Candidate invariants should be allowed to fail cleanly.
The protocol treats both successful and failed claims as useful artifacts. A low-defect claim can become stronger evidence. A high-defect claim becomes an obstruction record showing where a local pattern fails to compose globally.
Positive lifecycle states
hypothesizedwitnessedtransport_testedpersistentreplicatedcandidate_certifiedThe final state is deliberately named candidate_certified. It means the claim passed a declared evidence protocol. It does not mean regulatory certification or a guarantee of scientific truth.
Failure and retirement states
inconclusive
The available evidence does not support a clean pass or fail.
falsified
The candidate fails a decisive test under declared assumptions.
non_persistent
The candidate only survives a narrow or unstable filtration range.
high_defect
The closed-loop transport test produces excessive residue.
dataset_artifact
The result appears to depend on a particular dataset or sampling artifact.
model_artifact
The result appears to depend on a particular model representation or adapter.
Why failure states matter
Automated discovery should not only publish successes. It should also preserve structured failures. Obstruction records help future researchers avoid repeating brittle claims, identify missing transformations, and refine the taxonomy of admissible tests.
Failed invariants are scientific evidence.
A high-defect loop can show exactly where a candidate law stops generalizing: across scale, coordinate system, simulator modality, boundary condition, geometry class, causal intervention, or time evolution.