From discovered candidate to reproducible claim.
The Open Invariant Protocol is the thin waist between automatic discovery and reproducible evidence. It gives every candidate invariant a standard claim object, transport test, defect score, persistence summary, and verification path.
Core doctrine
No invariant without an evidence bundle.
A candidate law should not be promoted as an invariant unless it carries an executable evidence object describing what was tested, how it was tested, what passed, what failed, and how others can rerun the result.
Protocol chain
Mechanistic Invariant
A Mechanistic Invariant is a typed, reproducible claim that a candidate quantity, law, descriptor, equivalence class, or surrogate coordinate remains stable under a declared family of admissible transformations.
| Field | Purpose |
|---|---|
id | Stable registry identifier for the candidate invariant. |
domain_scope | Domain, program, benchmark, and operating context. |
observable_scope | Inputs, outputs, data sources, and observable quantities. |
candidate_quantity | The law, descriptor, coordinate, or relation being tested. |
admissible_transformations | Declared transformations under which stability is tested. |
validation_state | Current lifecycle state of the claim. |
evidence_pointers | Links to evidence bundles, datasets, code, runs, or reports. |
Transport-scored evidence
The protocol asks whether a candidate invariant survives closed-loop transport. Low defect means the candidate composes coherently under the declared loop. High defect produces an obstruction record: evidence that a locally plausible law does not glue globally under the tested transformations.
D(γ) = ||H_γ - I||_F low defect → coherent candidate under declared loop high defect → obstruction / non-gluable claim persistent low → stronger evidence across filtrations persistent high → brittle or spurious candidate