What Scotoma measures - and what it does not

Increasing reliance on AI outputs with reduced independent reasoning, verification, explanation, and disagreement.

Reduced opportunity to practice and maintain skills because AI handles tasks that previously built or refreshed expertise.

AI becomes background infrastructure, influencing decisions and workflows while becoming harder to notice, govern, or remove.

Temporary AI support intended to help learning becomes permanent support that prevents independent mastery from developing.

Explicit item mapping

Every scored item is mapped to one risk signal and one subfacet in the assessment schema. Attention checks and context items are excluded from scoring metadata.

Subfacet coverage

Each subfacet has at least two mapped scored items before it appears in the advanced report layer. Thin subfacet coverage is treated as an instrument-development issue.

Attention checks

Attention-check items are tracked separately and never contribute to profile, radar, risk-signal, or subfacet scores.

Reverse-coded review

Reverse-coded items are included as quality checks for response consistency and reviewed before stronger reliability claims are made.

Claim discipline

Public report copy is checked for overclaiming. Signals guide debrief and follow-up checks; they are not calibrated probabilities or validated diagnoses.

The Craftsperson

AI as an instrument under human skill and control.

Evidence basis: Tool metaphors, human agency, and augmentation framings in AI adoption discourse.

Caution: Can sound like a stable identity; report as current framing pattern instead.

The Symbiont

AI as a working human-AI system with mutual task coupling.

Evidence basis: Human-AI teaming, complementarity, and augmentation/automation paradox literature.

Caution: Do not imply balanced partnership; AI has no accountability or stakes.

The Accelerator

AI as speed, scale, throughput, and operational leverage.

Evidence basis: Productivity, work redesign, task decomposition, and automation literature.

Caution: Do not infer that productivity gains are net beneficial without capability and learning checks.

The Sentinel

AI as a system that needs boundaries, vigilance, and governance.

Evidence basis: Automation bias, safety, ethics, governance, and professional responsibility literature.

Caution: Do not frame caution as resistance or low AI maturity.

The Architect

AI as a force that redesigns expertise, learning, roles, and organizational capability.

Evidence basis: Work design, sociotechnical systems, learning design, and expertise-development literature.

Caution: Reflection is not itself evidence of good implementation; check enacted practices.

The Naturalist

AI as normalized background infrastructure in everyday work.

Evidence basis: Infrastructure, routines, invisible work, and cognitive extension literature.

Caution: This is the most exploratory profile; avoid hard claims until behavioral data supports it.

Content validity first

Every construct needs a clear definition, literature basis, item coverage, and intended interpretation before stronger measurement claims are made.

Response-process validity next

Cognitive interviews should test whether respondents understand items as intended and whether wording creates unintended shame, defensiveness, or moral ranking.

Reliability after dimensionality

Internal consistency should only be reported after checking whether items behave like reflective scales. Some constructs may be formative and should not be forced into alpha/omega claims.

Criterion evidence before prediction

Claims about deskilling, expertise preservation, or implementation outcomes require external behavioral evidence, not only self-report patterns.