Meta Muse assessment adds proposed evaluation and teacher candidates
The research proposes Wonder and Spark as evaluation baselines and self-hosted Glimmer as a conditional model or teacher candidate. It grants none approval under the project's strict provenance and model lineage requirements.
Proposed model roles
The research retains the proposed KL3M pilot and newly initialized OLMo-core foundation. It adds Glimmer alongside self-hosted gpt-oss as a conditional teacher candidate.
Proposed permission checks
The proposed hosted training route requires written authorization for competing-model development. The proposed Glimmer route requires separate license, source permission and lineage reviews, including its disclosed Spark ancestry.
Proposed contributor safeguards
The research proposes routing contributor material according to accepted creator permissions. It rejects treating Contributor discounts as creator compensation without an explicit agreement. The proposed access review includes subscription settings that govern training use.
Proposed evaluation requirements
The proposed evaluation measures source support, quotation accuracy, unsupported credit, abstention, latency and cost per accepted answer. Proposed accounting checks require durable receipts and tests for beneficiary splits, duplicate events, rounding and corrections.
Research index
The README now lists the Meta Muse assessment covering permissions, local deployment, lineage, Contributor pricing and proposed creator rewards.
Files in this change
| File | Change | Lines |
|---|---|---|
README.md | modified | +1 −0 |
docs/META_MUSE_SOVEREIGN_SOURCE_RESEARCH.md | added | +98 −0 |