Skip to content
Research

Measured, not marketed.

Peacefull is in BETA-1. There are no published outcomes to show. What there is: a public model card, a pre-registered research program in design, and a set of commitments about what we will publish when the evidence comes in — including the numbers that do not move.

Two documents

Start here.

Model card

Model card v0.2.2

What the model is and is not. Training and fine-tuning posture, the six-layer safety evaluation, named failure modes, dependency design commitments, escalation behavior, data and privacy, and the version log. BETA-1. Every change logged.

  • Six-layer safety evaluation (incl. bias & equity)
  • Ten named failure modes, plus the ones we haven't found yet
  • Deterministic escalation path — not model judgment
  • FDA Predetermined Change Control Plan structure
Read the model card
Research program

The studies in design

Why we publish nothing yet, what we are running, and the commitments each study carries. OBS-90 observational study first, then a hybrid effectiveness-implementation trial, then a micro-randomized trial of specific companion decision points.

  • OBS-90 · 90-day observational study (IRB submission in progress)
  • Hybrid effectiveness-implementation trial (in design)
  • Micro-randomized trial of decision points (months 4–6)
  • Pre-registration · intent-to-treat · bias reporting
Read the research program
Research program · pre-registeredThree studies in sequence. Null and negative results published as prominently as positive ones.
  1. Study 01Peacefull-OBS-90IRB submission in progress

    90-day observational study

    Question. How do patients actually use the companion day-to-day?

    Design. Prospective · observational · non-interventional. No randomization, no comparator arm.

  2. Study 02Hybrid-EIIn design · begins after OBS-90

    Effectiveness-implementation trial

    Question. Does it help, and can a practice actually run it?

    Design. Hybrid type-I. Outcome change (PHQ-9, GAD-7, PROMIS, alliance) + implementation conditions.

  3. Study 03MRTMonths 4–6 · in design

    Micro-randomized trial

    Question. Which companion actions, at which moments, move measures?

    Design. Within-person randomization at decision-point level. Proximal outcome measurement.

Modalities

Built on the work that works.

BA

Behavioral Activation.

The between-session scaffolding for depression — activity scheduling, values-linked action, and the small experiments that reconnect mood to behavior.

DBT

DBT distress-tolerance skills.

TIPP, radical acceptance, opposite action — the skills a companion can genuinely reinforce in the moment a patient needs them.

MI

Motivational Interviewing.

For ambivalence. Trained against the MI Treatment Integrity code, scored by clinical evaluators, with a trauma-informed stance throughout.

Instruments

Validated, between-session.

In current scope.

PHQ-9 (depression). GAD-7 (anxiety). Patient-initiated, surfaced to the supervising clinician.

Near-term roadmap.

PROMIS Anxiety and Depression short forms. The WHO-5 Well-Being Index. The Columbia Suicide Severity Rating Scale (C-SSRS) integrated with the deterministic escalation path.

Safety planning.

Where a supervising clinician has scoped it into care, the companion supports co-construction and review of a Stanley-Brown Safety Plan.

Evidence base

Every paper, every framework.

The guidelines, validated instruments, AI governance frameworks, and regulatory guidance the model card references — each with a one-sentence note on how we use it and a link to the canonical source. Twenty-five entries, grouped into seven categories.

Browse the evidence base
Access

Academic partners, reviewers, auditors.

Good-faith replication requests, protocol reviews, independent audits — write to clinical@peacefull-ai.io. Model card concerns — modelcard@peacefull-ai.io. Safety reports from patients, clinicians, educators, or researchers — safety@peacefull-ai.io.