Step 1 of 8

Agent Runs the Brenner Loop

7of 8~10 min
  • How to force discriminative (not confirmatory) tests
  • How potency checks prevent weak experiments
  • What a complete Brenner research artifact looks like
What you'll do
  • Give the full-loop prompt to your agent
  • Inspect the tests for discriminative power
  • Ask for revisions if anything is missing

This is where the method becomes operational: your agent turns the hypothesis slate into discriminative tests, ranks them by potency, and produces a complete research artifact.

The Full-Loop Prompt

Copy this prompt, fill in the bracketed sections from Steps 5–6, and give it to your agent:

Prompt to your agent
I want you to run a full Brenner Loop on my research question using the artifacts you already generated.
## Inputs
### Research question
[PASTE YOUR REFINED QUESTION FROM STEP 5]
### Hypothesis slate
[PASTE THE HYPOTHESIS SLATE FROM STEP 6]
### Assumption ledger
[PASTE THE ASSUMPTION LEDGER FROM STEP 6]
### Third alternatives
[PASTE THE THIRD ALTERNATIVES FROM STEP 6]
### Scale checks / plausibility notes (if any)
[PASTE ANY SCALE CHECK NOTES FROM STEP 6]
---
## Requirements (Brenner disciplines)
1) **Discriminative tests only (⊘ Exclusion Test):**
- Propose tests where at least two hypotheses predict different outcomes.
- Each test must specify: what observation would EXCLUDE which hypothesis.
2) **Potency check each test (🎭):**
- If the test comes back null/negative, what do we learn?
- If the test is likely to be ambiguous, rewrite it until it isn't.
3) **Keep the hypothesis space honest (Σ + ⟳):**
- Ensure hypotheses are not level-mixed (Σ Level Split).
- Ensure at least one genuine third alternative is included (⟳ Object Transpose).
4) **Scale sanity (⊙):**
- Where relevant, include order-of-magnitude or numeric plausibility checks.
---
## Output format (strict)
Produce a single markdown artifact with these sections:
1. **Single triangulated kernel** (3–6 bullets): your best current hypothesis + the single most discriminative next test + why.
2. **Hypothesis slate (revised)**: 4+ hypotheses, each with mechanism + 1–2 key assumptions.
3. **Discriminative tests (ranked)**: a table with:
- Test name
- What you do / measure
- Predictions if H1/H2/H3/... true
- Exclusion logic (what outcome rules out what)
- Potency check (🎭)
- Feasibility (time/cost/skill)
4. **Assumption ledger (updated)**: grouped by hypothesis and by type (theoretical / methodological / background).
5. **Adversarial critique**: attack the framing; propose 1–2 alternative framings.
6. **Recommended next steps**: the next 3 actions, ordered by discriminative power.
Take your time and be ruthless about discriminative power.

What to Look For

  • Tests that make different predictions under different hypotheses.
  • A potency check for every test (what you learn if the result is null).
  • A revised hypothesis slate that stays mechanistic and includes a real third alternative.
Pro Tip
If the agent produces a huge artifact, ask it to additionally output a one-screen "single triangulated kernel" summary: best current hypothesis + best next test + why.

Next up: In Step 8, you'll do a human review using a checklist that catches the most common failure modes.