07 — Tabletop exercise

Summary

The registry walked through on paper before it is walked through in code. Five seats (operator/issuer, agent, verifier, red, facilitator — every one runnable as an LLM session), a baseline timeline from keygen to a verified mandate, and six injects: revocation mid-exercise, a write into another's record, the flood, the compromised processor, the fixture mistaken for an identity, and session death. The facilitator's rule makes it a documentation test too: any question the published pages cannot answer is logged as a finding. Run three times — on paper before phase 0, with real fixtures after phase 2, and live as the phase-4 demo.

Key concepts

• I4 — detectable, not preventable, measured — the gap between commit and detection is the number the MVP's honesty depends on
• I5 — the flag before the signature — the most likely real-world implementation bug, run as a card
• Findings → comms — documentation findings become releases; rule findings go to the project lead

Key ideas

• The failure modes are procedural, not cryptographic — which is what a tabletop surfaces.
• An imagined population, run honestly, is the cheapest first test the four rules will ever get.
• Same script all three runs, so findings are comparable.

Read the document