🗺️ MOC — PUMA Master Map
Overview
Central navigation hub for the PUMA project. PUMA: Platform for Understanding and Management with Agents. “Can language models manage ICT projects?”
🎯 Project Identity
Full title: Pueden los modelos de lenguaje gestionar proyectos tecnológicos? PUMA: Plataforma de benchmark para la evaluación empírica de agentes en tareas de gestión de proyectos.
Research question: Do different LLM models and prompting strategies produce statistically significant differences in issue triage quality (F1-macro) and effort estimation (MAE) on real PM datasets with verified labels?
Hypotheses: EX-Hypotheses-H1-H2
MVP: Triage module (Stage 1) + statistical validation. Self-contained academic contribution.
📋 PUMA Project Structure → Vault Mapping
| PUMA Project | Content | Vault Location |
|---|---|---|
| 1. Introduction | Context, objectives, methodology, planning | PR-PUMA-Ch1-Introduction |
| 2. Materials & Methods | DSR + SLR + experiment design + stack | PR-PUMA-Ch3-Methods |
| 3. Results | F1-macro, MAE, Wilcoxon, carbon | PR-PUMA-Ch4-Results |
| 4. Conclusions | H1/H2 decision + future work (Smart PMO) | PR-PUMA-Ch5-Discussion |
| 5. Glossary | All definitions | Glossary-Master |
| 6. Bibliography | APA 7, ≥40 references | BIB-Master-APA7 |
| 7. Annexes | Templates, dataset prep, extended results | in project folders |
🔬 Experiment Design
| Stage | Task | Dataset | Metric | Status |
|---|---|---|---|---|
| 1 (MVP) 🟢 | Issue triage | Jira SR (200 stratified) | F1-macro ≥ 0.55 | 🔄 PEC2 |
| 2 🟢 | Effort estimation | TAWOS | MAE ≤ 3.0 SP | ⏳ PEC3 |
| 3 🟡 | Backlog prioritisation | TAWOS | Spearman ≥ 0.50 | ⏳ Conditional |
| 4 🔴 | RAG-enhanced triage | Jira SR | F1-macro > Stage 1 | ⏳ Optional |
| 5 🔴 | Smart PMO multi-agent | — | MTTD -30% | 🔭 Future work |
Prompting strategies: Zero-Shot · Few-Shot-3 · Few-Shot-6 · Chain-of-Thought
Models: Llama 3.2 8B · Mistral 7B · (Phi-3.5 Mini as fallback)
Reproducibility: seed=42, temperature=0, fixed requirements.txt
🏗️ Architecture
- SP-Architecture — 7-layer SwarmPM architecture
- SP-PUMA-Constitution — Non-negotiable principles
- SP-Triage-Agent — Triage agent spec
- BMAD-Agent-Roster — Multi-agent team
- BMAD-PRD-PUMA — Product requirements
📚 Key Literature
- LN-KeyPapers-CoGEE-Angermeir-Flyvbjerg — Core papers
- LN-Datasets-JiraSR-TAWOS — Datasets
- BIB-Master-APA7 — Full bibliography (42 refs)
Books — AI & Society:
- LN-Lawrence-2024-AtomicHuman — The Atomic Human: embodied intelligence, HITL theoretical basis
- LN-Suleiman-2023-ComingWave — The Coming Wave: AI governance and containment context
- LN-Shum-2025-PensarConPrompts — Pensar con Prompts: CO-STAR, prompt engineering taxonomy
Books — Agile & PM:
- LN-Beck-1999-XPExplained — XP Explained (2nd ed.): story points origin, TDD, adaptive development
- LN-Goldratt-2004-TheGoal — The Goal: Theory of Constraints; issue backlog as constraint system
Books — Business Systems (SmartPMO context):
- LN-Carpenter-2025-WorkTheSystem — Work the System: SOP documentation; systems mindset
- LN-Gerber-2009-EMythRevisited — The E-Myth Revisited: franchise prototype; working ON the business
- LN-Wickman-2012-Traction — Traction / EOS: execution operating system; Rocks; scorecard
- LN-Harnish-2022-ScalingUp — Scaling Up: Rockefeller Habits; Four Decisions framework
- LN-Price-2022-Frictionless — Frictionless Organization: CES; friction-free PM design
🧠 Key Permanent Notes
PM & Experiment concepts:
- PN-IssueTriage-StoryPoints — F1-macro, MAE, priority schema
- PN-CoT-FewShot-Prompting — Prompting strategies (S1–S4)
- PN-LLM-Local-vs-Cloud — Why local inference
- PN-RAG-Embeddings-VectorDB — RAG for Stage 4
- PN-ToolSelection-PUMA — Tool selection rationale for PUMA
Agent patterns & AI science:
- PN-KeyConcepts-Agents-Reproducibility-RedTeam — Agents, Reproducibility, Uniqueness Trap, Red Teaming
- PN-MultiAgent-ArchitecturePatterns — Specialisation (→ Smart PMO)
- PN-ReAct-AgentPattern — Stage 4 reasoning pattern
- PN-Agentic-Science-Paradigm — AI as active scientific agent
- PN-AI-Scientific-Knowledge-Generation — AI-generated scientific knowledge
- PN-PUMA-within-AgenticScience-Trajectory — PUMA’s place in the agentic science trajectory
- PN-ActiveReading-CognitivePractice — Active reading as cognitive practice
Research methods:
- PN-DSR-SLR-Methods — DSR + PRISMA
- PN-Wilcoxon-FINER-Cornell-PRISMA — Statistical protocol
Frameworks:
- PN-SDD-Framework — SDD + BDD + BMAD
- PN-RCOIF-Framework — Structured prompting
- PN-EGI-Framework — Exploratory guided interaction
- PN-AMI-DRCA-IIPR-Frameworks — AMI + DRCA + IIPR advanced prompting
- PN-MIT-Student-Method — MIT AI Lab active reading method
- PN-MIT-Student-Method-Complete — MIT AI Lab full Q1/Q2/Q3 + Keshav
- PN-PARA-GTD-Zettelkasten — PARA + GTD + Zettelkasten integration
Knowledge hub & Structure notes:
- ZK-Hub-PUMA — Full Zettelkasten index
- ST-Prompting-Strategies — Prompting strategies thematic cluster
- ST-Reproducibility-Cluster — Reproducibility crisis cluster
Sources & Persons:
- SRC-Keshav-2007-HowToReadPaper — Keshav 2007 Three-Pass paper
- SRC-MITAILab-WP316 — MIT AI Lab Working Paper 316
- PER-Keshav-Srinivasan — Three-Pass Method author
- PER-Flyvbjerg-Bent — Uniqueness Trap / Reference Class Forecasting
- PER-Yao-Shunyu — ReAct + Tree of Thoughts
- PER-Hong-Sirui-MetaGPT — MetaGPT multi-agent framework
- PER-Assalaarachchi-Nuwan — Agentic SPM vision
Results:
- RES-Results-Placeholders — Experiment results placeholders (Stage 1 & 2)
📊 Progress Dashboard
TABLE status AS "Status", deadline AS "Deadline", pec AS "PEC"
FROM "40 - Projects/PUMA"
WHERE type = "project-note"
SORT deadline ASC🔗 Linked MOCs
- MOC-Research-Pipeline — Research workflow
- MOC-Literature-Review — SLR state of the art
- MOC-LLM-Benchmarks-PM-AI — Benchmark landscape
- MOC-Methods-Frameworks — All methodologies
- MOC-Prompts-Library — Prompt templates
- MOC-Tools-Stack — Technology stack
MOC updated: April 2026 (PEC2)