BrennerBot

What Makes a Good Research Question?

Brenner was ruthless about question quality. Most questions scientists ask are actually disguised statements or lead nowhere testable. Here are his criteria:

Observable

There must be something you can actually measure or observe. If you can't specify what you'd look for, it's not a question yet.

Weak:Why does consciousness exist?

Strong:Does visual attention require intact prefrontal cortex?

Excludable

A good question can be answered 'no'. If every possible observation confirms your hypothesis, you're not doing science.

Weak:Does stress affect health?

Strong:Does cortisol elevation precede depressive episodes?

Consequential

The answer should change what you do or believe. If nothing changes either way, why ask?

Weak:Are there many genes involved in development?

Strong:Can we identify the minimal gene set for segment formation?

Formulating Your Question

Before giving your question to the agent, spend a moment thinking about it:

1.What phenomenon are you trying to explain?
2.What's the current consensus (if any)?
3.Why do you suspect the consensus might be wrong or incomplete?
4.What would you do differently if you had the answer?

Give Your Question to the Agent

Copy this prompt, replace [YOUR QUESTION HERE] with your actual question, and give it to your agent:

Prompt to your agent

I have a research question I'd like to explore using the Brenner methodology you just learned. Here's my initial question:
 
[YOUR QUESTION HERE]
 
Please help me refine this question by:
 
1. **Evaluating against Brenner's criteria:**
   - Is there a clear observable I could measure?
   - Is there potential for exclusion (could this be proven false)?
   - What would change if I knew the answer?
 
2. **Applying Level Split:**
   - What level of analysis is this question at? (molecular, cellular, systems, behavioral?)
   - Should the question be asked at a different level?
 
3. **Checking for implicit assumptions:**
   - What am I assuming to be true that might not be?
   - Are there hidden premises in my framing?
 
4. **Suggesting refinements:**
   - If the question is too broad, suggest narrower versions
   - If too narrow, suggest what broader question it addresses
   - Reframe to make it more testable
 
Take your time. Be critical. I want a question that will actually lead somewhere.

What to Expect

Your agent should respond with a thorough critique. Don't be discouraged if it finds problems — that's exactly what you want at this stage.

Level Analysis

Agent identifies what level your question operates at

Assumption Surfacing

Agent names things you're taking for granted

Refined Versions

Agent offers sharpened alternatives to your question

Testability Check

Agent evaluates whether the question can be answered

Warning

Don't skip the critique step! It's tempting to rush ahead with your original question, but taking 5 minutes now to sharpen it will save you from building elaborate tests for the wrong question.

Pro Tip

If your agent's critique reveals that your question is actually several questions bundled together, that's a valuable discovery. Pick the most important sub-question to proceed with.

Iterate Until Sharp

You may need 2-3 rounds of refinement. After the agent critiques your question, try a revised version:

Follow-up prompt

Based on your critique, here's my refined question:
 
[REVISED QUESTION]
 
Does this address the issues you raised? What's still weak about it?

Success Criteria

You're ready to proceed when:

You have a single, focused question (not a bundle)
You can specify what observation would answer it
The question could be answered "no" (it's falsifiable)
You know what you'd do differently with the answer

Next up: In Step 6, your agent will generate the formal inputs for the Brenner loop: hypothesis slate, assumption ledger, and more.