fix(vocab-quiz): UI cluster — anti-cheat + UX (#190 #192 #193 #194)

[davidortinau]

Summary

Stream A of the Vocabulary Quiz bug cluster — UI-only fixes in src/SentenceStudio.UI/Pages/VocabQuiz.razor plus resource entries.

Closes #189
Closes #190
Closes #192
Closes #193
Closes #194

Changes

#189 — Strip legacy field readouts from Learning Details panel

Per Jayne's service-side repro tests on main (Repro189_SingleCorrectRecognitionAttempt_ProducesExpectedPanelState and Repro189_SingleCorrectRecognition_LegacyProductionFieldsRemainZero both pass), the service math is clean. The "2 attempts / 50% accuracy" confusion was a UI-rendering artifact — the panel was showing pre-streak-truth metadata (IsKnown, IsUserDeclared, VerificationState) alongside the current streak fields, conflating two different mental models of "is this word known".

Restructured the two stat grids in the panel into one streak-truth grid showing only the allowlist:

• TotalAttempts, CorrectAttempts, Accuracy
• CurrentStreak, ProductionInStreak, EffectiveStreak
• MasteryScore
• status badge (already kept)

Schema fields (IsKnown, VerificationState, ProductionAttempts, RecognitionAttempts, etc.) on VocabularyProgress are unchanged for sync/back-compat — only the rendering is tightened.

Also audited RecordPendingAttemptAsync call sites (NextItem, override-to-correct, DisposeAsync): no double-invocation found. The method is idempotent via the if (pendingAttempt == null) return; guard followed by pendingAttempt = null; — calling it twice is a no-op on the second call. The sentence-shortcut path at line 980 deliberately records one attempt per credited sentence, which is correct.

#194 — Anti-cheat: hide terms in Learning Details panel

The info (ⓘ) panel is opened mid-quiz. Previously it rendered the target word as a heading and the native gloss right under it — completely defeating the purpose of testing recognition. Now the panel shows ONLY the status badge and learning metadata. The terms are removed from this view entirely.

#190 — Multiple-choice always has 4 options

Pre-fix, distractors were drawn from vocabItems (the active round). When the round was small (e.g. SRS DueOnly with 1–3 words), the user got 1–3 options, often making the answer obvious. Now we capture the FULL filtered+deduped vocabulary scope into a new distractorScope field during LoadVocabulary, and GenerateChoiceOptions samples from it. Verified e2e: with 7 review-due Korean words, the quiz now shows distractors like "to come out" / "to learn" / "cheese" alongside the correct answer drawn from the full 327-word vocab.

#193 — Prompt audio follows prompt direction

Audio play button on the prompt previously called GetTargetAudioText / GetTargetAudioLanguage regardless of direction. When the user has prompt direction set to native→target (English shown, Korean hidden), pressing play would speak the Korean answer — leaking the test. New helpers GetPromptAudioText and GetPromptAudioLanguage switch on promptUsesNativeLanguage and return the actually-shown side. PlayWordAudio and the audioPromptAvailable check now route through these helpers.

#192 — Submit button for text-entry mode

Text-entry quiz mode previously required pressing Enter to submit. Added a visible Submit button inside the form. Disabled while input is empty; hidden once feedback is shown in non-correction mode. New resource key VocabQuiz_SubmitAnswer ("Submit" / "제출"). The button is type="submit" and routes through the existing @onsubmit="SubmitTypedAnswer" handler.

Files changed

• src/SentenceStudio.UI/Pages/VocabQuiz.razor — all five fixes
• src/SentenceStudio.Shared/Resources/Strings/AppResources.resx — VocabQuiz_SubmitAnswer = "Submit"
• src/SentenceStudio.Shared/Resources/Strings/AppResources.ko.resx — VocabQuiz_SubmitAnswer = "제출"
• src/SentenceStudio.Shared/Resources/Strings/AppResources.Designer.cs — auto-regenerated

Verification

• ✅ SentenceStudio.UI.csproj builds clean (0 errors)
• ✅ SentenceStudio.MacCatalyst.csproj net10.0-maccatalyst builds clean (0 errors)
• ✅ SentenceStudio.WebApp.csproj builds clean (0 errors) — verified after Accurate and total attempt don't make sense #189 edit
• ✅ Code review (Stream A): APPROVE
• ✅ Every quiz multiple choice should have a full set of answer options #190 e2e — distractors from full scope visually confirmed in browser
• ✅ Learning Details modal leaks the answer #194 e2e — info panel shows only status badge, no terms (DOM + screenshot)
• ✅ Accurate and total attempt don't make sense #189 — service-side already proven clean by Jayne's repro tests on main; UI fix is mechanical (strip + reorganize fields). Visual surface is the same panel as Learning Details modal leaks the answer #194 — re-verification on the same offcanvas is low-risk.
• ⚠️ Add a button to text entry Quiz items #192 e2e — code path verified via build; full UI exercise requires a vocabulary word past MasteryScore >= 0.50 or CurrentStreak >= 3 to trigger text-entry mode.
• ⚠️ Audio plays the actual target language audio #193 e2e — code path verified via build and audio button click; TTS streams server-side via SignalR so no browser-visible network request to assert against.

Out of scope

• Any backend / service / repository changes — Jayne owns Stream B (PR test: failing repro for vocab quiz scoring bugs (#189, #191) #195 draft)
• Cleanup of unused resource keys (VocabQuiz_IsKnown, VocabQuiz_Yes, VocabQuiz_No are no longer referenced from this panel but remain in resx files; low-priority cleanup)