Deployable task coverage × Epistemic integrity

The enclave becomes a standards nursery — with two structural risks. Standards designed for hospitals, courts, and banks get exported with their economic assumptions intact, then imposed on schools, newsrooms, and small firms where the power relations differ. And the corner teaches the wrong lesson: that verifiable reality is too expensive to universalise — reality as a premium tier, optional outside elite sectors. Redeeming: enclaves produce the auditors, evidence lawyers, and model-risk teams who could later govern broad deployment, if the standards are kept open and portable.

Trace institutions become strategic infrastructure: whoever defines admissibility, audit standards, and certification shapes what counts as real. Accountability improves but gains procedural drag — and high integrity can paradoxically unlock more automation, because regulators and insurers can approve deployments they could otherwise never bound. The dark version is the verified cage: everything documented and certified, with the truth layer owned by platforms and states that ordinary people cannot challenge.

The fog teaches bad habits before the stakes arrive. While AI is cheap and common but workflows don't yet depend on it, verification goes interpersonal — people learn to trust the colleague, not the system — and everyone quietly normalises laundering outputs into ordinary documents, skipping source trails, treating hallucination as background noise. Institutions misread the low blast-radius as low risk and defer the boring standards work. The redeeming read: low coverage limits systemic dependence, so this corner can buy time to observe real failure before AI is doing the real work — if anyone treats it as the rehearsal it is.

Human review collapses from checking into triage. When useful AI covers whole production workflows and no social truth layer scales with it, oversight becomes sampling, vendor assurances, anomaly detection, and scandal response; liability migrates to whoever is visibly near the decision, not whoever caused it. Reality becomes subscription-tiered — wealthy institutions buy verification, ordinary citizens get disclaimers. Recursive-leverage domains (code, cyber, persuasion, AI R&D) deepen the fog by producing opaque machinery that then produces opaque facts. The redeeming pressure: the density of harm eventually creates surfaces that courts, insurers, and professions can act on.