Answering well once proves nothing. The system remembers where the athlete is weak, and drives the next session there.
A professional communication situation opens: a journalist insists on a refereeing decision after a defeat. The athlete answers in French — by voice or in writing. The exercise trains language and public communication, not football tactics.
Technical vocabulary, diplomacy, structure and risk are scored by a dedicated engine. Same answer, same score, always — no generative lottery.
Professional mines cap the score whatever the linguistic brilliance, and explain why: an accusation, a premature transfer confirmation, a fitness claim against medical advice.
Mastery per scenario category is tracked in a signed profile, weighted by media pressure. Any tampering puts the profile in quarantine.
The next session is chosen automatically: mandatory groups first, then the weakest category — never the same scenario twice in a row.
Every evaluation is chained to the previous one by SHA-256. Ordinary deletion or alteration breaks verification visibly; the local ledger is tamper-evident, not claimed to be tamper-proof. Any optional human-signed private attestation is handled separately from the standard Lemon software licence.
A local model (Ollama) and local voice can enrich dialogue on the machine. The deterministic evaluation core does not depend on either one or on a cloud API. Quantum Excellium Core manages the local-first execution path, fallback behaviour, memory and integrity controls.