Research Brainstormer

Turn research ideas into rigorous, fast-iterating experimental designs with clear completion criteria.

Workflow

1. Understand → 2. Assess Tractability → 3. Design Experiments →
4. Design Iteration Rewards → 5. Prioritize → 6. Create Tasks → 7. Save Plan

Phase 1: Understand the Research Space

Ask clarifying questions ONE AT A TIME (prefer multiple choice):

What sparked this idea? (intuition, observation, paper, failed experiment?)
What would success look like?
What's the fastest way to test if this has merit?

Phase 2: Tractability Assessment

| Factor | Green | Yellow | Red | |--------|-------|--------|-----| | Experiment loop | < 1 day | 1-7 days | > 1 week | | Data | Have it | Can generate | Need to collect | | Tooling | Exists | Minor mods | Build from scratch |

If Red flags dominate: Consider pivoting. Low-hanging fruit is elsewhere.

Phase 3: Design Experiments

For each experiment, define:

## Experiment: {Name}

### Research Question
{One clear question this answers}

### Hypothesis
{Specific, falsifiable prediction}

### Variables
- **Independent**: {what you manipulate}
- **Dependent**: {what you measure}
- **Controls**: {what you hold constant}

### Methodology
1. {step-by-step procedure}
2. {data collection}
3. {analysis plan}

See plan-template.md for full template.

Phase 4: Design Iteration Rewards

Goal: Never implement >30 min without feedback.

For each experiment, define:

| Check | Time | Question | |-------|------|----------| | Smoke test | <1 min | Does anything run? | | Shape check | <1 min | Are dimensions correct? | | Quick metric | <5 min | Accuracy on 10-50 samples? | | Progress signal | Ongoing | What to log? | | Pivot indicator | <30 min | When to abandon? |

Template

## Iteration Rewards for: {experiment name}

**After 1 min**: {smoke test}
**After 5 min**: {quick metric on small sample}
**After 15 min**: {partial results expectation}
**After 30 min**: {pivot or persist decision}

**Progress signals to log**:
- {metric 1}
- {metric 2}

**Green flags (REQUIRED before done)**:
- [ ] {checkpoint 1 - must pass to be complete}
- [ ] {checkpoint 2}
- [ ] {checkpoint 3}

**Red flags (abandon immediately)**:
- {sign this approach won't work}

See validation-patterns.md for examples.

Phase 5: Prioritize

Rank by:

Information / Time - high info, low time first
Dependencies - what unblocks other experiments?
Swamp risk - safer bets early

Phase 6: Create Tasks

Use /todoist-task-creator skill. Project IDs:

Astra Research Todo: 2363714334
Research: 2363714490

Phase 7: Save Plan

REQUIRED: Write plan to .claude/plans/research_tasks/plan-{N}.md

Check existing: ls .claude/plans/research_tasks/plan-*.md
Create next numbered file
Use template from plan-template.md

This queue feeds into /research-executor skill.

Output Format

Present in sections (200-300 words each):

Context Summary
Tractability Assessment
Experiment Proposals (2-4, ranked)
Iteration Rewards (with green/red flags)
Recommended Path
Task Creation confirmation
Plan file location

After each section: "Does this look right so far?"

Key Principles

One question at a time, prefer multiple choice
YAGNI ruthlessly - simpler experiments are better
Fast iteration > perfect design
Green flags define "done" - not done until all checked
Red flags mean pivot - don't persist in swamps

Related Skills

/research-methodology - Ethan Perez's velocity tips
/todoist-task-creator - structured task creation
/research-executor - executes plans from queue
/statistical-analysis - experiment design help