MindFrame's coach isn't a chatbot you ask for motivation. It's a system of scheduled, scoped, transparent AI passes over your own training data. Here's exactly what it does, what it sees, and where it stops.
Each one runs on its own cadence. Each one has a clear input, output, and a thumbs-feedback loop back to us.
After every session
A 3-sentence story of what just happened: where you were calibrated, where confidence outran , and what the arc of your session says about today.
Input: Your 10–20 attempts in this session — confidences, accuracies, timings, and per-dimension score deltas.
Output: Prose, not scores. Written to be read, not skimmed.
The one thing to do tomorrow
A single focused recommendation — which mode to play next, or which concept in your review deck needs another pass before it fades.
Input: Your weakest dimension this session vs. your trailing 30-day average + which Review Deck items are entering the 're-forget' zone.
Output: One sentence. One link. Designed to fit inside your day.
Weekly proactive insights
A short letter from the coach every Monday: one pattern, one celebration, one recommendation. You never have to ask — it writes itself.
Input: Your full 7-day session history + any manually flagged attempts + prior Coach Insight outcomes (which ones you dismissed).
Output: Inbox-style cards with thumbs feedback — the model learns which insights land.
Monthly essay
A long-form essay naming how you think now — what's changing, what's stubbornly the same, and the working theory of your cognitive style.
Input: Your full month of sessions, reflections, prediction outcomes, and Coach Insight history.
Output: Full essay (600–900 words) with its own shareable OG image and audio narration.
High-confidence-wrong diagnosis
Finds the cluster of attempts where you were most certain and most wrong, names the bias pattern, and hand-picks 3 bespoke challenges to retrain it.
Input: Attempts where confidence ≥ 75 and = 0, across the last 30 days.
Output: A short diagnosis + 3 specific challenges wired for one-tap start.
Real-time, on wrong answers
When you miss a challenge without writing a justification, a soft prompt asks why — with 4 seconds to skip. Pure , no scolding.
Input: The attempt state — wrong, no reasoning captured.
Output: One line of copy. No AI call.
AI is the last step, not the first. Deterministic scoring runs first; the coach reads what it computed.
Answer challenges, rate confidence, optionally write reasoning.
Deterministic scoring computes , , reflection slope, adaptation delta.
When a coach surface triggers (session end, weekly cron, etc.), a scoped prompt runs against the relevant slice of your data.
Every output is visible on-surface with a thumbs widget + link back to the attempts it was built from.
Thumbs-down outputs are down-weighted in future prompt context. The model learns your taste without learning your identity.
Every AI feature on MindFrame is held to the same five rules.
Generic AI coaches produce generic advice. The MindFrame coach works against five measured dimensions — , reasoning, , reflection, adaptation — so every recommendation can be traced to data and evaluated later.
The coach sees attempt-level stats and free-text justifications you chose to write. No cross-site tracking, no guesses about mood. What the coach doesn't have, it doesn't pretend to.
Session narratives cache for 24h per session id. Weekly insights regenerate on the Monday cron. Theory-of-You regenerates monthly. You don't pay — in latency or in noise — for reruns that would say the same thing.
Every AI card has a thumbs widget. Every insight links back to the mode or challenge it's rooted in. You can click through to see the attempts the coach was looking at.
The coaching layer is additive. Turn it off in settings and the dashboard still gives you scores, , and raw analytics — the product still works without a single AI call.
Metacognition training produces a g = 0.63 effect on academic performance (de Boer, Donker & van der Werf, 2018) — but only with specific, immediate, reasoned feedback. Vague praise produces no improvement (Halpern, 2014). The coach exists because the research is clear: feedback quality is the lever, not feedback quantity.
Answer one challenge. No sign-up. You'll see how confidence-rated attempts become the input that the coach reads.
Your first session produces your first Session Narrative within 10 seconds of finishing.