Measurement and Limits

This series opened with a claim: "The evidence points to a testable answer." An honest framework owes the reader the test. This article gives it — how the claim would be measured, and what would prove it wrong.

The field's most significant evidence gap is the lack of controlled workplace outcome data (Bury et al., 2020). The way to close it is for adopting facilities to generate that data themselves, through a common schema measured across four domains and published annually in anonymized aggregate:

The metrics track those already used by corporate neurodiversity programs, ensuring comparability (Juicebox, 2026). Each adopting facility becomes an evidence generator — converting the field's biggest weakness into a structural contribution. Facilities are encouraged to register their protocols and contribute anonymized data to a common repository, building the controlled evidence base the field currently lacks.

The framework specifies its own failure conditions. That is deliberate. Here they are, stated plainly:

1. The productivity figures are corporate self-reports. The headline numbers come from corporate programs, not peer-reviewed controlled studies. Peer-reviewed confirmation exists for specific strength domains; controlled workplace outcome studies across neurodivergent and neurotypical teams do not — yet.
2. The strengths are trait-associated, not universal. They are population-level tendencies surfaced by skills-based matching, not guarantees about any individual. The distinction between "trait-associated tendency" and "individual capability" is load-bearing.
3. AI complementarity requires design. Providing AI tools without deliberate design for neurodivergent working styles does not produce the gains described.
4. The values-alignment mechanism is a hypothesis, not a finding. It rests on converging evidence — justice sensitivity, systemizing cognition, moral distress and burnout — but has not been tested in the specific configuration proposed: whether verified environmental integrity correlates with retention and wellbeing among neurodivergent operators working in AI-paired roles. A second, analytically separable question concerns the pairing itself: the observed finding that neurodivergent operators prefer and recommend AI assistance (UK DBT, p = 0.035) is established; the proposal that AI pairing lowers the cost of cross-neurotype information transfer is untested. The measurement framework above is designed to examine them jointly via a pre-specified mediation analysis.
5. The geographic evidence is uneven. Published evidence is concentrated in North America, Europe, East Asia, and Australia. Generalization elsewhere requires locally generated evidence.
6. The diagnostic scope is narrow. The evidence base concentrates on ADHD, autism, and their co-occurrence — commonly called AuDHD in the community, though not yet an official term in the diagnostic manuals. Other neurodivergent configurations are acknowledged but less represented in the cited literature.

A framework that names what would prove it wrong is structurally harder to greenwash, because it has already said exactly where to look. The same systemizing drive this series describes — the drive to detect the gap between a claim and a utility bill — is the drive this framework invites the reader to turn on itself.

This is not hedging. It is the thesis applied to itself. A community data center is not defined by what it claims; it is defined by what it measures and publishes. The next article shows what one looks like.