Harmonic announced Aristotle's Rust support on November 16, 2026. It describes checking correspondence to specifications, including support for unsafe code. These are the developer's capability claims; this digest has not tested the system.
A dated debate about proof visibility
The November 16 Hacker News archive lists the launch. Its discussion showed 112 points and 61 comments when accessed November 17. Participants asked which Rust semantics were covered and whether specifications and proofs could be inspected. This is evidence of attention on one forum, not a measure of agreement across developers.
The documented verification boundary
The official documentation describes translation at Rust's MIR level into a proprietary Lean model. Results identify verified functions, assumptions and excluded scope. The documentation acknowledges incomplete coverage of undefined behavior and excludes termination, concurrency, leaks, liveness and deadlock freedom.
Functional postconditions describe behavior if a function returns. Proof source is not returned, and compiling supplied specifications does not replay the proofs. The translation, compiler and checker remain assumptions. These limitations make the exact property and its conditions essential parts of any verification claim.
A parser specification exercise
Original illustrative exercise, not executed: specify a parser over a synthetic byte buffer. Record its permitted input lengths, promised output relation, panic conditions and dependencies in separate fields. Add an explicit entry for whether termination is established. This exercise designs a reviewable specification; it does not claim Aristotle verified the parser.
Sources accessed November 17, 2026. Views are my own and do not represent any employer, past or present.
I would review the verification boundary alongside the result. For that parser, I would want the output property, allowed inputs and panic behavior stated separately, with excluded dependencies visible. Easier proof generation is useful, but the documentation's limits and the discussion's questions make me reluctant to describe the whole program as safe after one property succeeds. Inspectable specifications help; they should not be mistaken for independently replayed proofs.
Sources and evidence
Read the verified property, assumptions and exclusions together, and distinguish an inspectable specification from access to a replayable proof.