Decision Intelligence in mental wellbeing

A risk model estimates an outcome. Decision Intelligence asks what action could be appropriate, given evidence, alternatives, constraints, uncertainty and human choice.

This is a research direction for Dzvin, not an autonomously validated clinical decision system.

QDIP is a distinct Decision Intelligence technology track exploring hybrid rule/model signals, policy governance, explanation, audit and monitoring.

Dzvin is an application context for testing those ideas in wellbeing. A QDIP component's presence does not prove fully qualified production integration into Dzvin.

Personal trajectories and changing baselines may provide richer signals than a one-off score when data are intermittent.

Evaluation: simple baseline comparison, false alarms, detection delay, missing-data robustness.

When observations are sparse or noisy, the system should express uncertainty and be able to withhold a risky recommendation.

Evaluation: calibration, selective risk, abstention frequency and robustness on incomplete time series.

Explore a transparent path from observations to alternative actions, constraints, an explanation, user choice and feedback.

Evaluation: rule-only versus ML-only versus hybrid baselines, audit completeness, explanation comprehension and user override.

Test whether decision-aware, user-controlled support improves an agreed proximal outcome relative to fixed suggestions.

Start with feasibility work. Randomized or micro-randomized evaluation requires adequate protocol, ethics and participant protections. No study outcome is claimed.

Adaptive interventions and Bayesian methods already exist. We do not claim inventing decision science, causal inference or mental-health personalization.

Potential contribution lies in the domain-specific, testable combination of policies, uncertainty, explanations, feedback and human control. Novelty remains subject to prior-art assessment and experimental evidence.