AI coach modes

CombatScore has three AI assistants, all grounded in real data and a curated technique corpus. They're not chatbots — they cite their sources and refuse to make things up.

1. Game plan (athlete-facing)

Dashboard → Game plan.

Tap the AI to draft a short match strategy for today (or a date you pick). It uses:

• Your recent training history (what you've been working on).
• Your weight, belt, and primary discipline.
• Tags you've used on recent sessions (what's working, what isn't).
• The technique corpus (so it doesn't suggest moves that don't exist).

Output is a 5-block plan: warmup, primary game, backup plan, what to avoid, mental cue. Editable; save as your game plan for the day.

2. Opponent analysis (coach-facing)

Coach → Game plan → New opponent.

Drop in a description of an opponent: belt, weight, school, known techniques, recent results. The AI drafts a counter-game-plan grounded in:

• The opponent profile you provided.
• Technique corpus (grappling, striking, wrestling — whichever discipline is in scope).
• The athlete's recent training (if you tied the plan to one of your athletes).

Output is an append-only document. Revisions go to a new version — the history isn't overwritten, so you can compare drafts week-over-week if you're prepping for a long arc.

3. Athlete coaching (coach-facing)

Coach → Athletes → [Athlete] → AI insights.

A daily-refreshed summary of one athlete:

• Patterns in their recent sessions (what they're drilling, what they're skipping).
• Gaps relative to their belt's expected curriculum.
• Suggested talking points for your next 1:1.
• Risk flags (overtraining, injury patterns, attendance drift).

Cites the sessions it pulled from so you can verify. If it suggests something based on three sparring rounds, you'll see those three rounds linked.

What the AI is NOT doing

• Not training on your data. Our model providers (Anthropic, OpenAI) operate under enterprise terms that prohibit training on customer data. See Your data rights.
• Not making up techniques. Output is grounded against the curated technique corpus. If a move isn't in the corpus, the AI won't suggest it.
• Not replacing you. Every output is editable; nothing the AI says is auto-applied to the athlete.

Rate limits

20 AI messages per day per account. Generous; most coaches don't come close. If you hit the cap, you'll see a clear "daily limit reached" message and it resets at midnight your local time.

When the AI says "I'm not confident"

It will sometimes refuse to answer or flag low confidence. Trust that signal — it's better at knowing what it doesn't know than at confabulating. If you want a better answer, give it more context (paste in recent session notes, link an opponent's public profile, etc.).