THE BOUNDARIES
Safety, confidence and privacy.
CREST works with information about your body and mind. The boundaries have to be explicit, and they have to hold even when the model is wrong. So they are not written in a policy nobody reads — safety, permissions, storage and what can change your plan are deterministic rules in code, running before and after the model.
A rule written in a prompt is a suggestion. A rule written in code is a rule.
The rules run in code, not in a prompt.
Deterministic checks before and after the model. A recommendation that fails validation is never shown.
It stops rather than guesses.
No diagnosis, no medication decisions, no crisis response. It routes you to someone qualified.
Your data is never sold or shared identifiably.
You can see what CREST remembers, correct it, and delete it.
HOW IT BEHAVES
Confidence is stated, and medication is never prescribed.
Confidence, stated.
Every observation carries a confidence level based on how complete, recent and relevant the underlying information is.
High confidence produces a clear recommendation. Low confidence produces a question. CREST is designed to say “I don’t have enough to tell you that” rather than generate a plausible answer.
Medication.
You can tell CREST what you’re prescribed and when you took it. It’s important information — medication interacts with food, sleep and training in ways that are hard to see from the inside.
CREST will track what you log, correlate it with everything else, and explain why you may be feeling as you do. It will act on the things you control: what to eat, when to sleep, how hard to train.
CREST will not tell you what dose to take, when to take it, or whether to change anything about your prescription. When a pattern points at your treatment rather than your routine, CREST says so and routes you to your prescriber, with the pattern summarised so the conversation starts from evidence.
THE LIMITS
What CREST does not do, and where that is enforced.
- No diagnosis.
- No advice on medication dose, timing or changes.
- No recommendation of clinical interventions, tests or procedures.
- No emergency response — CREST is not a crisis service.
- No treatment of clinical conditions.
- No claims about preventing disease or injury.
When something falls outside what CREST can responsibly handle, it stops and points you to someone qualified. Mental health follows the same rule: CREST helps you regulate, and routes you to a professional at the boundary.
Safety rules run in code before and after the model, not inside the prompt. A recommendation that fails validation is never shown. High-risk signals follow a fixed escalation path that a model cannot override.
YOUR DATA
Your data.
- Never sold. Never used for advertising.
- Never shared with an employer, club or venue in a form that identifies you.
- Where CREST works with an organisation, they receive aggregate patterns only.
- You can see what CREST remembers, correct it, and delete it.
CREST is in development.
