]bool{}
 		for _, row := range rows {
+			if Withheld(attempts[row.AttemptID]) {
+				if scope.AttemptID != 0 {
+					return ErrAttemptOpen
+				}
+				continue
+			}
 			t := &reviewTarget{Answer: row, Question: questions[row.QuestionID], Attempt: attempts[row.AttemptID]}
 			done, err := s.confirmTarget(ctx, tx, t, by)
 			if err != nil {
diff --git a/controllers/api/reviews.go b/controllers/api/reviews.go
index 7a158bf..e250837 100644
--- a/controllers/api/reviews.go
+++ b/controllers/api/reviews.go
@@ -33,6 +33,8 @@ func writeReview(w http.ResponseWriter, status int, data any, err error) {
 		reviewConflict(w, "The answer is already reviewed; set the score instead.", "already_reviewed")
 	case errors.Is(err, classes.ErrGraderManual):
 		reviewConflict(w, "The quiz's grader is manual or not available, so nothing can be graded again.", "grader_manual")
+	case errors.Is(err, classes.ErrAttemptOpen):
+		reviewConflict(w, "The test attempt is still open; review its answers once it ends.", "attempt_open")
 	case errors.Is(err, classes.ErrNotFound):
 		wire.WriteJSON(w, http.StatusNotFound, map[string]any{"error": true, "message": "Open answer, attempt or quiz not found."})
 	default:
diff --git a/controllers/reviews.go b/controllers/reviews.go
index 4a1cb5f..a400b0f 100644
--- a/controllers/reviews.go
+++ b/controllers/reviews.go
@@ -245,6 +245,8 @@ func reviewError(err error) error {
 		return &cabana.ForbiddenError{Message: reviewsLang + "already_reviewed"}
 	case errors.Is(err, classes.ErrGraderManual):
 		return &cabana.ForbiddenError{Message: reviewsLang + "grader_manual"}
+	case errors.Is(err, classes.ErrAttemptOpen):
+		return &cabana.ForbiddenError{Message: reviewsLang + "attempt_open"}
 	case errors.Is(err, classes.ErrNotFound):
 		return &cabana.ForbiddenError{Message: reviewsLang + "not_found"}
 	}
diff --git a/lang/en/lang.yaml b/lang/en/lang.yaml
index 7aee6a0..f03319a 100644
--- a/lang/en/lang.yaml
+++ b/lang/en/lang.yaml
@@ -128,5 +128,6 @@ reviews:
     no_grader_score: This answer has no grader's score yet; enter the points yourself.
     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.
     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 d9d20dd..a467153 100644
--- a/lang/pl/lang.yaml
+++ b/lang/pl/lang.yaml
@@ -128,5 +128,6 @@ reviews:
     no_grader_score: Ta odpowiedź nie ma jeszcze wyniku oceniającego; wpisz punkty sam.
     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.
     not_found: Nie znaleziono tej odpowiedzi.
     not_created: Odpowiedzi otwarte pochodzą od studentów; tutaj nie można ich dodawać ani usuwać.
diff --git a/mcp_reviews.go b/mcp_reviews.go
index 83d2ca8..babd342 100644
--- a/mcp_reviews.go
+++ b/mcp_reviews.go
@@ -49,7 +49,7 @@ func reviewTools() []pact.MCPTool {
 		{
 			Name:        "review_list",
 			Title:       "List open answers to review",
-			Description: "Lists the answers students wrote to open questions, oldest change first, for you to review. status: open (default: pending, graded and failed, everything not reviewed yet), pending (waiting for the grader or for you), graded (the grader gave a score you have not confirmed), failed (the grader could not score it; set the score yourself), reviewed or all. Filters: quiz (slug), user_id, attempt_id; page and per_page (1-100, default 20). Each row: answer_id, status, quiz {id, slug, title, mode}, question {id, text, points}, attempt {id, completed, completed_at}, student {user_id, email, name, surname}, text (the first 300 characters, text_truncated when cut), grader_points, points, graded_at, reviewed_at, updated_at. Read one answer in full with quizzes_review_show. It shows student emails, names and answers." + needsRead,
+			Description: "Lists the answers students wrote to open questions, oldest change first, for you to review. status: open (default: pending, graded and failed, everything not reviewed yet, except answers of a test attempt still in progress), pending (waiting for the grader or for you), graded (the grader gave a score you have not confirmed), failed (the grader could not score it; set the score yourself), reviewed or all. Filters: quiz (slug), user_id, attempt_id; page and per_page (1-100, default 20). Each row: answer_id, status, quiz {id, slug, title, mode}, question {id, text, points}, attempt {id, completed, completed_at}, student {user_id, email, name, surname}, text (the first 300 characters, text_truncated when cut), grader_points, points, graded_at, reviewed_at, updated_at. Read one answer in full with quizzes_review_show. It shows student emails, names and answers." + needsRead,
 			Method:      http.MethodGet,
 			Path:        ReviewPrefix,
 			Input:       reviewListInput{},
@@ -67,7 +67,7 @@ func reviewTools() []pact.MCPTool {
 		{
 			Name:        "review_confirm",
 			Title:       "Confirm the grader's score",
-			Description: "Confirms the grader's score of one graded answer: status becomes reviewed and points become grader_points, final over any later grading, and a finished attempt's score is recalculated. A pending or failed answer has no score to confirm and is refused with 409 code no_grader_score (set the score with quizzes_review_set_score instead); confirming a reviewed answer changes nothing. Returns the answer as quizzes_review_show does." + needsWrite,
+			Description: "Confirms the grader's score of one graded answer: status becomes reviewed and points become grader_points, final over any later grading, and a finished attempt's score is recalculated. A pending or failed answer has no score to confirm and is refused with 409 code no_grader_score (set the score with quizzes_review_set_score instead); confirming a reviewed answer changes nothing. An answer of a test attempt still in progress is refused with 409 code attempt_open. Returns the answer as quizzes_review_show does." + needsWrite,
 			Method:      http.MethodPost,
 			Path:        ReviewPrefix + "/{id}/confirm",
 			Input:       reviewAnswerInput{},
@@ -77,7 +77,7 @@ func reviewTools() []pact.MCPTool {
 		{
 			Name:        "review_set_score",
 			Title:       "Set the score of an open answer",
-			Description: "Sets your own final score on one open answer, whatever its status (pending, graded, failed or reviewed): points from 0 up to the question's points in steps of 0.5 (anything else is refused with 422), and an optional comment of at most 2000 characters that the student sees with the answer; leaving the comment out clears an earlier one. status becomes reviewed and the score is final over any later grading; the grader's score and feedback stay for the record, and a finished attempt's score is recalculated. Returns the answer as quizzes_review_show does." + needsWrite,
+			Description: "Sets your own final score on one open answer, whatever its status (pending, graded, failed or reviewed): points from 0 up to the question's points in steps of 0.5 (anything else is refused with 422), and an optional comment of at most 2000 characters that the student sees with the answer; leaving the comment out clears an earlier one. status becomes reviewed and the score is final over any later grading; the grader's score and feedback stay for the record, and a finished attempt's score is recalculated. An answer of a test attempt still in progress is refused with 409 code attempt_open. Returns the answer as quizzes_review_show does." + needsWrite,
 			Method:      http.MethodPost,
 			Path:        ReviewPrefix + "/{id}/score",
 			Input:       reviewSetScoreInput{},
@@ -87,7 +87,7 @@ func reviewTools() []pact.MCPTool {
 		{
 			Name:        "review_confirm_bulk",
 			Title:       "Confirm grader scores in bulk",
-			Description: "Confirms the grader's score of every graded answer of one quiz (quiz, its slug) or of one attempt (attempt_id); give exactly one of the two. Each becomes reviewed with points = grader_points, as quizzes_review_confirm does; pending, failed and reviewed answers are left alone, so check failed ones with quizzes_review_list first. Finished attempts are recalculated. Returns {confirmed: the number of answers confirmed}; a second call confirms nothing new." + needsWrite,
+			Description: "Confirms the grader's score of every graded answer of one quiz (quiz, its slug) or of one attempt (attempt_id); give exactly one of the two. Each becomes reviewed with points = grader_points, as quizzes_review_confirm does; pending, failed and reviewed answers are left alone, so check failed ones with quizzes_review_list first. Answers of test attempts still in progress are skipped (an attempt_id of one is refused with 409 code attempt_open). Finished attempts are recalculated. Returns {confirmed: the number of answers confirmed}; a second call confirms nothing new." + needsWrite,
 			Method:      http.MethodPost,
 			Path:        ReviewPrefix + "/confirm",
 			Input:       reviewConfirmBulkInput{},