CBT-inspired moves
Track repeated statements, surface alternatives, and preserve uncertainty until enough evidence accumulates.
/science
The stack behind the room.
This is the long-form explanation: what ideas shape behavior, how continuity is computed, and where this system intentionally stops.
foundational intent
What the system is
Moc. is a reflective continuity environment. It is built for repeated thinking, not for proving someone right about your mental state. It should help you carry useful context and call out when context becomes stale.
- Scope is bounded. No treatment claims. No diagnostic claims. No hidden personality coaching.
- Continuity is explicit. What carries over is intentionally tagged and versioned.
- Correction is structural. User changes are treated as first-class state changes.
Psychology foundations (borrowed, not claimed)
MI-OARS influence
Open questions, reflective statements, and short summaries replace persuasion-style prompting.
Person-centred style
Warmth and autonomy-first language reduce coercion and keep the user in control of trajectory.
Schema-aware tracking
Recurring self-descriptions are tracked as hypotheses, not facts, until confirmed by user continuity.
Architecture callout
Think of moc. as a layered control system: capture, vet, compose, constrain, and persist only what can be defended later.
Signal route (architecture)
Safety gate decision map
The system always keeps an explicit branch where risk or uncertainty diverts to deterministic care-safe responses.
Technical architecture (practical overview)
| Layer | Responsibility | Why it exists |
|---|---|---|
| Input surface | Web and app clients collect text/voice/camera signals and hand them to the request layer. | Keep capture simple and user-controlled. |
| Safety and routing | Run risk checks, hard-coded crisis paths, and channel policy before continuity logic. | Never let optimization logic skip critical safety gates. |
| Response kernel | Build the turn response with context limits and claim boundaries. | Balance usefulness with confidence honesty. |
| Continuity ledger | Persist only bounded thread candidates with provenance, confidence, and timestamps. | Make long threads debuggable and user-correctable. |
| Repair path | Apply corrections, reconcile stale claims, and reduce overconfident resurfacing. | Protect the user from stale narratives becoming truth. |
Infrastructure sketch
- Client surfaces -> API ingress -> app backend services.
- Backend request orchestrator -> safety checks -> continuity graph service.
- Continuity graph -> session store -> retrieval layer for next turn.
- Observation records -> export/deletion surfaces -> user-initiated review path.
When claims are constrained
Moc. does not claim certainty unless evidence, context, and user confirmation agree.
- No diagnosis, no prognosis, no treatment replacement path.
- No hidden scoring claims. Confidence is explicit and revisable.
- No silent memory inflation. Carrying context is a gated decision.
How this is different from a generic chatbot page
Most conversation systems are optimized to continue dialogue. moc. is optimized to continue your thread with correction control and a bounded memory model.
- Continuity is visible: sources and provenance are preserved.
- Uncertainty is allowed: silence or lower-confidence responses are acceptable outcomes.
- Safety is deterministic: there are fixed escalation branches, not improvised escalation.