How LARI research is tested and reviewed.

This six-step workflow describes how experiments are framed, checked, documented, and approved. It is the research method—not LARI’s technical architecture.

1. State a question and observable outcome

   Define behavior in checkable terms, including success, failure, and inconclusive outcomes.

2. Build or select a candidate behavior

   Keep candidate behavior distinct from existing infrastructure and authored procedures. Reusing a pre-existing routine alone does not show that a capability was learned.

3. Compare against a baseline

   Record the starting behavior and environment so readers can see what changed and which controls separate the candidate’s effect.

4. Verify scope, transfer, and failure cases

   Test the conditions named by the claim. Identify fixtures, fresh-process checks, unseen cases, and important paths that remain untested.

5. Retain or reverse the change

   When persistence or rollback is part of a study, measure those operations directly. Do not assume one surface’s result applies to every path.

6. Publish only an approved record

   An authorized reviewer checks public wording, measurements, provenance, privacy, and limitations. Only allowlisted fields enter the static snapshot.

A passing workflow does not establish that every kind of learning, verification, persistence, or rollback works generally.