Governance Thresholds: How Tight Is Tight Enough? | Geometry of Trust | Governance - Lesson 3

The structural governance from the first three talks decides whether two AI agents are allowed to interact. This talk is about the quantitative layer sitting on top — the thresholds that decide how strictly their evidence is held to numerical bounds once they do.

An important admission upfront: every number in this talk is illustrative. Not prescriptive. The values — 0.02 for critical infrastructure, 0.03 for healthcare, 0.05 for finance, 0.10 for commercial supply chain, 0.25 for research — are placeholders chosen to show the *shape* of a tiered framework. They are not recommendations. The real values have to come from domain regulators working with operators, auditors, and standards bodies, informed by actual deployment data.

What the framework contributes is the shape, not the specific numbers. Per-domain thresholds. Per-peer enforcement. Governance-set rather than maths-produced. Revisable as evidence accumulates.

We walk through four scenarios that show how this plays out. A self-driving tractor operating in two domains at once — solved by inventing a dual-purpose vehicle domain and letting each peer apply its own thresholds to the interaction. A same-domain healthcare pair — diagnostic agent advising a drug checker, with both held to full healthcare thresholds despite the asymmetric roles. A supervised finance interaction where the regulator demands and the trader proves. And the case where thresholds don't even get consulted — when a structural exclusion rejects the interaction before any attestation is evaluated.