AEA-STATE/1 version 1.0 / Evidence
Inspect it. Run it. Check it.
The public record format comes with source, reference vectors and repeatable checks. The results below describe host tests and compiled binaries. Hardware timing and stack depth are not yet measured.
TRL 3, with fully reproducible public evidence; not yet run on target hardware or in flight; seeking evaluation partners.
Reproduction
Start with the public repository.
Use the source archive from the public GitHub repository. It includes the verification scripts, Rust source, C header and test vectors. Python runs without external Python dependencies; full build checks also need the pinned Rust toolchain and the relevant cross-compilation targets.
Open the GitHub repository ↗Windows, Linux or macOS
python verify.pyReference result: 7 passed, 0 failed. The script reports skipped build checks if the Rust toolchain is unavailable; a partial run is not a full reproduction.
Shell and Rust checks
bash verify.sh
cargo testReference results: shell verification 6/6 and cargo tests 14/14. Use a Linux-compatible shell and the documented build tools.
Canonical reference
One Cambridge record. Four ways to check it.
The test vector digest matches the Rust reference, C caller, Python verifier and a standalone SHA-256 calculation over the canonical first 104 bytes.
9f6cc1096562729a902167fd622e66a264a239f3e9a9d804414659e3ab97800eThe 20 independently generated vectors are compared with the Rust implementation. Host tests also run on big-endian s390x under QEMU; this is emulation, not a flight or hardware result.
Digest behavior
The six exceptions are documented.
The unkeyed SHA-256 digest is a check for accidental corruption of the record's first 104 bytes (bytes 0–103). In the exhaustive single-bit test it detected 1,082 of 1,088 flips; the 6 exceptions are documented ±0.0 sign flips. It does not protect against deliberate changes, because anyone can recompute it.
The six documented exceptions in the exhaustive single-bit test are sign changes on zero-valued floating-point fields. The format canonicalizes negative zero to positive zero before hashing.
CI includes fuzz targets for byte verification and sealing round trips. Passing tests describe the tested inputs; they are not a guarantee for every possible input.
Compiled footprint
Approximately 2.3 KB on the listed Cortex-M targets.
| Core | Target | Footprint |
|---|---|---|
| Cortex-M0 | thumbv6m-none-eabi | 2,329 B |
| Cortex-M4 / M7 | thumbv7em-none-eabihf | 2,315 B |
| Cortex-M33 | thumbv8m.main-none-eabihf | 2,319 B |
These are cross-compiled size results. Execution time, worst-case execution time, stack high-water marks and on-target behavior are not yet measured.
Inspect size results ↗Technical evaluation
Inspect the record. Reproduce the evidence.
An evaluator integration kit is available on request.