Overview
/silver:deep-research-multi-ai is the explicit multi-model variant of canonical deep research. The host drives the research phases once; eligible retrieval, triangulation, and critique work is delegated through AF-MULTI-AI-TASK.
Use canonical /silver:deep-research by default. Select this workflow when parallel independent evidence gathering justifies the added cost and reconciliation work.
Catalog Composition
The composition preserves the normal research contracts. Multi-AI dispatch does not bypass need-profile capture, source quality rules, deterministic consolidation, or final validation.
Research Modes
- General research: parallel retrieval, triangulation, and critique contributions.
- Solution landscape: evidence-backed candidate inventory and weighted comparison artifacts.
- Solution compare: two or more named solutions evaluated against the accepted need profile.
Artifacts
The canonical run_manifest.json uses manifest v4 and spine: multi-ai-task-v2. Consolidation can emit report.html, landscape-report.html, and comparison JSON, Markdown, and spreadsheet artifacts. Older v3 manifests migrate copy-on-write with an audit chain.