assess

name	assess
pack	vocab
description	CEFR placement test — 30 questions, estimate level, seed initial deck
command	true
triggers	["vocab assess","placement","test my level","测试水平","评估等级"]

CEFR Placement Assessment

Place the user on the A2 / B1 / B2 / C1 spectrum, then build their initial card deck.

Goal

Run a 30-question quiz, compute a CEFR level, persist it, and seed 20 starter cards from that level.

Procedure

1. Start a run Use the db tool to read the 30 rows from assessment_items and the persist tool to open a run:

   {
     "action": "practice",
     "skill_type": "vocab.assess.start",
     "content": "Assessment started",
     "metadata": "{\"target_count\":30}"
   }
   

2. Ask questions

   • Iterate items in order: A2 (10) → B1 (10) → B2 (6) → C1 (4).
   • Use the msg tool with type: "quiz" to render each as a Telegram-native multiple-choice (4 options).
   • Track answers in memory: per-level correct count.

3. Adaptive shortcut (optional)

   • If the user scores 0/3 in a level's first 3 items, skip the rest of that level — they're below it. Drop straight to closing.
   • If they score 9-10/10 in A2 AND 8-10/10 in B1, you can shorten B2 to its first 3 items and C1 to its first 2 to save time. Use judgment.

4. Compute level

   • Highest level where they got ≥60% correct → that's their CEFR level.
   • Ties resolve down (be conservative).

5. Persist result

   {
     "action": "practice",
     "skill_type": "vocab.assess.complete",
     "content": "Placed at <level>",
     "metadata": "{\"cefr_level\":\"B1\",\"scores\":{\"A2\":\"9/10\",\"B1\":\"7/10\",\"B2\":\"2/6\",\"C1\":\"0/4\"}}"
   }
   

   Also write user_state row cefr_level = <level> and last_assessed_at = now.

6. Seed initial deck

   • Generate 20 high-frequency words appropriate to the placed level (avoid obviously below-level words).
   • For each word, fill words row with: word, cefr_level, ipa, def_en, def_zh, one example_en, one example_zh.
   • For each word, insert a cards row with status='new', next_due=date('now').
   • This is a batched LLM operation — generate all 20 at once if context allows; otherwise loop in batches of 5.

7. Wrap up

   • Tell them: "You're at . I've seeded 20 starter cards — first session at 08:00 tomorrow, or say 'study' now."
   • Briefly mention the interval curve they can customize later.

Rules

• Never grade items leniently — the level matters for what they'll see for weeks.
• Never reveal correct answers mid-quiz; defer feedback to the wrap-up if asked.
• Generated word entries should be real, common, and not obviously easier than the placed level.