Methodology
How we calculate it
The technical side of Hermacles, for anyone who wants to check our work.
Numbers come from code, not from the AI
Specialists for readiness, patterns, running, strength and body composition compute every number in code, the same way every time.
We call this the deterministic core. The AI coach puts its numbers into words; it never calculates them.
Patterns are associations, shown with their evidence
A pattern is shown only with its sample size n and its FDR-corrected p, and it is described as an association, never as a cause.
One source per metric
Each metric has one source of authority, and readings from different devices are never averaged. When two devices disagree, Hermacles shows which one wins and why.
A reviewer checks every plan
A reviewer checks each plan against thresholds with public sources, such as hard-day spacing and long-run growth. A plan that fails goes back to the planner.
Thresholds come from public sources and record who approved them. You see a day-by-day diff, and the active plan changes only when you say yes.
A coach or physiologist you invite can tune your thresholds only within fixed safety limits.
Each week learns from the last one
Completed sessions are matched to the plan. A missed session tells the planner something about your week; it is never treated as a failure.
Your history, encrypted per user
Free text, raw payloads and source tokens are encrypted with your own key. Delete your account and the key is destroyed, so the data cannot be read again.
Safety red flags
Every message to the coach passes a safety check first.
Warning signs, such as chest pain or fainting, stop the coaching and hand you off to a health professional, with a 24-hour pause on training advice. It is a safety stop, not a diagnosis.