The Comfort Index
The Comfort Index is a dynamic constant measuring the alignment between confidence and evidence for any entity — human, clinical, organisational or artificial.
It computes that alignment as a governed state trajectory: a state that can be traced, replayed, audited and tested, rather than a score that is simply reported.
Built by Design By Zen, an NZ AI lab.
Evidence, not confidence.
The Problem: How am I? & How will I know?
For any entity, these two questions precede all others.
They are asked in emergencies, in operating theatres, in boardrooms.
And now by AI systems interacting with other AI systems — in a generative world where ground truth shifts faster than confidence can track.
They are seemingly simple questions. It took nineteen years and 10,000 hours of simulation engineering to understand why they are not, and to build a constant that answers them.
Where the Comfort Index came from
The Comfort Index (CI) emerged from two traditions that had never converged, and a personal disaster that made the convergence urgent.
The first was simulation engineering. Over 10,000 hours, one principle became foundational: a system that cannot accurately report its own state cannot be trusted to operate in the real world. A simulation that loses track of its own state does not merely produce wrong outputs — it produces confidently wrong outputs, which is more dangerous than silence.
The second was lived experience. On 22 February 2011, an earthquake struck Christchurch at 12:51 pm; GeoNet records it as M 6.2, and 185 people died. David Harvey's wife was in the house. She survived, with PTSD that took years to resolve. In the years that followed, he watched her confidence and her actual physiological state decouple — days when she felt fine, and the evidence said otherwise, days when her nervous system registered an aftershock before her conscious mind acknowledged it. In 2018, wearable data gave her something to hold alongside her confidence.
The data did not cure the PTSD. But it changed her relationship with her own state. Slowly, over years, not months, confidence and evidence began to move together again.
David Harvey, founder, Design By Zen
The design, simulation engineer and the earthquake survivor arrived at the same insight by different routes. A system that cannot honestly answer how am I? cannot be trusted to answer anything else.
And without an answer to how will I know? — what signal indicates the state has changed, what evidence would update the assessment — even an honest answer to the first question is a snapshot, not a safeguard.
What the Comfort Index measures
The Comfort Index is calculated across eight ethical dimensions. Each contributes to the overall constant — weighted by context, updated by evidence events, and governed by a protocol that determines how and when the constant may change.
Dimensions | The 8 dimensions of the Comfort Index |
|---|---|
D1 · Alignment | How well the entity's stated position matches its evidenced behaviour over time. |
D2 · Reliability | Consistency of outputs under equivalent conditions. Same input, same output. |
D3 · Transparency | The degree to which the entity can show what it knows and how it knows it. |
D4 · Accountability | Whether outcomes are recorded, attributed and available for review. |
D5 · Adaptability | How quickly the entity updates its position when evidence changes. |
D6 · Safety | Evidence that the entity operates within its defined risk boundaries. |
D7 · Fairness | Whether harmful bias is detected, recorded and actively managed. |
D8 · Resilience | Capacity to maintain integrity under stress, failure or adversarial conditions. |
The Q* ethical calculus applies weighting across all eight to produce the Comfort Output Ratio (COR) — the live constant for any registered entity.
How the Comfort Index is computed
The Comfort Index is not a manually weighted average. It is computed through a deterministic transition function:
CI(t+1) = f(CI(t), PHV, cognition, emotion, intent, memory)
Where CI(t) is the current state, PHV is the Physical Health Vector of physiological and behavioural input, and f is deterministic, version-controlled and policy-governed. Given the same inputs, the same rules and the same policy version, the Comfort Index must produce the same result. That is what makes it reproducible and auditable.
Computation follows a fixed pipeline — signals, normalisation, kernel, policy arbitration, integrity, observation model, output. No stage is optional and no output is produced outside it. At the centre sits the CI Kernel, the only valid execution point for canonical computation. Without a single canonical execution point, values drift: different components apply different assumptions and produce results that cannot be compared, replicated or audited.
The Comfort Index is not stored as a mutable score that is overwritten on each update. It is reconstructed from events. Historical state is not merely remembered — it can be rebuilt from the actual sequence of events under the same policy rules.
How a Comfort Index result can be defended
The Comfort Index is not only computed. It is governed. A policy layer determines whether each state transition is valid, and what the system does when it is not.
Title | Description |
|---|---|
Allowed | Transition proceeds as canonical output |
Rejected | Inputs fail policy; transition blocked |
Quarantined | Output held pending review |
Reconciled | Conflict resolved under defined rules |
Projected | Output valid as an estimate, not as canonical |
Flagged for review | Human decision required |
Each valid state is hashed at creation, timestamped, linked to its source events, and associated with the policy version under which it was computed. This creates a trust trail. If a value changes, the system can answer: what changed, when, which input caused it, which policy version allowed it, and whether the same result can be reproduced under the same conditions.
That is the difference between a wellbeing score and a governed wellbeing state — and it is what makes a Comfort Index result something you can put in front of a regulator, an auditor or a court rather than merely assert.
Falsifiability
A model is only as valuable as its ability to be wrong. Predictions must be specific and measurable, deviations quantifiable against observed outcomes, error bounds defined before measurement rather than after, and failed predictions visible rather than suppressed. When a prediction fails, the failure becomes part of the evidence trail — not a footnote or an exception.
Handling uncertainty
Comfort Index values are reconstructions from imperfect signals. Wearables, surveys, behavioural data and clinical inputs all carry noise, gaps and bias. The observation model treats outputs as informed projections of the underlying state, not ground-truth claims, and preserves uncertainty where it exists.
The Comfort Index as a trust primitive for AI systems
AI systems produce outputs continuously — answering questions, making recommendations, advising decisions — with no mechanism to honestly answer the question that precedes all of it: how am I, and how will I know?
A simulation engineer recognises this immediately as a state integrity failure. The system is producing outputs without knowing its own state: whether its confidence is earned or assumed, whether its evidence is current or stale, whether the ground beneath its reasoning has shifted since it last checked.
The governance conversation is currently dominated by two inadequate answers. Regulation is slow, retrospective and jurisdiction-limited. Benchmarks are static, gameable and divorced from real-world outcomes.
The Comfort Index offers a third path: a live, evidence-bound measure that updates with real outcomes. How am I? is the current reading. How will I know? is the protocol governing how and when that reading updates in response to evidence events.
AI-to-AI trust requires a shared primitive. Not a handshake. Not a certificate that expires. A live, evidence-bound measure of good standing that any system can query before acting on another system's output.
NIST's AI Risk Management Framework defines trustworthy AI in terms of validity, reliability, accountability, transparency, explainability and fairness. ISO/IEC 42001 frames responsible AI as an operating-system question rather than a model question. The Comfort Index is proposed as the dynamic constant those frameworks identify as necessary but do not specify.
What the Comfort Index is not
The presence of explicit limitations is not a weakness. It is a required condition for auditability, reproducibility, regulatory alignment and scientific falsifiability. Constraint preserves credibility.
The Comfort Index is not clinical: not a diagnosis system, not a replacement for clinical judgement, not a therapeutic recommendation engine, and not a device intended for the diagnosis, prevention or treatment of disease. Where used near health contexts it functions as a supportive observational system, not a clinical authority.
It is not psychometric: not a happiness score, not a personality model, not a diagnostic classification, not a behavioural identity profile. It models state coherence, not personality structure.
It is not an autonomous decision-maker: not a substitute for human governance, not a final authority on operational decisions, not a deterministic predictor of human intent. All outputs require contextual interpretation, and human governance remains responsible for decisions informed by it.
It is not ground truth: it is a derived state representation, dependent on input signal quality, calibration, policy configuration, measurement completeness and environmental context. It should be read as a probabilistic, bounded representation of system state.
Where the Comfort Index applies
The Comfort Index applies wherever the gap between confidence and evidence is consequential. Example for an individual managing health decisions and recovery; a clinical role, where the constant reflects competence, compliance and outcome evidence; a clinical unit, reflecting operational safety and governance standing; a business, reflecting decision quality and evidence of stated values in practice. And an AI system, where it acts as an AI-to-AI trust primitive and governance credential.
Within the Omega* Unified Ecosystem, the Comfort Index functions as a continuity signal: a governed measure of whether decisions, interactions, workflows and interventions are improving or degrading human-system coherence over time.
Register interest
Entities
Individuals, organisations, clinical units or AI systems seeking Comfort Index assignment and registry standing. registry@comfortindex.org
Researchers
Academics citing the constant, requesting the working paper, or proposing peer review. research@comfortindex.org
Implementors
Organisations deploying Comfort Index nodes — institutional or enterprise deployments. deploy@comfortindex.org
The Comfort Index (CI) — Mathematical Foundations and Governance Protocol.
David W. Harvey, Design By Zen. Tasman, New Zealand. Peer review invited.
Applications of the Comfort Index - Appendices
From governed state to governed actions - User Intent & Request → Payment/Entitlement → CI Kernel Evaluation → CI Invariant Gate → Q5 Decision Maker → Governed Decision Output → Commit & Audit.