Deterministic AI research coordination through structured scientific methodology.

Diogenes makes AI agents produce defensible, auditable, evidence-based research instead of sycophantic confirmation of what you already believe. It combines nine frameworks from intelligence analysis, clinical medicine, climate science, systematic review methodology, institutional standards, and the philosophy of science into a structured process that constrains AI behavior through enforcement language.

A structured research methodology that makes AI agents produce defensible, auditable, evidence-based research. It combines frameworks from intelligence analysis (ICD 203), clinical medicine (GRADE, Cochrane, CONSORT), climate science (IPCC), systematic review methodology (PRISMA, ROBIS), institutional standards (NAS), and the philosophy of science (Chamberlin/Platt).

The methodology was developed to solve a specific problem: AI agents, when asked to "research this," default to building a case rather than conducting an investigation. They confirm what you expect, minimize contradictions, and present uncertain conclusions as settled. This methodology constrains that behavior through enforcement language — telling the AI not just what to do, but what it is prohibited from doing and why.

The detailed background behind this methodology — the framework evaluation, the design decisions, and the evidence for every feature — is discussed in a pair of articles:

• The Truth is Out There. But How Do You Find It? — the what and the why (Part 1)
• The Truth is Out There. Now Go Find It. — the how and how to get it (Part 2)

• Claims — factual assertions to verify. Each claim is tested against competing hypotheses with evidence scored for reliability, relevance, and bias.
• Queries — research questions to answer. Each question generates hypotheses ranked by evidence strength.
• Axioms — facts declared by the researcher that must be assumed true during the investigation. Not tested — they function as constraints that frame the research.

All three can be combined in a single research run.

Research produces complete evidence archives: source scorecards, search logs, hypothesis evaluations, collection-level synthesis, gap identification, and a five-domain self-audit.

1. Claim/query received and clarified — ambiguities surfaced, assumptions identified, axioms acknowledged
2. Vocabulary exploration — map terminology across domains before searching
3. Competing hypotheses generated (Chamberlin/Platt) — minimum three
4. Discriminating searches designed — what would disprove each hypothesis?
5. Searches executed and logged (PRISMA) — every search documented
6. Per-source scoring (GRADE + Cochrane) — reliability, relevance, six bias domains
7. Collection-level synthesis (IPCC) — evidence quality, source agreement, independence
8. Probability assessment (ICD 203) — nine-point calibrated scale (including deterministic endpoints)
9. Gap identification (NAS) — what's missing and what it means
10. Process self-audit + source-back verification (ROBIS + net-new) — five-domain bias check including interpretation verification
11. Report with revisit triggers (ICD 203) — every claim sourced, every judgment explicit, specific conditions for re-research identified
12. Temporal revisitation archive — enable periodic re-execution

The methodology includes explicit behavioral constraints that target AI's most dangerous default behaviors:

• Evidence from research outranks training data
• The researcher's claims are inputs to test, not truths to confirm
• Contradictory evidence must be highlighted, not minimized
• Embedded assumptions must be surfaced and tested
• Uncertainty must be stated explicitly
• The AI cannot declare victory early — the full process runs every time

• Claude Code plugin — installs as a skill, integrates directly into Claude Code sessions
• Standalone prompt — works with any capable LLM (Claude, ChatGPT, Gemini, or others)
• dio CLI — command-line interface for scripted research workflows

The main repository has everything you need: diogenes-project/diogenes

Contributions are welcome. See the contributing guidelines for development setup, workflow, and the AI contributor identity model.

Built by Phillip Moore. The detailed background behind this methodology — the framework evaluation, the design decisions, and the evidence for every feature — is discussed at The Infrastructure Mindset.