Proposal 206: runtime co-reviewer model resolution

[alonf]

What changed

• adds Proposal 206 for runtime co-reviewer model resolution
• separates stable harness adapter data from volatile account-visible model selection
• defines CLI-first discovery, advisory official-web enrichment, explicit model threading, and evidence-backed resolved identity
• replaces host-only independence with harness/provider/model-aware classifications
• updates the proposal index

Why

The current reviewer catalog records descriptive model labels without consistently controlling or proving the model that actually runs. Model availability varies by CLI version, account, plan, provider, and region, while aggregator harnesses can expose the same underlying model through different executables. This can make catalog data stale and overstate reviewer independence.

Proposal 206 composes with Proposals 102, 197, 200, 068, 040, 187, and 203. It replaces only Proposal 102 Pillar 5's static-only model-resolution mechanism while preserving its authorization and fallback-policy goals.

User impact

The proposal targets a future flow where users authorize reviewer hosts through the supported CLI, choose explicit models or capability policies, and receive evidence naming the requested and resolved model. Users do not manually edit reviewer configuration, and web research cannot authorize or invoke a model.

Validation

• npx --yes markdownlint-cli proposals/206-runtime-co-reviewer-model-resolution.md proposals/INDEX.md
• git diff --cached --check
• governance validator exited successfully; it reported only existing missing-dashboard warnings unrelated to this proposal