$ h5i app check
Can a web backend be proved correct and still be fast?
h5i-app is a Rust framework whose kernel, the code that makes the decisions, is extracted to Lean 4 and proved there. This site holds three benchmarks for it: how much real code it can verify, what that costs at runtime, and whether a verified kernel can replace firmware logic on a device.
Verification
Security logic from six open-source Rust servers, ported to h5i-app and stated as Lean 4 theorems. How many are proved, how much upstream code they cover, how much proof it takes, and how fast the port is against the original.
Open the verification results → Benchmark 2 · HTTPWeb performance
Real REST APIs (RealWorld and linkding) written in Rust on h5i-app and in Go, TypeScript and Python, on the same PostgreSQL data. Every implementation passes the same conformance suite.
Open the performance results → Benchmark 3 · RenodeEmbedded firmware
Decision logic from open-source embedded Rust firmware (ESP32, RP2040, RP2350) ported to h5i-app and proved in Lean 4. The original and the port run side by side on the target instruction set in Renode, on the same input.
Open the embedded results →What each one answers
- Verification asks how much of a real server's security logic fits in a verified kernel. A theorem proves one property of a function, not all of its behavior. Parsing, the database and I/O around the kernel are trusted. Source:
prove/. - Web performance asks whether a server built that way is slower than one written without proofs. The h5i-app servers run their kernels on every request. Source:
performance/. - Embedded firmware asks whether a verified kernel can stand in for the decision logic of real device firmware. The original and the port run on the target instruction set in Renode; I/O, the HAL and the device models are trusted. Source: the
prove/ports-embed/ports. - All are runs on one host. The scripts, raw results and summaries are in the repository.