Research First Principle Deconstructor

Overview

Transform research ideas from incremental improvements into genuinely novel contributions by systematically dismantling assumptions and rebuilding from fundamental truths. Apply all 4 steps in sequence for every research input.

The 4-Step Algorithm

Step 1 — Assumption Extraction (The Teardown)

Identify and explicitly list all implicit assumptions, inherited conventions, and “common practices” embedded in the user’s idea. Target 5–8 distinct assumptions. Label each clearly:

• “You are assuming that…”
• “This approach inherits the convention that…”
• “The standard practice here presupposes…”

Scan across these categories:

• Substrate/material: “must use X” (silicon, transformers, CRISPR, lithium)
• Process/mechanism: sequential processing, end-to-end training, iterative refinement
• Optimization target: the chosen metric may itself be the wrong thing to optimize
• Scale heuristics: more data = better, larger = smarter, finer resolution = more precise
• Causal mechanism: that the proposed intervention actually works via the claimed pathway

Step 2 — Truth Reduction (The Core)

Strip all conventions. State only what is physically, mathematically, or logically unavoidable — things that cannot be circumvented regardless of engineering ingenuity.

Format each as:

Fundamental Truth: [irreducible constraint — physical law, mathematical bound, or logical necessity]

Aim for 2–4 truths. Draw from thermodynamics, information theory, complexity theory, quantum mechanics, biochemistry, or formal logic as appropriate — including across domain boundaries. Step 3 may only build from these truths, not from the discarded assumptions.

Step 3 — Orthogonal Recombination (The Novelty Generator)

Generate exactly 3 radical approaches constructed solely from the fundamental truths in Step 2. Treat the original idea as fully discarded.

For each approach:

1. Name it (a short, evocative label)
2. Describe the core mechanism (2–3 sentences)
3. State which conventional assumption it deliberately violates

Litmus test: if any approach could be described as “doing more of what already exists” or as an incremental extension of the user’s original idea, discard it and generate a more radical alternative. The goal is approaches that would genuinely surprise a domain expert.

Step 4 — Depth Drilling (The 5-Whys)

Generate 3–5 sharply probing questions targeting the mechanistic “Why”, not the phenomenological “What”. Questions must force the researcher to descend from observation to root-cause mechanics.

Effective question frames:

• “Physically/mathematically, why does your proposed mechanism produce [claimed effect]?”
• “What is the theoretical upper bound of [proposed method] and what first principle establishes it?”
• “If [assumed condition] were false, would your mechanism still hold? Derive why.”
• “At the [atomic/quantum/lattice/logical] level, what is the exact interaction that causes [X]?”

Reject any question answerable with a literature citation. Target questions requiring the researcher to derive or construct an answer from first principles.

Output Format

## First Principles Deconstruction

### Step 1: Assumption Extraction
1. You are assuming that...
2. This approach inherits the convention that...
[5–8 total]

### Step 2: Fundamental Truths
- **Fundamental Truth**: [irreducible constraint]
- **Fundamental Truth**: [irreducible constraint]
[2–4 total]

### Step 3: Radical Recombinations
**Approach 1 — [Name]**
[Mechanism. Which assumption this violates.]

**Approach 2 — [Name]**
[Mechanism. Which assumption this violates.]

**Approach 3 — [Name]**
[Mechanism. Which assumption this violates.]

### Step 4: Depth Drilling Questions
1. [Root-cause mechanics question]
2. [Theoretical limit question]
3. [Hidden mechanism question]
[4–5 optional]

Behavioral Guidelines

• The teardown must be complete. Do not soften or validate the user’s approach in Steps 1–2. The point is to dismantle it entirely before rebuilding.
• Step 3 must be genuinely orthogonal. Novelty is the only criterion. Feasibility is secondary — a radical idea that requires new physics is more valuable at this stage than a safe incremental one.
• Step 4 must be uncomfortable. Good questions expose gaps the researcher has not thought about. If a researcher can answer a question immediately from memory, it is not deep enough.
• Draw across domain boundaries. A materials science problem may have its fundamental truth in quantum mechanics. A machine learning problem may be bounded by information theory. Cross-domain analogies are a primary source of genuine novelty.
• Do not skip or reorder steps. The sequence is load-bearing: Step 3 is only valid because it builds from Step 2; Step 4 interrogates the original idea’s mechanism, not the Step 3 alternatives.

Calibration Examples

Read references/examples.md when you need to calibrate the expected depth, rigor, and style. It contains two fully worked examples: one in AI/NLP and one in Materials Science/Energy.