🏗️ Claude — Theoretical Framework Construction for PUMA
Tool: Claude (Anthropic) Phase: Phase 1 — Research Step: 04 — Conceptual Synthesis Methodology: RCOIF + DRCA (Deconstruction and Reconstruction Conceptual Approach)
Prompt A — Conceptual Axis Identification (RCOIF)
Role: Senior theoretical researcher in AI systems and project management.
Context: I am building the theoretical framework for PUMA, a DSR-based TFG evaluating local LLM agents (Llama 3.2 8B, Mistral 7B) on project management tasks: issue triage (classification on Jira SR) and effort estimation (story points on TAWOS). The study follows Design Science Research and produces a reproducible benchmark with statistical validation (Wilcoxon test).
Objective: Identify and structure the theoretical axes that underpin PUMA's research design.
Instructions:
1. Identify 4–5 main theoretical axes. Each axis is a cluster of theories, concepts, and debates sharing a common fundamental question.
2. For each axis, provide: (a) axis name and central question, (b) key theories or models with founding authors, (c) precise concept definitions (2–3 lines each), (d) current active debate — what is disputed?, (e) direct relevance to PUMA's research question, (f) 2–3 essential references to cite [mark uncertain ones with VERIFY].
3. Identify relationships between axes: complementary, tensioning, or hierarchical.
Format: Structured sections per axis. Relationship diagram in text. Total length 800–1200 words.
Prompt B — Concept Key Matrix
Role: Academic lexicographer specializing in AI and software engineering.
Context: PUMA uses the following concepts that require precise operational definitions for the thesis: LLM Agent, Multi-Agent System, Issue Triage, Effort Estimation, Story Point, F1-macro, MAE, Few-Shot Prompting, Chain-of-Thought, RAG, Reproducibility, Human-in-the-Loop, Bounded Autonomy, Benchmark.
Objective: Build a definitional matrix for these 14 concepts.
Instructions: For each concept provide: (1) Operational definition for PUMA (50–70 words), (2) Canonical definition from most-cited source [with author and year — mark UNVERIFIED if uncertain], (3) Why this concept is indispensable for PUMA, (4) Relationship to other concepts in the matrix, (5) Empirical indicator or measurable proxy, (6) English search term for Semantic Scholar.
Format: Structured table with 6 columns. 1 row per concept. Final paragraph (100 words) describing how all concepts articulate together.
Prompt C — DRCA: Deconstruct the Field for PUMA
Role: Benchmark architect applying DRCA methodology.
Context: The literature on AI agents for project management is fragmented across three communities: NLP/LLMs, software engineering, and project management science. Each community uses different terminology and evaluation standards.
Objective: Deconstruct the field into atomic components and reconstruct a coherent benchmark blueprint for PUMA.
Instructions:
1. DECONSTRUCTION: Break the field into fundamental components: tasks, datasets, evaluation methods, agent architectures, human oversight mechanisms, reproducibility requirements.
2. HIDDEN ASSUMPTIONS: For each component, identify the unstated assumptions that limit generalizability.
3. RECONSTRUCTION: Rebuild a benchmark blueprint for PUMA that addresses the identified assumptions.
4. For each component: what does the literature support, what it does NOT solve, what PUMA should operationalize.
Format: Structured report with sections. At least one comparison table. Explicit PUMA design decisions derived from each component.
Prompt D — MOC Generation from Synthesis
Based on the theoretical framework and concept matrix above, generate a Map of Content (MOC) structure for the PUMA Obsidian vault. The MOC should:
1. List the main conceptual nodes (permanent notes) needed.
2. Show links between nodes using → notation.
3. Identify which concepts belong in: Zettelkasten (permanent ideas), Resources (external reference), Projects (PUMA-specific application).
4. Propose 10 atomic permanent note titles in declarative sentence format (each captures one reusable insight).
Format: Bulleted MOC structure + permanent note titles list.
PUMA Relevance
This prompt generates the intellectual backbone of PUMA Section 2 (Materials and Methods) and informs the Zettelkasten permanent notes in 31 Concepts. The DRCA deconstruction directly justifies why PUMA’s benchmark design differs from prior work.