package services import ( "context" "encoding/json" "fmt" "testing" "time" "speech-learning-api/internal/domain" "speech-learning-api/internal/store" ) func statisticalResult(word string, score float64) domain.PronunciationResult { return domain.PronunciationResult{Provider: "azure", Locale: "en-US", Alphabet: "IPA", Method: "scripted", MeasurementVersion: "azure-rest-v1", RecognizedText: word, OverallScore: score, WordScores: []domain.WordScore{{Word: word, Accuracy: score, PhonemeScores: []domain.PhonemeScore{{Phoneme: "k", Accuracy: score}}}}} } func profileSound(t *testing.T, p domain.PronunciationProfile) domain.PronunciationTarget { t.Helper() for _, x := range p.Targets { if x.Kind == "phoneme" && x.Symbol == "k" { return x } } t.Fatalf("/k/ missing: %+v", p) return domain.PronunciationTarget{} } func TestPronunciationStatisticsPostgresPracticePlansAndTransfer(t *testing.T) { db, repo := adaptiveTestDB(t) ctx := context.Background() user, err := repo.CreateUser(ctx, "statistical@example.test", "hash", "de-DE") if err != nil { t.Fatal(err) } other, err := repo.CreateUser(ctx, "statistical-other@example.test", "hash", "de-DE") if err != nil { t.Fatal(err) } if _, err = repo.SaveAttempt(ctx, domain.Attempt{UserID: user.ID, ExerciseID: "ex_a1_water", TargetLanguage: "en-US", WordScores: []domain.WordScore{{Word: "legacy", Accuracy: 0}}}); err != nil { t.Fatal(err) } mock := statisticalResult("mock", 1) mock.Provider = "mock" if _, err = repo.SaveAttempt(ctx, domain.Attempt{UserID: user.ID, ExerciseID: "ex_a1_water", TargetLanguage: "en-US", Measurement: &mock}); err != nil { t.Fatal(err) } for i, w := range []string{"cat", "cup", "school"} { _, ex, err := repo.SaveGeneratedExercise(ctx, user.ID, domain.GeneratedExercise{TargetLanguage: "en-US", CEFRLevel: "A1", ReferenceText: fmt.Sprintf("I see %s.", w), Source: "ai"}) if err != nil { t.Fatal(err) } r := statisticalResult(w, float64(30+i*10)) if _, err = repo.SaveAttempt(ctx, domain.Attempt{UserID: user.ID, ExerciseID: ex.ID, TargetLanguage: "en-US", WordScores: r.WordScores, Measurement: &r}); err != nil { t.Fatal(err) } } now := time.Now().UTC().Add(time.Second) p, err := repo.PronunciationProfile(ctx, user.ID, "en-US", now) if err != nil { t.Fatal(err) } k := profileSound(t, p) if k.Samples != 3 || k.Words != 3 || k.State != "needs_practice" || p.UnverifiedHistory != 1 { t.Fatalf("bad profile: %+v %+v", p, k) } isolated, err := repo.PronunciationProfile(ctx, other.ID, "en-US", now) if err != nil || len(isolated.Targets) != 0 { t.Fatal("user isolation", err, isolated) } isolated, err = repo.PronunciationProfile(ctx, user.ID, "de-DE", now) if err != nil || len(isolated.Targets) != 0 { t.Fatal("language isolation", err, isolated) } plan, err := repo.PlanPronunciation(ctx, user.ID, "en-US", now) if err != nil { t.Fatal(err) } if plan.Target == nil || plan.Target.Symbol != "k" || plan.RequiredWords[0] != "cat" { t.Fatalf("not a bundled target: %+v", plan) } generated := domain.GeneratedExercise{TargetLanguage: "en-US", CEFRLevel: "A1", ReferenceText: "My cat is very small.", Source: "ai", TrainingPlan: &plan} _, ex, err := repo.SaveGeneratedExercise(ctx, user.ID, generated) if err != nil { t.Fatal(err) } _, again, err := repo.SaveGeneratedExercise(ctx, user.ID, generated) if err != nil || again.ID != ex.ID || again.TrainingReason == "" { t.Fatal("assignment replay", err, again) } // A transfer stays unconfirmed until the expected symbol is actually measured. transfer := plan transfer.ID = store.NewID("transfer") transfer.Mode = "transfer" transfer.Category = "explore" transfer.RequiredWords = []string{} transfer.CandidateWord = "kite" transfer.TransferStatus = "unconfirmed" raw, _ := json.Marshal(transfer) if _, err = db.Exec(`INSERT INTO pronunciation_plans(id,user_id,language,data,status) VALUES($1,$2,'en-US',$3,'pending')`, transfer.ID, user.ID, raw); err != nil { t.Fatal(err) } generated.ReferenceText = "I see a kite." generated.TrainingPlan = &transfer _, ex, err = repo.SaveGeneratedExercise(ctx, user.ID, generated) if err != nil { t.Fatal(err) } r := statisticalResult("kite", 80) if _, err = repo.SaveAttempt(ctx, domain.Attempt{UserID: user.ID, ExerciseID: ex.ID, TargetLanguage: "en-US", WordScores: r.WordScores, Measurement: &r}); err != nil { t.Fatal(err) } if err = db.QueryRow(`SELECT data->>'transfer_status' FROM pronunciation_plans WHERE id=$1`, transfer.ID).Scan(&transfer.TransferStatus); err != nil || transfer.TransferStatus != "confirmed" { t.Fatal("transfer not verified", err, transfer) } var count int db.QueryRow(`SELECT count(*) FROM pronunciation_assessments`).Scan(&count) if err = repo.Init(ctx); err != nil { t.Fatal(err) } var after int db.QueryRow(`SELECT count(*) FROM pronunciation_assessments`).Scan(&after) if after != count { t.Fatal("restart duplicated evidence", count, after) } // Unknown historical raw values are never silently promoted. p, err = repo.PronunciationProfile(ctx, user.ID, "en-US", time.Now().Add(time.Minute)) if err != nil || p.UnverifiedHistory != 1 { t.Fatal(err, p) } } func TestPronunciationStatisticsPostgresCourseProvenanceAndExamPrivacy(t *testing.T) { db, repo := adaptiveTestDB(t) ctx := context.Background() user, err := repo.CreateUser(ctx, "stat-course@example.test", "hash", "de-DE") if err != nil { t.Fatal(err) } adaptive := store.NewAdaptiveStore(db) insert := func(kind, exam string) domain.AdaptiveTask { t.Helper() task := domain.AdaptiveTask{SchemaVersion: 2, ID: store.NewID("stat_task"), ModuleID: "a1_family_v1", TargetLanguage: "en-US", CourseVersion: domain.CourseVersion, CEFRLevel: "A1", Skill: "speaking", Kind: kind, ReferenceText: "My cat is small.", Source: "curated", ExamID: exam} raw, _ := json.Marshal(task) _, err := db.Exec(`INSERT INTO course_v2_tasks(id,module_id,language,version,owner_id,source,skill,kind,data,fingerprint,exam_id) VALUES($1,$2,$3,$4,$5,'curated','speaking',$6,$7,$1,NULLIF($8,''))`, task.ID, task.ModuleID, task.TargetLanguage, task.CourseVersion, user.ID, kind, raw, exam) if err != nil { t.Fatal(err) } return task } r := statisticalResult("cat", 30) task := insert("repeat_sentence", "") attempt, err := adaptive.Submit(ctx, user.ID, "same-request", task, "My cat is small.", &r) if err != nil { t.Fatal(err) } if _, err = adaptive.Submit(ctx, user.ID, "same-request", task, "ignored", &r); err != nil { t.Fatal(err) } // An ungraded content answer still contains a real pronunciation measurement. if _, err = db.Exec(`UPDATE course_v2_attempts SET status='ungraded' WHERE id=$1`, attempt.ID); err != nil { t.Fatal(err) } free := insert("free_speech", "") r.Method = "unscripted" if _, err = adaptive.Submit(ctx, user.ID, "free-request", free, "My cat is small.", &r); err != nil { t.Fatal(err) } r.Method = "scripted" exam := store.NewID("stat_exam") if _, err = db.Exec(`INSERT INTO course_v2_exams(id,user_id,language,version,status,request_id) VALUES($1,$2,'en-US',$3,'in_progress',$1)`, exam, user.ID, domain.CourseVersion); err != nil { t.Fatal(err) } exTask := insert("repeat_sentence", exam) if _, err = adaptive.Submit(ctx, user.ID, "exam-request", exTask, "My cat is small.", &r); err != nil { t.Fatal(err) } now := time.Now().Add(time.Minute) p, err := repo.PronunciationProfile(ctx, user.ID, "en-US", now) if err != nil { t.Fatal(err) } if k := profileSound(t, p); k.Samples != 1 { t.Fatalf("duplicates/free speech/unfinished exam counted: %+v", k) } if _, err = db.Exec(`UPDATE course_v2_exams SET status='completed' WHERE id=$1`, exam); err != nil { t.Fatal(err) } p, err = repo.PronunciationProfile(ctx, user.ID, "en-US", now) if err != nil { t.Fatal(err) } if k := profileSound(t, p); k.Samples != 2 { t.Fatal("finished exam not counted", k) } // Rebuild confirmed course evidence; archived copies must never be counted twice. if _, err = db.Exec(`DELETE FROM pronunciation_assessments WHERE source='course'`); err != nil { t.Fatal(err) } if err = repo.BackfillPronunciation(ctx); err != nil { t.Fatal(err) } if err = repo.BackfillPronunciation(ctx); err != nil { t.Fatal(err) } p, err = repo.PronunciationProfile(ctx, user.ID, "en-US", now) if err != nil { t.Fatal(err) } if k := profileSound(t, p); k.Samples != 2 { t.Fatal("backfill lost/duplicated evidence", k) } }