OrchestrUI v0.3.0 — Explainable candidate ranking

OrchestrUI 0.3.0 — Explainable candidate ranking

OrchestrUI now explains not only which UI ecosystem wins, but how every admissible candidate was ranked and why a hard compatibility gate rejected an option.

Highlights

• Multi-candidate routes consider policy order, installed-package evidence, current-plan reuse, dependency cost, bundle cost and role overlap.
• Supplied host versions are checked against bounded compatibility constraints before a candidate can win.
• recommend_stack returns candidate_rankings with factor-level scores and evidence.
• A candidate already present in the host can outrank an otherwise preferred addition when it remains compatible.
• Eight independently specified adversarial goldens run separately from the 50-case internal policy benchmark.
• All three executable fixtures now include installed dependency versions and captured ranking evidence.
• The npm archive excludes heavyweight promotional renders that are unnecessary at runtime.
• The orchestrui-mcp executable is normalized for current npm clients and checked against its built entrypoint.

Compatibility

• The MCP server remains read-only and keeps the existing six-tool surface.
• Existing text and partial-profile inputs remain supported; full HostProfile and TaskProfile input provides the strongest evidence.
• New response fields are additive. Existing selected/rejected and role-ownership fields remain available.
• Node.js 20 or newer is required.

Safety and licensing

• Ranking scores never override framework, version, ownership, conflict or asset-rights gates.
• No upstream component collection is mirrored or redistributed.
• Paid, Pro, authenticated and credentialed upstream content remains excluded.
• Rive runtime licensing remains separate from rights to individual .riv assets.

Documentation: README · Routing · MCP specification

Contact

For questions about OrchestrUI, integration, consulting, or collaboration: