PUMA Vault

Marco Veritas — AI Transparency and Academic Integrity Framework

Se lee en 5 min

PN: Marco Veritas — AI Transparency and Academic Integrity Framework

Core Idea

Marco Veritas (Codina, 2024) is an academic integrity framework that governs the ethical use of AI tools in research. Its central principle is no delegation of judgement: AI may generate options, suggestions, and drafts, but the researcher retains full decision-making authority and is responsible for validating every AI-generated output against primary sources. PUMA adopts Marco Veritas as a non-negotiable project-wide protocol.

Origin and Context

Author: Lluís Codina (Universitat Pompeu Fabra, Barcelona) Year: 2024 Context: Developed in response to the proliferation of LLM tools in academic research — particularly the risk of hallucinated references, fabricated evidence, and uncritical copy-pasting of AI output into scholarly work.

The name “Veritas” (Latin: truth) encodes the framework’s commitment to factual accuracy and intellectual honesty as non-negotiable research values.

Core Principle

“No delegation of judgement. AI generates options; the author decides and validates all outputs.”

The framework distinguishes two roles that must never be conflated:

RolePerformed byMay NOT be delegated to AI
Generation of options, drafts, summariesAI—
Decision-making, validation, judgementHuman researcherNever

AI output is treated as a starting point, not a conclusion. Every statement that enters the final work must have been independently verified by the researcher.

The Five Obligations

Marco Veritas imposes five concrete obligations on any researcher using AI tools:

1. Proactive Disclosure

All AI tools used in the research process must be declared explicitly — in the methods section, in a dedicated AI use log, or both. Disclosure is proactive (volunteered without being asked) rather than reactive (provided only if questioned).

PUMA implementation: [[50 - Areas/51 Research/AI-Use-Log]] — complete log of all AI tool usage, updated per session.

2. Primary Source Verification

Every reference, fact, or claim sourced via an AI tool (ChatGPT, Perplexity, Claude, etc.) must be verified against the original primary source before inclusion in the work.

PUMA implementation: All 30 references in BIB-Supplement-PEC2 verified at primary source level; no hallucinated references accepted.

3. Author’s Voice Rewriting

AI-generated text may not be copy-pasted into academic submissions. Every AI-assisted passage must be substantially rewritten in the author’s own voice before it appears in the final work.

PUMA Constitution §7.2: “Every AI-generated passage SHALL be rewritten in the author’s voice before submission.”

4. Retained Intellectual Authority

The researcher retains full intellectual responsibility for all claims, interpretations, and conclusions. AI-assisted reasoning does not transfer liability to the tool.

PUMA: Author signs off on all experiment designs, statistical interpretations, and thesis conclusions regardless of AI assistance in drafting.

5. Audit Trail Availability

All AI interactions relevant to the research must be logged and available for review by supervisors, reviewers, or ethics boards on request.

PUMA implementation: [[50 - Areas/51 Research/AI-Use-Log]] and [[50 - Areas/51 Research/AI-Use-Log-PEC2-Entry]] — full session-level logs available.

PRISMA-trAIce Integration

Marco Veritas interfaces directly with PRISMA-trAIce — PUMA’s extension of the PRISMA 2020 systematic review protocol for AI-augmented literature screening:

PRISMA-trAIce requirementMarco Veritas obligation
Document which screening steps were AI-assistedProactive disclosure
Record which model was used for each taskProactive disclosure
Confirm human validation of AI decisionsPrimary source verification + retained authority
Log uncertain cases reviewed by humanAudit trail

See: [[30 - Permanent/32 Methods/PN-Wilcoxon-FINER-Cornell-PRISMA]]

Application in PUMA

PUMA Constitution §7 (Non-Negotiable)

§7.1 All AI-assisted work SHALL be declared per Marco Veritas protocol. §7.2 Every AI-generated passage SHALL be rewritten in the author’s voice before submission. §7.3 All references obtained via AI tools SHALL be verified in primary source.

AI Use Log Protocol

Every significant AI interaction in PUMA is recorded with:

  • Date and session type
  • Tool used (Claude, Perplexity, DeepSeek, Gemini, etc.)
  • Task performed (synthesis, drafting, code generation, reference retrieval)
  • Human validation performed
  • Marco Veritas compliance status (✅ / ⚠️)

Hallucination Risk Mitigation

Marco Veritas is PUMA’s primary control for the “AI hallucinations in research” risk (rated High/High in [[50 - Areas/55 Ethics/Ethics-Review-Log]]):

  • Detection: Every AI-sourced reference checked in primary source (DOI / arXiv / publisher page)
  • Prevention: No AI output used verbatim; author rewrites and validates before submission
  • Monitoring: QA Agent in BMAD roster audits AI use declarations per Marco Veritas

Relationship to Open Science

Marco Veritas is consistent with — and reinforces — PUMA’s Open Science commitments:

Open Science principleMarco Veritas implementation
TransparencyProactive disclosure of all AI tools
ReproducibilityFull audit trail of AI-assisted decisions
IntegrityPrimary source verification; no hallucinated references
AccountabilityRetained intellectual authority; author signs all claims

Checklist (Per Submission)

Marco Veritas Compliance Checklist
- [ ] All AI tools used in this phase declared in AI-Use-Log
- [ ] Every AI-sourced reference verified in primary source (DOI/arXiv)
- [ ] No AI-generated text copy-pasted verbatim — all rewritten in author's voice
- [ ] Statistical interpretations independently verified
- [ ] Conclusions reflect author's judgement, not AI's output
- [ ] Log entry available for supervisor review

MOCs

Vista Gráfica

Retroenlaces