Antithesis, TLA+ and Veil verification pipeline
Source Code, e.g. etcd or HashiCorp Raft (Documented Properties) -> Antithesis SUT (Assertions; Deterministic Simulation Testing) via the Antithesis plugin. Source Code -> TLA+ Model (Safety Invariants; Model Checking + TLAPS Proofs) via AI inference. Antithesis SUT <-> TLA+ Model: Trace Validation. TLA+ Model -> Veil Spec (Safety Invariants; Lean Proofs) via LLM translation. TLA+ Model <-> Veil Spec: Bounded Equivalence Checking. All three checks pass.
Source Code
(e.g., etcd or HashiCorp Raft)
Documented Properties
Antithesis SUT
Assertions
Deterministic Simulation Testing
Passing
Failing
TLA+ Model
Safety Invariants
Model Checking + TLAPS Proofs
Passing
Failing
Veil Spec
Safety Invariants
Lean Proofs
Passing
Failing
Antithesis plugin
AI inference
Trace Validation
LLM translation
Bounded Equivalence Checking
fix bugs
fix bugs
repro counter-example
fix bugs
fix model
How did we get here?