The packet is the evidence boundary
An evidence packet keeps a coaching claim from floating away from the data that supports it. It carries the driver, outcome, effect estimate, interval, sample context, caveats, evidence state, and allowed wording in one small object.
That matters for AI-assisted coaching because a language model should not decide which statistical relationship is actionable. It should read the packet, respect the boundary, and phrase only the licensed version.
Fields in a packet
- Driver or question: the signal or training lever being evaluated.
- Outcome: the athlete-specific performance or readiness target.
- Effect and interval: the estimate and uncertainty range, not a bare number.
- Evidence state: supported, exploratory, borrowed, withheld, or insufficient.
- Sample context: event count, window, exclusions, and coverage notes.
- Allowed wording: what the coach or agent may say without overclaiming.
Decision licenses
ADVISEThe relationship is strong enough to use, with caveats visible.
BORROWA prior or comparable pattern can guide observation, but it is not this athlete's own settled effect.
WITHHOLDThe pattern is not strong enough to claim, even if it looks tempting.
DECLINEThe available data is insufficient, so no effect is shown.