Invariant truths

OKF's central claim is that anyone can consume a bundle, any framework, any model provider, any serving system. Nothing in the format lets you test that.

The purpose of OKF is to embed your knowledge in the corpus, what is true to you , how things connect and operate. This needs to be invariant independently of the consumer engine.

The practical version of this problem is a model upgrade. An organisation moves its analytics agent from one model version to the next and has no way to answer the only question that matters: did the corpus keep behaving the same way? Today the answer is a hope, discovered later through incidents.

"Run all invariants under model XYZ; no change" should be a result you can produce.

Attestation checks that a single run computed a single value the sanctioned way, a mechanical comparison. It is deliberately scoped to the moment a number is produced.

An invariant asks something different: does the corpus entail a conclusion, and does a consumer reading it arrive there? That is a property of the corpus and its reader together, not of one execution.

There is a need for a regression suite embedded in the corpus authoring: outcomes that must keep reproducing.

The issue does not name files, tests, or an entry point; begin by locating the corpus authoring and attestation implementations in the repository. Define how invariant outcomes should be represented and executed across model versions, then establish what a regression result must demonstrate before implementation.

Written by the indexing model from the issue text.