All blog posts
Research 3 min read

Protocol Standards Research: Why Test Vectors Come First

Secure Net's standards research found that independent implementations need named registries, canonical encoding and tests of what must be rejected.

By Comutability Published Update from
Illustration of a protocol specification being checked against test vectors and validated across multiple systems.

Research record from 13 November 2026. This article discusses published project research and implementation evidence, not independent certification.

Secure Net's standards research found that independent implementations need named registries, canonical encoding and tests of what must be rejected.

A protocol is more than the code that currently runs

If two sites agree to exchange messages but calculate different bytes for the same object, a signature can verify on one machine and fail on another. Secure Net's early specification work found that many shared names lived only inside source code. Record kinds, event names, headers, algorithm suites and domain-separation labels needed a language-neutral definition.

The standards proposal also identified the absence of clear compatibility rules. Without explicit versioning and deprecation, contributors cannot know whether an apparently small change will break another client's implementation.

The approach proposed in SNIP-0004

The work adds a framework for normative language and precedence among documents, registries, vectors and code. The registry lists the parts and shared names so one identifier cannot be quietly reused for a different meaning. Conformance profiles let a client state which subsets of the protocol it actually supports.

A part template requires security and compatibility information to be written consistently. The proposal also introduces a lifecycle for deprecated features, rather than allowing them to disappear without notice.

Canonical JSON as a practical test

Digitally signed structured data must have one exact byte representation. Canonical JSON provides ordering and encoding rules that independent implementations can reproduce. The protocol added language-neutral test vectors for signed records, including malformed or forbidden cases that a conforming implementation must refuse.

The engineering notes record a second canonical-JSON encoder written in Python from the specification. It produced a different result from Python's default key sorting on one case, illustrating why merely calling a standard library sort is not enough to establish canonical agreement.

Checks against documentation drift

The initial standardisation work introduced a checker that compared shared names, declared feature versions and the specification registry with the implementation. If a developer added an unregistered label, changed an endpoint without a corresponding specification entry or left a generated file stale, the check was designed to fail.

Later Rust migration work moved test and standards tooling into the Rust implementation. That makes the original principle more important: test vectors describe a cross-language contract, not a preference for whichever implementation happens to be current.

What remains to prove

The proposal itself was recorded as a draft, and a source-level checker does not establish that independent developers can build compatible software. The remaining work includes filling in missing part documentation, generating vectors for all relevant operations and getting third-party conformance evidence.

The research finding is simple but consequential: an open standard is one that another team can implement without guessing. Clear rejection cases are just as important as successful examples.

Primary sources and further reading

Explore other Comutability research ยท Read about the organisation

Sources and further reading

Explore the evidence and original project records behind this article.