- Models repeat the same secondary sources
- Contrary evidence disappears during synthesis
- Confidence is reported without coverage limits
research workflow hub
Multi-model workflows for research
Use separate evidence, verification, and challenge stages to produce one transparent conclusion instead of model consensus theater.

- Preserve direct URLs and event dates
- Label model synthesis as inference
- Never use agreement between models as independent evidence
- Record coverage and unreviewed sources
One flow, bounded owners.
| Stage | Purpose | Suggested ownership |
|---|---|---|
| Protocol | Define the question, source hierarchy, dates, and exclusions | Planner |
| Evidence | Collect primary sources and direct findings | Researcher + note-taker |
| Verification | Check authority, dates, and claim support | Verifier + fact-checker |
| Counter-case | Find limitations and alternate explanations | Critic + devil’s advocate |
| Decision brief | Separate evidence, inference, recommendation, and unknown | Reviewer + formatter |
Authored recipes
Start from a tested structure.
Source-grounded research brief
Split evidence discovery and counter-analysis into two supervised tracks, then combine them into one cited brief.
Roles + task guides
Understand each responsibility.
Researcher
Finds and organizes evidence under a defined source, date, and relevance standard.
Verifier
Checks whether claims, calculations, links, states, and completion evidence are true.
Researcher aide
Organizes source discovery and evidence extraction under the primary role’s research rules.
Devil’s advocate
Constructs the strongest credible challenge to the primary role’s assumptions or conclusion.
Choose Solo, Collaboration, or Dual
Match the orchestration topology to the dependency structure of the work.
Configure Majesty Obmil supervisor rules
Give the supervisor explicit review, stop, takeover, and team rules.