Insurance

The formula that prices a claim’s cost of corruption also prices insurance against it. Given a failure probability for each atom, and assuming the atoms fail independently:

P(φ1 ∧ φ2 fails) = 1 − (1 − p1)(1 − p2)
P(φ1 ∨ φ2 fails) = p1 · p2

A broken atom fails with probability 1; an atom without a given probability is treated as never failing.

When an atom occurs in several parts of a formula, the parts are not independent. sion_assurance::insurance::failure_probability handles that exactly by conditioning on each shared atom (up to a caller-chosen limit, and never more than 30 shared atoms). Correlated atoms, such as two TEEs from one vendor, need an explicit correlation model, which this module does not have.

Probabilities are floating point. That is allowed here because premiums are reports, not consensus values; no consensus-critical code uses floats.

Two limits of the model

• Undetectable lies stay cheap. Insurance and cost of corruption both depend on falsehoods being provable. Claims that are rarely disprovable carry a detection factor below 1.
• Prices come from outside. The ledger proves formulas and bonds. The atom probabilities an insurer uses are its own judgment.

In this implementation

• Built: failure_probability as a library function, with tests.
• Not built: any interface to it (no CLI command, HTTP route or MCP tool), insurance premiums or an assurance marketplace.