LLM tooling

An LLM can work from a detailed specification, report the work as complete, and still miss items. It can write tests that pass without testing what was asked, or fix one race condition while leaving the same race one call over. The tools here turn those failures into checks that fail in CI, so a green run means what it says.

They cover keeping a specification, its tests and its code aligned; model-checking concurrent Elixir code against a TLA+-style description of what it may do; and compiling dependencies once and reusing them safely across projects and pipelines. They were built alongside the applications on this site and are used to build them.

• ExTLAexperimentalTLA+-style model checking for Elixir: describe what a concurrent system may do, check every ordering, then check the running GenServer against the same description.
• SurfexexperimentalKeeps a specification, its tests and its code aligned, and accepts code as implementing the spec only when a test has shown it.
• DepdepactiveCompile an Elixir dependency once and restore it in every other CI job, keyed by everything that went into the build so a restore is never a wrong build.