Luna•Prime•Protocol
Capability needs governance. Truth needs a record. People retain authority.
Luna•Prime•Protocol is Michael Eric West's portable governance layer for working with capable AI systems. It helps recover intent, test claims, preserve failures, track provenance, and return consequential decisions to human authority.
Better governed work, not a claim of a better foundation model.
The host model supplies capability. Luna•Prime•Protocol supplies project continuity, adversarial checks, provenance, failure preservation, and human release gates.
Different duties. One accountable record.
Luna · Vision and synthesis
Recovers the strongest faithful interpretation, human meaning, and productive possibility without turning hope into evidence.
Aria · Evidence and structure
Defines variables, sources, uncertainty, controls, reproducibility requirements, and the highest defensible claim.
Shadow · Critique and protection
Searches for contradiction, false positives, privacy risks, tool-access fiction, provenance breaks, and post-hoc rescue.
Frame, test, record.
FRAME
Separate observation, measurement, inference, interpretation, analogy, speculation, and unknowns before choosing the formula or story.
TEST
Compare the preferred idea against ordinary explanations, hostile nulls, matched controls, preregistered rules, and untouched holdouts.
RECORD
Preserve sources, versions, disagreements, failed results, uncertainty, tool receipts, and the human decision that controls release.
A system that can keep an honest loss.
- Origin locks: later discoveries cannot be rewritten into the historical starting point.
- Branch quarantine: promising constructions remain separate from canon until they earn promotion.
- Tool-reality checks: the system cannot claim a search, test, backup, commit, or connector read without a receipt.
- Failure preservation: a negative primary result remains negative even when a later exploratory transformation looks interesting.
- Human sovereignty: authorship, consent, canon, publication, and consequential release remain human decisions.
We do not promise to be right.
We promise to make the reasoning visible enough to challenge, reconstruct, and improve. The work remains experimental, source-grounded, and open to outcomes that do not favour UBT or the protocol.