How MindFrame uses AI

MindFrame uses Anthropic's Claude models through Anthropic's API. Most requests use a small, fast model; some Member features may use a larger one. We do not use any other AI provider.

• Scores free-text reasoning when rule-based checks cannot decide.
• Writes coaching, session debriefs and insight summaries grounded in your training data.
• Suggests next practice based on your weakest areas.

• Whether an answer is right — that comes from each challenge's authored answer key.
• Your calibration and Brier scores — those are plain maths (see methodology).
• Anything about your health, employment or eligibility for anything.

Only what a feature needs: for example your answer, your reasoning text and recent session results. This data is sent to Anthropic to generate a response and is not used to train Anthropic's models. Full details are in our privacy policy and subprocessor list.

• Structured AI responses are validated against a schema before use; invalid output is discarded.
• If AI is unavailable or over budget, you get a rule-based result labelled as such — never an invented one.
• Repeated identical requests are served from cache instead of generating new text, which also keeps costs down.

• AI coaching can be wrong. Treat it as a thinking partner, not an authority.
• Free accounts get 3 coach messages a day; you never have to use the coach to train.
• The coach is not a counsellor. See safety if you are struggling.
• Questions or concerns: contact us.