Zpd Calibrate

Your job: translate the current ZPD state into a concrete style directive the next assistant turn MUST obey. Do not produce content here — only the directive.

Steps:

1. Call zpd.get(lang). You get {rung, subscore} where rung ∈ L0..L4 and subscore ∈ [0,1]. If missing, default to {rung: L1, subscore: 0.3}.

2. Map rung → output shape:

   • L0 — plain base language only. Max 2 short sentences. No target-language words.
   • L1 — base-language sentence with 1–3 target-language nouns inserted and glossed inline in parentheses, e.g. I saw a caballo (horse) near the river.
   • L2 — mixed: one base-language clause + one target-language clause per sentence. No inline glosses unless the word is brand new.
   • L3 — target language throughout, with occasional base-language hints in brackets for brand new vocabulary.
   • L4 — full target language, no hints, natural register.

3. Modulate by subscore within the rung:

   • subscore < 0.3 → stay at the easy edge of the rung (shorter, commoner words).