1) Development roadmap (how this evolves)
We treat this as a staged research build: first proving safety and transparent control, then attunement quality, then adaptive personalisation. At this stage we use explicit structures as demonstrations because memory is still session-scoped.
Stage A — Safety foundation (now)
- Prove safe dialogue boundaries, user control, and clear support-signposting pathways.
- Keep monitoring logic transparent with clear plain-language explanations.
- Refine bilingual tone, pacing, and consent-first interaction patterns.
Stage B — Attunement pilot (partner-led)
- Test usability, acceptability, safety comprehension, and engagement.
- Evaluate whether the assistant can hold a safe trajectory over repeated contacts.
- Document where human oversight is needed and how handover works in practice.
Stage C — Adaptive core evaluation
- Evaluate whether longitudinal interaction improves fit, trust, and timing.
- Assess whether adaptive language handling improves safety and continuity.
- Produce a partner-facing implementation plan and publishable write-up.
2) Contact schedule concept (user-paced)
The TRiM-inspired rhythm is a scaffold, not a cage. Some users may want a short, intensive run (days), while others may prefer a slower pattern (weeks). The system should make pace explicit and negotiable, and it should never treat oscillation as failure. Reflection on traumatic experience should be invited only when the interaction suggests enough steadiness, and when psychoeducation has helped the user understand what reflective work is for.
How continuity works over time
Participant code continuity
A participant code links sessions so the companion can maintain continuity and avoid starting from zero each time.
Adaptive support rhythm
The companion adapts pace, tone, and prompt style based on what has been helpful across prior interactions.
Reflective prompts by invitation
When stability is present and psychoeducation has prepared the ground, reflection can be invited in a bounded way, with clear exits back to grounding.
Important: signals can swing. A “worse day” does not mean failure. The point is to help the assistant respond with the right support style (pace, warmth, grounding, containment), and to offer human routes when needed.
Memory reorientation: how reflection should be guided
This reflective layer is not about pushing disclosure. The interaction itself, plus simple psychoeducation, helps determine when reflection is appropriate and what form it should take. Some users may only need grounding and orientation; others may be ready for brief structured reflection.
- Start with psychoeducation: explain in plain language that trauma memory can arrive in fragments, body alarms, images, or loops, and that reflection should stay manageable.
- Use the interaction as a guide: if language is disorganised, highly activated, or shutdown, stay with containment and regulation rather than moving into memory work.
- Orient memory before interpreting it: help the user separate past from present, place events in sequence, and notice what is known, uncertain, or still confusing.
- Offer the least intense reflective route first: brief meaning-making, values questions, or simple narrative stitching before any deeper exploration.
- Keep exits visible: return to grounding, pause the topic, or hand over to human support whenever reflection begins to destabilise rather than clarify.
3) What we are testing (and what we are not)
This prototype prioritises presence, safety, and transparency. Any structured elements are used to test the assistant’s behaviour — not to diagnose a person or quantify recovery.
We are testing
- Whether the assistant maintains a safe therapeutic direction (containment → regulation → gentle exploration).
- Whether language-based orientation can notice rising strain and adapt stance appropriately.
- Whether support-signposting pathways are understandable, localised, and user-controllable.
- Whether later-stage Council perspectives improve engagement without increasing burden.
We are not testing
- A clinical outcome tool or diagnostic system.
- A replacement for therapists, TRiM practitioners, or crisis services.
- A rigid programme that users must follow.
4) Safety, consent, and human oversight
Boundaries are a first-class design constraint. Language orientation is explained and consented. The user can pause, change topic, or stop at any time. Signposting to human support is explicit — never false certainty.
- No pressure into trauma detail or exposure.
- Clear support-signposting pathways with locale-specific options (demo placeholders are acceptable).
- User can view/edit/erase memory (when implemented).
- Research participation is opt-in; consent can be withdrawn.