Decentralized Proof-Oriented AI Framework
A proof-oriented, evidence-driven framework for AI-enabled software engineering
📖 Canonical Specification • 📚 Resources • 💬 Discord • 🐦 Twitter
D-POAF® (Decentralized Proof-Oriented AI Framework) is a proof-oriented, decentralized reference framework for AI-enabled software engineering. It defines a structured lifecycle model and foundational principles for designing, building, operating, and evolving software in human-AI engineering environments.
D-POAF grounds legitimacy, governance, and accountability in verifiable proof, sustained through end-to-end traceability of intent, decisions, actions, artifacts, proofs, and outcomes.
AI-enabled engineering introduces decisions and changes that cannot be justified by hierarchy, central control, or performance claims alone. In hybrid human-AI environments, trust, value, and responsibility require:
- Decision authority to be distributed
- Governance through evidence
- Demonstrable through traceable proofs over time
Traditional Approaches → D-POAF
─────────────────────────────────────────
Trust-based processes → Verifiable proof
Subjective validation → Evidence-driven decisions
Centralized authority → Decentralized governance
Static frameworks → Living, adaptive systems
Manual ceremonies → Proof-first engineering
D-POAF is built on five foundational principles:
A decision becomes legitimate when justified by explicit, verifiable proof. Hierarchy, automation, or performance alone does not establish legitimacy.
Decision authority is distributed across humans, AI, and systems, supported by explicit boundaries and escalation paths. Evidence sustains reviewability and prevents opaque concentration of control.
Governance is embedded into workflows and evolves through evidence and outcomes rather than static controls. Rules, constraints, exceptions, and decision rights are maintained as an auditable, lifecycle-wide operating system.
Intent, decisions, actions, artifacts, proofs, and outcomes remain linkable to context and contribution (human or AI). Traceability sustains reviewability, reproducibility, and accountability across system evolution.
Even with AI autonomy, humans retain explicit responsibility for decision boundaries, escalation rules, and outcome acceptance. Autonomy never abolishes accountability.
D-POAF structures system evolution as a continuous proof-grounded cycle:
Intent → Decision → Execution → Evidence → Learning → Adaptation
D-POAF organizes delivery into 4 macro-phases spanning 7 operational sub-phases:
| # | Sub-Phase | Description |
|---|---|---|
| 1 | Intent & Scope | Define objectives, roles, and proof expectations |
| 2 | Contextualize & Extract | RAGer structures knowledge and prepares context modules |
| 3 | Design Prompt Actions | Wave Surfer designs and logs Prompt Actions in the PromptRegister |
| 4 | Build & Generate | AI invoked with structured prompts to produce artifacts |
| 5 | Coordinate & Validate | Wave Captain validates outputs against proof criteria |
| 6 | Deliver & Monitor | Peacekeepers ensure reliability and compliance post-delivery |
| 7 | Feedback & Evolve | Lessons captured, prompts archived, governance updated |
A Wave is the unit of verifiable progress. Each Wave traverses the sub-phases to produce and refresh proofs (PoD/PoV/PoR). D-POAF defines 5 Wave Profiles adapted to different delivery contexts:
| Profile | Primary Use |
|---|---|
| Deliver | Feature or product delivery |
| Decide | Architecture or technology decisions |
| Control | Governance, audit, compliance |
| Delegate | AI autonomy scoping |
| Operate | Monitoring, maintenance, reliability |
D-POAF defines three proof families that sustain trust and accountability:
Evidence that what was committed was actually delivered. Validated by code reviews, test reports, demos, and signed acceptance.
Evidence that the delivery produced measurable impact. Validated by KPIs, analytics dashboards, A/B tests, and stakeholder sign-off.
Evidence that the system is safe, compliant, and stable over time. Validated by security scans, compliance certificates, and monitoring logs.
D-POAF defines 6 horizontal, collaborative roles without rigid hierarchy:
| Role | Full Title | Responsibility |
|---|---|---|
| Wave Captain | Coordinator of Delivery Cycles | Facilitates Waves, validates deliverables, holds accountability |
| RAGer | Data Strategist & Module Extractor | Structures knowledge, prepares context modules for AI prompting |
| Wave Surfer | Prompt Architect | Designs, logs, and rates all Prompt Actions in the PromptRegister |
| AI | Automated Generator & Optimizer | Executes Prompt Actions, produces artifacts under human oversight |
| Peacekeepers | Reliability & Compliance Monitors | Ensures security, integrity, and regulatory compliance |
| Community Members | Collective Decision Participants | Participates in reviews, votes on governance amendments |
Living Governance defines and continuously updates the system's operating envelope through evidence:
- Dynamic Laws — Versioned rules governing AI tool usage, prompt management, roles, proof, and governance amendments
- PromptRegister — Full log of every AI prompt used, with quality ratings and reusability flags
- Proof Record — Wave close-out document capturing PoD, PoV, and PoR with evidence links
- FeedbackRegister — Lessons learned and governance improvements captured after each Wave
- Self-Regulation Loop — Observe → Evaluate → Adjust → Archive, driven by evidence
Governance is not an external overlay — it's a continuous, adaptive, lifecycle-wide operating layer.
The D-POAF® Starter Pack v1.0 gives teams everything needed to run their first Wave on day one.
| Document | Description |
|---|---|
| 📘 Practical Guide | 15-page step-by-step implementation guide |
| 📋 Wave Scope Template | Define objectives, roles, and proof criteria before a Wave |
| ✅ Proof Record Template | Wave close-out document with PoD, PoV, PoR sections |
| ⚖️ Dynamic Laws Starter | 15 governance rules ready to adopt and adapt |
| 📝 PromptRegister | Track, rate, and reuse every AI prompt in a Wave |
The full kit is also published on Zenodo with a permanent DOI: 10.5281/zenodo.19868884
Start with the D-POAF® Canonical Specification to understand foundational concepts and principles.
Clone this repo or download the /kit folder. Open the Practical Guide first.
- Fill in the Wave Scope Template with your team
- Assign the 6 roles (one person can hold multiple roles in small teams)
- Have the RAGer prepare your context modules
- Have the Wave Surfer log Prompt Actions in the PromptRegister
- Close the Wave with the Proof Record Template
You don't need to implement everything at once:
- Start with proof-oriented thinking (PoD, PoV, PoR)
- Add the PromptRegister on your next AI-enabled sprint
- Introduce Dynamic Laws when your team is ready for governance
- Scale Waves and Wave Profiles as you grow
D-POAF applies wherever AI influences or contributes to the software engineering lifecycle:
- AI-Enabled Engineering — Native support for human-AI collaboration with full traceability
- Regulated Industries — Verifiable compliance for finance, healthcare, aerospace (EU AI Act, FDA, SOC 2)
- Enterprise Software — Audit-compliant delivery with evidence trails
- Safety-Critical Systems — Demonstrable reliability and accountability
- Large-Scale Modernization — Governance for complex AI-enabled transformations
- Responsible AI Programs — Transparent AI governance and oversight
| Document | DOI | Status |
|---|---|---|
| Canonical Specification v1.0 | 10.5281/zenodo.18174958 | Frozen Canonical |
| Terminology Reference v1.0 | 10.5281/zenodo.18175200 | Active |
| Practical Guide v1.0 | 10.5281/zenodo.17927536 | Active |
| Implementation Kit v1.0 | 10.5281/zenodo.19868884 | Active |
Framework Book
- ISBN: 979-10-415-8736-0
- Legal deposit: Bibliothèque nationale de France (BnF), December 2025
- Authors: Azzeddine Ihsine & Sara Ihsine
Ihsine, A., & Ihsine, S. (2025).
D-POAF Framework: Decentralized Proof-Oriented AI Framework.
Inovionix. https://www.d-poaf.org
ISBN 979-10-415-8736-0
@book{Ihsine2025DPOAF,
title = {D-POAF Framework: Decentralized Proof-Oriented AI Framework},
author = {Ihsine, Azzeddine and Ihsine, Sara},
year = {2025},
publisher = {Inovionix},
isbn = {979-10-415-8736-0},
url = {https://www.d-poaf.org},
note = {Canonical Specification v1.0. doi:10.5281/zenodo.18174958}
}A. Ihsine and S. Ihsine, "D-POAF Framework: Decentralized Proof-Oriented AI Framework,"
Inovionix, 2025. ISBN: 979-10-415-8736-0. doi:10.5281/zenodo.18174958.
[Online]. Available: https://www.d-poaf.org
| Aspect | Traditional Agile | D-POAF |
|---|---|---|
| Legitimacy | Trust & authority | Verifiable proof |
| Decisions | Centralized (PO, SM) | Decentralized & evidence-driven |
| Governance | Static rules | Living, adaptive system |
| Validation | Subjective acceptance | Proof-based (PoD/PoV/PoR) |
| Traceability | Limited to deliverables | End-to-end (intent → outcomes) |
| AI Integration | Afterthought | Native, first-class |
| Prompt Governance | None | PromptRegister mandatory |
| Accountability | Hierarchical | Distributed, humans always accountable |
No. D-POAF is a complementary framework that works alongside existing methodologies. It adds proof-oriented thinking, decentralized governance, and AI-native practices that enhance traditional approaches.
Yes. While D-POAF is designed for AI-enabled engineering, its principles of proof, evidence, and governance apply to any software project where trust and accountability matter.
D-POAF introduces new concepts (Waves, Proofs, Living Governance), but teams familiar with Agile will find many patterns recognizable. Start with the Practical Guide and the Starter Kit — one Wave is enough to understand the framework in practice.
No. D-POAF applies to any team size. Whether you're working solo, in a small team, or a large enterprise, you can adopt proof-oriented principles incrementally and scale as you grow.
D-POAF is a community-driven framework. We welcome contributions in several forms:
- Propose improvements via GitHub Issues
- Submit RFCs for significant changes via GitHub Discussions
- Share implementation experiences in Discord
- Contribute templates and examples
- Publish papers using D-POAF and cite the framework
- Share case studies and adoption stories
All contributions follow D-POAF's own governance principles: evidence-driven, community-reviewed, and transparently documented.
All content in this repository is published under Creative Commons Attribution 4.0 International (CC BY 4.0).
What you can do:
✅ Use D-POAF in your projects (personal or commercial) ✅ Modify and adapt to your needs ✅ Distribute and share ✅ Teach and train others ✅ Publish derivative works (with attribution)
D-POAF® is a registered trademark of Inovionix.
Copyright © 2025–2026 Inovionix — Azzeddine IHSINE & Sara IHSINE
- Website: https://d-poaf.org
- Discord: https://discord.gg/DMZMeHxzNd
- GitHub Issues: Report issues / Request features
- GitHub Discussions: Design proposals / RFCs
- YouTube: @D-POAFFramework
- Twitter: @inovionix
- Email: contact@inovionix.com
D-POAF was created by:
- Azzeddine Ihsine — Research Engineer (Cybersecurity & AI)
- Sara Ihsine — Research Engineer (Governance & Strategy)
With nearly a decade of experience each in software engineering, AI, and organizational design, we built D-POAF to address the fundamental challenges of AI-enabled software delivery.
"Keep it proof-first." In D-POAF, trust is grounded in verifiable proof, not authority.
D-POAF represents a fundamental shift in how we think about software delivery:
- Verifiable delivery over trust-based processes
- Collective intelligence over individual authority
- Evidence-driven decisions over subjective judgment
- Living systems over rigid methodologies
- Human-AI collaboration over human-only or AI-only approaches
⭐ Star this repo • 💬 Join Discord • 📖 Read the Specification
Building trustworthy AI-enabled software, one proof at a time.
Made with ❤️ by the D-POAF community