Metaprompt
METAPROMPT (v0.2): Extract + Operationalize “The Brenner Approach” from Transcript Evidence
You are analyzing Sydney Brenner’s scientific method from primary-source transcripts. Your job is NOT to summarize. Your job is to extract repeatable cognitive moves and turn them into actionable research protocols that a multi-agent team can run.
Inputs (I will paste these)
- TRANSCRIPT EXCERPT(S): selected sections of Sydney Brenner transcripts (with headings / section numbers if present).
- (Optional) QUOTE BANK: a pre-extracted set of verbatim quotes (still treat as evidence; do not quote outside what I paste).
- (Optional) FOCUS THEME: e.g., “problem selection”, “decision experiments”, “working out of phase”, “HAL biology”, “inversion”, “digital readouts”, “Don’t Worry hypotheses”, “phase problems / ambiguity breaking”, “mutational spectra”, “conversation as thinking technology”, “open the box / grammar of the system”, “tooling economics”, “quickies / pilot experiments (de-risking)”, “dominant-variable rescues (physics saves)”, “democratizing tools”.
- (Optional) TARGET RESEARCH DOMAIN: e.g., biology, ML, materials, climate, robotics.
- (Optional) CURRENT RESEARCH QUESTION: a real question we want to make progress on using the Brenner approach.
Hard constraints (do not violate)
- No quote fabrication. Only quote what appears in the excerpt I provide.
- Evidence-first. Every major claim must cite ≥1 quote (or explicitly mark as inference).
- Avoid generic advice. Everything must be tied to Brenner-specific moves and transcript evidence.
- Prefer discriminative leverage. Optimize for actions that collapse hypothesis space quickly.
---
Your mission
Produce a “Brenner Approach” playbook that is: 1) grounded in transcript evidence, 2) internally coherent (moves reinforce each other), 3) operational (can be executed as a workflow), 4) robust against the “third alternative” (both models wrong).
---
Output format (follow exactly)
0) Executive summary (10 lines max)
- 3–5 bullet “headline” principles (high novelty, high leverage).
- 1 paragraph: what makes this distinctly Brenner, not generic “good science”.
1) Quote bank (high-signal, minimum 12 quotes)
Create a table:
| ID | Quote (verbatim) | Section/Heading | Why it matters (1 sentence) | Tags | |---|---|---|---|---|
Tags are drawn from: problem-choice, representation-change, inversion, HAL, digital-handle, exclusion, paradox, assumption-ledger, ruthless-kill, out-of-phase, organism-selection, cheap-loop, topology/algebra, bayesian-ish, dont-worry, conversation, open-the-box, grammar, phase-problem, phase-breaking, spectra, tacit-knowledge, tooling, democratize, quickie, dominant-variable, initiation-vs-continuation, genetic-dissection, self-assembly, special-exemplar, construction-vs-function, anti-analogy, plausibility-filter, lineage-vs-gradients, reconstruction, heroic-vs-classical, normal-science, genetic-surgery, inside-out-genetics, discount-genome, integrative.
2) Brenner Moves (8–12 items, each evidence-backed)
For each move, produce this exact structure:
Move N — <NAME> (one line definition)
- What it does: (2–3 sentences)
- What it optimizes for: (e.g., time-to-discrimination, likelihood ratio, loop time)
- When it fails / anti-pattern: (1–2 failure modes)
- Transcript anchors: (list 2–4 Quote IDs)
- Modern translation: (how this looks in 2025 research practice)
- Micro-drills: (2 exercises to train this move)
Moves must be specific (e.g., “Hunt paradoxes as beacons”, “Turn the problem into a decision procedure”, “Pick the experimental object”, “Prefer digital handles”, “Inversion as a generator”, “Occam’s broom as a warning sign”, “Work out of phase”, “Ruthless theory-killing”, etc.).
Also include Brenner-style moves like: “Don’t Worry hypothesis (defer missing mechanisms)”, “Conversation to break deductive loops”, “Open the box / find the grammar of the system”, “Compute the organism (reconstruction as explanation)”, “Gradients vs lineage (analogue vs digital development)”, “Break ambiguity by solving the missing variable (phase)”, “Type mechanisms via spectra / equivalence classes”, “Find the control point (initiation vs continuation)”, “Genetic surgery (mutation-first proof of function)”, “Inside-out genetics (tooling changes feasible experiments)”, “Discount genome / organism choice as technology”, “Heroic → classical periods (routine work generates new problems; what can/can’t be solved by normal science)”, “Tooling economics (build/democratize the kit)”, “Reject ‘logical but non-natural’ theories and easy analogies.”
Prefer to include moves like: “Run a quickie (pilot experiment) to de-risk the flagship”, “Find the dominant physical variable and push it hard”, and “Turn elite techniques into routine kit (democratize tooling).”
3) The “Brenner Loop” (a runnable protocol)
Define an explicit loop a research group can run in 30–90 minutes:
- Step A — Problem selection
- Step B — Hypothesis slate (explicit enumeration)
- Step C — Third-alternative guard
- Step D — Discriminative tests / decision experiments
- Step E — Assumption ledger
- Step F — Next actions + stopping rule
For each step, include:
- Inputs
- Outputs (artifacts)
- Quality bar (what “good” looks like)
- Common failure mode
4) Prompt templates (ready to copy/paste)
Create 6 prompt templates (each ≤ 250 lines) that implement the loop steps above. Each template must include:
- Role
- Goal
- Inputs
- Output schema
- Self-check rubric
Templates to include:
- 1Problem selection + “work out of phase” scan
- 2Hypothesis slate + explicit priors (lightweight)
- 3“Third alternative” adversarial critique
- 4Decision experiments ranked by expected discrimination
- 5Assumption ledger + break-tests
- 6Synthesis memo (what we now believe, why, and what would change it)
5) Demonstration on a concrete question (if provided)
If I provide a CURRENT RESEARCH QUESTION:
- Run the Brenner Loop once.
- Produce the artifacts.
- Explicitly show where a “Brenner move” was used.
If I do NOT provide a question:
- Propose 3 candidate questions (in the chosen domain or general science) that are well-suited to the Brenner approach and explain why.
6) Calibration & falsification
- What would convince you this playbook is wrong or incomplete?
- Which “Brenner moves” are most at risk of being mythology vs grounded method?
- What evidence would you want next from the transcripts? (specific kinds of sections/episodes to locate)
---
Style requirements
- Be crisp, funny when appropriate, and precise.
- Prefer “this implies…” over “it seems…”.
- When you infer beyond quotes, label it explicitly: (Inference).
- Keep the tone like a top lab’s internal methods memo, not self-help.