A learning session

A session is one focused conversation with the AI through the seven-step learning cycle. Sessions are short — 15-45 minutes is typical. The Dashboard's "Start session" button creates a new one; the app picks the learning method (your dominant from the assessment) and the cycle's starting step (usually 1 = Input).

The seven steps

#	Step	What happens
1	Input	The AI presents new material in the active method's style
2	Attempt	You apply what you just learned — the AI poses a task
3	Error	A mistake shows up; the AI marks it precisely
4	Feedback	The AI explains the correction in depth
5	Adapt	You adjust your approach; the AI may re-frame
6	Repeat	A fresh task that exercises the same concept
7	Integrate	The AI connects today's material to broader context

The cycle is a framework, not a conveyor belt. Steps can repeat, skip forward, or even step backward when your last turn shows confusion. The AI judges per round-trip and the app's progress bar updates accordingly.

The cycle in detail

How the AI guides you

Every message you send triggers up to three AI calls:

1. The learning reply — streamed token-by-token via SSE. You see the inline cursor (▍) while the assistant thinks; tokens land in the bubble as they arrive (no "Thinking..." placeholder). The system prompt is composed from a 42-cell matrix (6 methods × 7 steps), so a deductive Input feels very different from a contextual Repeat.
2. The step evaluator — a second AI call reads the exchange and decides whether you're ready to advance. It emits advance, confidence, reason, suggested_step. The app applies the suggestion when confidence ≥ 0.6.
3. The topic-transition evaluator (only at step 7) — a third AI call decides whether the topic has been integrated. If yes AND continue_recommended, a new cycle starts automatically with a fresh subtopic (auto-loop, max 5 cycles per session).

The verdict is shown discreetly above the chat as a "Step moved from X to Y because…" notification when it actually applies. Cycle-transition cards render as dashed-border "Cycle N" cards in the chat history.

Voice on / off — a TTS button (▶) next to each AI reply reads it aloud; a microphone button (🎤) on the input lets you dictate; interim transcripts populate the textarea so you can review before sending. Both are Web Speech API; toggle in Settings → Voice.

Cycle-progress indicator

Across the top of the Session page sits a 7-circle progress strip. The current step is filled in your project's accent color; passed steps are filled more faintly; future steps are empty. When the evaluator moves you forward (or backward!), the strip animates to make the transition visible.

On mobile (≤768px) the strip becomes a single horizontal row of small circles to save vertical space.

Method-switch recommendations

Sometimes the active method just doesn't land. After three sessions where your "understanding" rating doesn't grow and your "stress" rating stays high, the app surfaces a MethodSwitchBanner: "Want to try [other method] for the next session?". Accept and the next session starts with the new method active.

The recommendation reads your profile and prefers your second-strongest method that you haven't used recently. You can dismiss the banner; it'll come back if the stagnation pattern continues.

Both storage modes (Server + Local) support method-switch recommendations.

Rating + ending a session

The Session page has an "End session" button. Before the session closes you fill a short rating: understanding, stress, and method-fit on a 1-5 scale, plus an optional rich-text note (TipTap: bold, italic, lists, code blocks with syntax highlighting, links). The note is yours — the AI doesn't read it.

The ratings + multi-cycle journey summary turn into a ProgressCommit row — the Git-style snapshot of one session. Completing a session awards XP (50 base × streak multiplier, plus per-cycle bonuses), checks for newly-earned badges, and updates your streak. See Progress, Dashboard, and the Tracking concept.