Research Architect Brainstorm

Research Architect Brainstorm

Use this branch after intake. When target references exist, run the target-reference pass of research-architect-literature first so brainstorming can adapt known research logic rather than guess from the topic alone.

Inputs

Read:

paper_output/project_config.json
paper_output/source_map.md
paper_output/source_inventory.md
paper_output/exemplar_logic_profile.md        # when target references exist
paper_output/exemplar_adaptation_plan.md      # preliminary version when available

Brainstorming lenses

Generate candidates using these lenses:

1. Reference-logic lens: Which question-forming, theory, design, evidence, analysis, and argument principles make the target paper work?
2. Adaptation lens: What is the user's project analogue, and which context-dependent assumptions require replacement?
3. Problem lens: What unresolved tension, contradiction, practical concern, or conceptual limitation is visible?
4. Material lens: What can the user measure, observe, interpret, compare, trace, derive, review, or test?
5. Method and inquiry lens: Which design family and logic of inquiry can generate a defensible answer?
6. Warrant lens: What empirical, interpretive, comparative, theoretical, or argumentative warrant can support the claim?
7. Disconfirmation lens: Which negative case, rival explanation, counterexample, sensitivity test, or boundary condition would clarify scope?
8. Reader-value lens: What would a skeptical reader learn that existing work does not already establish?

Required outputs

Create:

paper_output/problem_landscape.md
paper_output/idea_candidate_matrix.md
paper_output/research_question_options.md
paper_output/feasibility_filter.md

Candidate matrix

Use:

| Candidate ID | Problem or Tension | Candidate Research Question | Reference Logic Inherited | Required Adaptation | User-Specific Contribution | Needed Evidence or Warrant | Available Material | Missing Material | Feasibility | Mismatch Risk | Claim Risk | Next Action | |---|---|---|---|---|---|---|---|---|---|---|---|---|

Option requirements

Each target-driven option must state:

• which strength of the reference paper it preserves;
• which source-specific assumption it replaces;
• how the user's context creates an independent contribution;
• what evidence or argumentative warrant is required;
• which design family is provisionally selected;
• what action the user should take first;
• what claim remains supportable if the full target design is infeasible.

Options should differ in research logic, evidence path, or contribution. Topic paraphrases do not count as distinct options.

Filtering

Classify candidates as:

• advance: coherent, answerable, and feasible enough for literature positioning;
• needs_more_material: promising with a clearly identified missing input;
• defer: useful after a simpler or narrower study is completed;
• reject: already answered, unsupported, structurally mismatched, or lacking a defensible inquiry path.

Conceptual, critical, interpretive, and normative projects can advance when they have explicit premises, source-selection logic, argumentative warrants, counterarguments, and scope.

Output behavior

Present 2-5 candidate directions and explain the reference-derived logic behind each. The final research spine remains a user decision after literature positioning.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: mmTheBest
• Source: mmTheBest/Research-Architect
• License: MIT

Install and usage instructions live in the source repository linked above.