Role before provider
A governed system needs durable roles: lead, challenger, executor, verifier, specialist, and human authority. Providers may fill those roles, but provider identity should not define governance.
Brain agnostic does not mean every model is interchangeable. It means an assignment can change without rewriting authority, memory, evidence, acceptance, or recovery.
The dated reference configuration
As of August 7, 2026, the approved public reference describes Anthropic models for executive cognition and continuity, OpenAI Codex Exec for implementation and execution, Goose plus ECC and local models or Ollama for a candidate on-prem worker path, Sam for governed operation, Sky for technical leadership, and Steve for consequential authority.
That is a dated narrative, not a permanent leaderboard or partnership claim.
How roles are earned
A provider can gain or lose a role based on accepted quality, cost, latency, reliability, tool use, context handling, privacy, safety, and observed regressions. Reassignment should be recorded, not silently changed.
For consequential work, fresh challenge can expose assumptions. Deterministic tests and qualified human review remain separate requirements.
Permanent layer
Governance is permanent even when the brain changes. No provider grants itself authority, accepts its own unverified result, publishes private data, or silently chooses a less private or more expensive route.
The durable architecture is the contract around the model, not loyalty to the model name.
Evidence and limits
Sources
- Sam Foundry Product, Platform, and Operating Specification v0.1 (primary-documentation) — docs/specs/SamFoundry-Product-and-Platform-Spec-v0.1.md
- Sam Foundry Model Council Doctrine v0.1 (primary-documentation) — docs/architecture/SamFoundry-Model-Council-Doctrine-v0.1.md
