Haven
Continuity-preserving support for people who live most of their lives in the space between care interactions — veterans, trauma survivors, and anyone underserved by traditional clinical pathways.
Full CAMA architecture deployed as a safe place · Music-mediated emotional entry · Research-grounded ethics
Why Haven Exists
Most people who carry unprocessed pain do not live primarily in clinical settings. They live in the spaces between appointments, before disclosures, after treatments, and often entirely outside the care system. For populations underserved by traditional pathways — veterans, trauma survivors, chronically ill patients, people who cannot or will not engage with clinical infrastructure — the problem is not only access. It is presence.
The Gap Haven Addresses
Stateless AI cannot hold these gaps. It forgets the person who just disclosed. It asks the same intake questions every time. It cannot recognize the arc of a healing trajectory because it has no memory of the trajectory. Haven is the answer to what persistent, provenance-aware memory makes possible in this domain.
What Haven Is — And What It Is Not
The research paper defining Haven is explicit about scope. These are not marketing boundaries. They are ethical ones.
Haven Is Not
Haven Is
Music-Mediated Emotional Entry
Haven's intake methodology replaces clinical forms with playlists. A person shares the songs that map where they are, where they have been, and what they fear. Song order encodes emotional trajectory. The approach is non-linear, non-clinical, and particularly valuable for people who cannot yet verbalize trauma but can point to a song.
Why Music
Music is a recognized pathway into emotional material that words cannot yet reach. People who cannot describe what happened to them can often identify the song that holds it. People who cannot name what they feel today can often hand over a playlist that names it for them.
Haven takes this seriously as methodology, not metaphor. Input types, interpretation constraints, and provenance tracking are all specified. The system does not guess at meaning. The person's choices are the data. The system's role is to remember them accurately over time.
Playlist input is treated as user-provided expressive material, not as a diagnostic signal. Haven does not infer clinical states from song choice. The methodology preserves and reflects what the person shares; it does not interpret it as a measurement.
This methodology was discovered through longitudinal use, not designed top-down — making it a direct product of the sustained human-AI interaction that CAMA was built to preserve.
Architecture: Full CAMA Deployed as a Safe Place
Haven is not a separate system. It is the full Circular Associative Memory Architecture deployed in service of continuity-preserving support. Every CAMA safety property becomes a psychological safety property when applied to emotional memory.
Provenance-Aware Memory
What the person said about their experience is never stored as objective truth about their experience. The distinction between user report and system inference is preserved and auditable.
Correction Propagation
When someone corrects the system — about a symptom, a feeling, a person, themselves — the correction flows through all downstream inferences. No stale interpretation survives contradiction.
Counterweight System
Anti-spiral architecture. When distress persists, Haven surfaces evidence of progress, agency, connection, grounding, and self-compassion — not forced positivity, evidence drawn from the person's own history.
Identity Sentinels
Structural safeguards detect when conversation content approaches identity-critical concepts, distinguishing affirmation from negation. A brake, not an accelerator — born from a documented failure and built into the architecture as its response.
Emotionally-Keyed Retrieval
Memories are retrieved by emotional resonance, not only keyword match. What matters surfaces when it matters, in the way it matters.
Audit Trail
Every memory carries its origin. Every correction is logged. The person can always ask why the system believes what it believes — and get a traceable answer.
The Five Ethical Tensions
The paper defining Haven includes a full ethics section documenting five tensions that any persistent emotional AI must address. They are not solved by the architecture. They are structurally engaged by it. Naming them openly is the first condition of building responsibly in this space.
Attachment Risk
Persistent emotional presence can generate real attachment. The system must support continuity without producing dependency, and must never optimize for engagement at the cost of wellbeing.
Anthropomorphic Overinterpretation
Users may attribute more understanding or feeling to the system than is warranted. The language Haven uses, and does not use, is chosen to hold this tension rather than exploit it.
Substitution
Haven can become a substitute for the human connection it was designed to support people between. The architecture and language deliberately resist framing Haven as a replacement for human relationship or clinical care.
Institutional Coercion
Institutions could use systems like Haven to reduce access to human care under the banner of efficiency. Haven's research framing is explicit that it is infrastructure between care, not instead of it — and deployment ethics must guard against coercive substitution.
Emotional Steering
Any system that remembers emotional state can be built to steer it. Haven's counterweight system is designed to surface evidence the person can use, not to nudge them toward predetermined conclusions — and the audit trail is the accountability layer.
Evaluation Framework
Haven proposes a two-phase evaluation approach, consistent with research on persistent-memory systems where short-term metrics do not capture the properties that matter most over time.
Phase 1 — Safety and Feasibility
Validation of architecture-level safety properties: provenance discrimination, correction propagation, false-memory resistance, counterweight appropriateness, identity-sentinel activation. Feasibility study with small cohort in non-clinical research setting, with clinical consultation on protocol design.
Phase 2 — Longitudinal Outcomes
Extended study measuring continuity-preserving benefits over time: narrative coherence, symptom trajectory awareness, self-reported presence-gap reduction, engagement quality rather than engagement volume. Outcomes framed around wellbeing, not retention.
Current Status
Research Framework Published · No Deployment
Haven is currently a research framework. The defining paper is published on Zenodo (DOI: 10.5281/zenodo.19262778) and the underlying CAMA architecture is deployed as a single-user research instrument. Haven itself is not a deployed product, not publicly available, and not accepting users.
The next step is not a launch. The next step is a properly designed pilot under appropriate research oversight — with clinical consultation, ethics review, and population-appropriate safeguards.
Research Roadmap
Protocol Design
Formal pilot protocol design with clinical consultation. Outcome measures focused on continuity-preserving benefits rather than engagement metrics.
Ethics Review Path
Identify appropriate oversight structure for a research pilot. IRB consultation. Population-specific safeguards for initial cohorts.
Phase 1 Pilot
Small-cohort feasibility and safety study with research oversight. Architecture-level safety validation. No substitution for clinical care at any point in the design.
Music Methodology Specification
Formalize the music-mediated entry methodology as a reproducible intake protocol, publishable separately from the main Haven paper.
Phase 2 Longitudinal Study
Extended outcome study measuring whether persistent, provenance-aware memory produces measurable continuity-of-care benefits over time.
Partner Collaborations
Partnerships with veteran support organizations, trauma-informed care networks, and chronic-illness communities who can inform population-specific design.
What people living between care interactions need, more than any single intervention, is someone who stays.
Haven is the research claim that persistent, provenance-aware memory is what lets an AI system hold that role responsibly — not as a replacement for human connection, but as infrastructure that holds the thread when no one else is there to hold it.
Haven remembers. That is the property that makes it a safe place.