BrennerBot

What We're Building

The Brenner loop needs structured inputs to work with. Your agent will generate these using the operators it learned earlier:

Hypothesis Slate

Multiple competing explanations, not just the obvious one

Uses Object Transpose (⟳)

Assumption Ledger

Every hidden premise that could be wrong

Uses Level Split (Σ)

Third Alternatives

Options beyond the obvious binary choices

Uses Object Transpose (⟳)

⊙

Scale Checks

Reality checks on effect sizes and plausibility

Uses Scale Check (⊙)

The Input Generation Prompt

Replace [YOUR REFINED QUESTION FROM STEP 5] with your actual refined question and give this to your agent:

Prompt to your agent

Now I need you to build the formal inputs for a Brenner research loop. My refined research question is:
 
[YOUR REFINED QUESTION FROM STEP 5]
 
Please generate the following artifacts:
 
## 1. Hypothesis Slate (minimum 4 hypotheses)
 
For each hypothesis:
- State it clearly and specifically
- Explain the mechanism it proposes
- Note what evidence would support it
- Note what evidence would falsify it
 
Include at least:
- The "obvious" hypothesis (what most people would assume)
- A mechanistic alternative (different underlying mechanism)
- A reversed causation hypothesis (effect causes apparent cause)
- A third-variable hypothesis (both are caused by something else)
 
Use **Object Transpose (⟳)** to generate the reversal and third-variable options.
 
## 2. Assumption Ledger
 
List every assumption underlying each hypothesis. Categorize them as:
- **Theoretical**: Assumptions about how things work
- **Methodological**: Assumptions about how we'd measure/observe
- **Background**: Taken-for-granted facts that might be wrong
 
Use **Level Split (Σ)** to check if assumptions hold at different levels.
 
## 3. Third Alternatives
 
For any pair of hypotheses that seem like the only options, generate a third possibility that:
- Is neither hypothesis A nor hypothesis B
- Could explain the same observations
- Comes from applying Object Transpose or Level Split
 
## 4. Critical Evaluation
 
For each hypothesis, apply **Scale Check (⊙)** to verify:
- Is the proposed effect size plausible?
- Do the numbers make physical/biological sense?
- What would have to be true for this mechanism to work?
 
Take your time. Be thorough. This is the foundation for everything that follows.

What Good Outputs Look Like

Good Hypothesis Slate

H1 (Obvious): X directly causes Y through mechanism M.

H2 (Alternative mechanism): X causes Y, but through mechanism N, not M.

H3 (Reversed causation): Y causes X, and we've misidentified the direction.

H4 (Third variable): Z causes both X and Y; they're correlated but neither causes the other.

H5 (Level shift): The relationship holds at molecular level but not at systems level (or vice versa).

Good Assumption Ledger

Theoretical: "We assume X and Y are distinct entities, not different manifestations of the same process."

Methodological: "We assume our measurement of X doesn't itself affect Y."

Background: "We assume the standard model of [domain] is correct in this context."

Important

The quality of your hypothesis slate determines everything. If you only consider one explanation, no amount of testing will save you from confirmation bias. Demand at least 4 genuinely different hypotheses.

Reviewing the Output

After your agent generates the inputs, check each one:

Hypothesis slate

Are the hypotheses genuinely different, or variations on a theme?

If weak: Ask for alternatives that would require different tests to distinguish.

Assumption ledger

Are there assumptions so obvious you almost missed them?

If weak: Ask the agent to list 'background' assumptions about your field.

Third alternatives

Do any feel like they were generated just to fill a slot?

If weak: Ask the agent to explain what observation would make each plausible.

Scale checks

Has the agent done actual calculations, or just hand-waved?

If weak: Ask for order-of-magnitude estimates with explicit numbers.

Pro Tip

Save the agent's output! Copy it to a file or note. You'll reference these artifacts throughout the rest of the loop, and you'll want them for documentation when you're done.

Success Criteria

You're ready to proceed when you have:

At least 4 genuinely distinct hypotheses
Assumptions identified for each hypothesis
At least one "third alternative" you hadn't considered
Scale checks that include actual numbers

Next up: In Step 7, you'll have your agent run the full Brenner loop — designing discriminative tests and ranking them by potency.