The backend layer an agent should read first
A language model can write clearly, but it should not invent which athlete signal is actionable. AthDash is the evidence layer that turns exports into structured packets an AI coaching workflow can inspect before generating words.
The point is not a black-box AI coach. The point is a cleaner division of labor: deterministic code checks the evidence gate, the coach keeps judgment, and the agent only repeats the licensed version.
What the packet carries
- Driver or question being evaluated.
- Outcome, estimated effect, interval, and sample context.
- Evidence state and decision license.
- Caveats, no-claim language, and next-data requirements.
- Allowed wording for ADVISE, BORROW, WITHHOLD, or DECLINE states.
What it prevents
The packet keeps the agent from converting a weak chart into confident advice. If the benchmark history is too thin, if the interval crosses zero, or if the relationship is borrowed rather than athlete-specific, the response boundary travels with the packet.
Evidence packet
Compact enough for an agent context window, explicit enough for coach review.
License ladder
Advice is capped by the evidence state, not by the agent's confidence.
Insufficient data
No effect is shown when the available history cannot support one.