AMI — Autodiagnosis & Iterative Improvement
Atomic Claim
Overview
AMI is a structured feedback loop where you present your own work to an AI, request a critical diagnosis using defined criteria, then iterate — using the AI as an external critic while retaining full authorship.
💡 The Three-Step Loop
YOUR DRAFT → AI DIAGNOSIS → YOUR REVISION → (repeat)
Step 1 — Submit with explicit criteria
"Here is my [hypothesis / section / prompt / argument]:
[YOUR CONTENT]
Please diagnose it against these criteria:
1. Internal consistency — Are there contradictions?
2. Evidence sufficiency — Are claims supported?
3. Falsifiability — Can each claim be tested?
4. Clarity — Is the argument followable?
5. [Domain-specific criterion]
For each criterion: rating (1-5), specific problem identified,
suggested fix. Do NOT rewrite for me."
Step 2 — You revise (never paste AI text directly)
Step 3 — Second pass
"Here is the revised version: [REVISION]
What problems remain from your previous diagnosis?"
🧩 Application to PUMA
Used to refine: research hypotheses H1/H2, prompt templates for experiments, Chapter 1 argumentation. See: 60 - Resources/61 Prompts/61.1 LLM-Tools/PT-Claude-AMI-Review
🔗 Connected Ideas
Part of: PN-MIT-Student-Method | Pairs with: PN-IIPR-Framework
id: PN-DRCA-Framework title: “DRCA — Deconstrucción y Reconstrucción Conceptual Avanzada”
DRCA — Advanced Conceptual Deconstruction & Reconstruction
Atomic Claim
DRCA is a four-step cognitive procedure for deeply processing any complex concept or paper: Deconstruct its components, Reconstruct in your own framework, perform Critical analysis, then Advance by generating original output.
💡 The Four Steps
| Step | Action | Output |
|---|---|---|
| Deconstruct | Break into atomic parts: problem, method, evidence, claim, limits | Structured outline |
| Reconstruct | Rebuild using YOUR vocabulary and mental model | Restatement in own words |
| Critical | Identify: assumptions, gaps, rival explanations, validity threats | Critical notes |
| Advance | What new permanent note, hypothesis, or design decision does this generate? | Zettelkasten atom |
🧩 Application to PUMA
Every paper in the SLR goes through DRCA. The “Advance” step produces the permanent notes in 30 - Permanent/. See: 60 - Resources/61 Prompts/61.1 LLM-Tools/PT-Claude-DRCA-Paper
🔗 Connected Ideas
Part of: PN-MIT-Student-Method | Produces: Template-Permanent-Note
id: PN-IIPR-Framework title: “IIPR — Ingeniería Inversa de Prompts y Respuestas”
IIPR — Inverse Prompt Engineering
Atomic Claim
IIPR is the practice of reverse-engineering why a prompt produced a poor result, then systematically redesigning the prompt structure to achieve the desired output — treating prompts as testable specifications.
💡 The IIPR Protocol
When a prompt fails or produces suboptimal output:
Step 1 — Diagnose
"I asked you: [ORIGINAL PROMPT]
You responded: [AI RESPONSE]
I wanted: [DESIRED RESPONSE]
What is wrong with my prompt that caused this mismatch?
Be specific about which component (Role/Context/Objective/Instructions/Format)
needs improvement and why."
Step 2 — Reconstruct Based on AI diagnosis, rewrite the prompt addressing each identified weakness.
Step 3 — Version control Log both versions in the prompt note with the change reason. See: Template-Prompt-Note (Refinement History section).
🧩 Application to PUMA
Critical for iterating on the Ollama experiment prompts (triage and estimation strategies). Each refinement iteration is logged in the experiment notes.
🔗 Connected Ideas
Part of: PN-MIT-Student-Method | Uses: PN-RCOIF-Framework | SDD analogy: Iterating on specs