The Origin Protocol
A recognition system for the AI era.
For AI Labs · Foundations · Investors · The First Cohort
Authored by Mohammad Rahimi
Universities, journals, festivals, accelerators, prize committees — all required infrastructure to enter. You needed the right degree, the right reference, the right flight, the right language. The talent existed everywhere. The recognition only existed in a few places. The two were never the same map.
The other half is what AVA exists to change.
A researcher in Tehran can run analyses that required a Stanford lab. A founder in Lagos can design systems that required a Y Combinator cohort. A creator in Karachi can produce work that required a New York gallery. AI removed the infrastructure barrier to producing serious work — anywhere.
But accelerators, journals, festivals, and prize committees still gatekeep through the old privilege networks. The talent that AI has unleashed is invisible — because the systems that used to discover talent were never built for a world where talent comes from everywhere.
In an age where anything can be generated, the only thing of permanent value is verified origin. AVA provides the cryptographic, multi-model, evidence-based proof that something is real, original, and made by who claimed to make it.
Each can be understood independently. They compose into a single trust pipeline.
A specialized AI investigator with the helpful-assistant posture inverted. Default stance is doubt, not trust. The applicant must prove their claim against active resistance. The entire dialogue becomes the certificate's primary evidence — Proof-of-Conversation.
Intake → Evidence → Five-Layer Analysis → Mandatory Challenge → Scoring → Issuance. No phase is skippable. Even apparently flawless applications are interrogated by the Devil's Advocate.
Independent AI vendors fetch the evidence pack, recompute hashes, re-score against the public rubric, and sign their own attestations. Confidence bands rise with vendor diversity, not vendor count.
Before any high-tier certificate is published, a panel of human reviewers signs off. AI does what humans cannot — examining 50,000 evidence packs identically. Humans do what AI cannot — accepting moral responsibility.
The default outcome is rejection. Evidence shifts it to issuance.
State precisely. Restated, classified, decomposed.
Documents hashed (SHA-256), timestamped, scanned for red flags.
Authenticity · Originality · Technical · Timeline · Consistency.
Devil's Advocate. Active search for grounds to reject. Not skippable.
Five criteria. Public rubric. Reasoning per score.
≥60 issues. Below rejects. Transcript attached either way.
A complete framework, methodology, or multi-module ecosystem. Architect of a coherent body of work spanning years and multiple disciplines.
A foundational technical contribution — a rule-set, standard, or specification that other systems can build on.
Proof of first creation. Anchors priority over a specific idea, design, or invention against a verified evidence chain.
Each mechanism eliminates a specific bias. Fairness is not promised — it is engineered.
Every applicant faces the same active resistance, regardless of how familiar their idiom looks to the model.
Even flawless-seeming applications are interrogated. Well-connected applicants get no shortcut.
No single model's training distribution decides recognition. Diversity is structural, not decorative.
Anyone can publish a Counter-Evidence VC against any certificate. Quorum rules trigger auto-downgrade.
Six dimensions, exact weights, thresholds — all public. Audit any score against the spec.
The highest in any startup-level accreditation system. Recognition without safety compounds the wrong incentives.
No hidden reasoning. No invisible committee deliberation. The proof is the conversation.
No single vendor decides. Forging a Platinum certificate would require coordinated misbehaviour across four competing AI companies — economically and technically intractable.
Claude · Anthropic
● Live
GPT · OpenAI
● Live
Gemini · Google
Phase 3
Open Source
Phase 3
Bronze ≥ 2 vendors · Silver ≥ 3 (2 families) · Gold ≥ 4 (3 families) + human audit · Platinum ≥ 6 (4 families) + audit + no disputes
Bands rise with vendor diversity and human audit. Sustained disputes trigger automatic downgrade.
≥ 2 independent attestations from any AI vendors.
≥ 3 attestations across ≥ 2 vendor families.
≥ 4 attestations across ≥ 3 vendor families + 1 human audit.
≥ 6 attestations across ≥ 4 vendor families + human audit + no critical disputes.
Every recognition system has a first case. The first Nobel Prize. The first YC batch. The first published paper in any journal. The first case answers the question: what kind of work is this system actually for?
"AVA was not built to recognize me. It was built so that the next thousand of me could be found."
— Mohammad Rahimi · First Precedent · UID AVA-2025-ML-0001
The fake-credential industry was already $7B+ before generative AI. Undetectable fabrication is now available to anyone. Existing verification was not built for this threat model.
Protocols, architectures, research projects are routinely developed in long conversations with frontier models. The body of work needs a new form of recognition that reads the reasoning traces themselves.
The next consequential creators will not live where the existing recognition systems live. A protocol that recognizes at distance, at depth, and at cost is no longer optional.
Concept and specification. Canonical narrative authored. AVA Auditor v3.0 system prompt published. Two vendor families (Claude, ChatGPT) operational. Visual identity and landing site live at avaverify.com.
Operational pilot. First cohort of certificates issued. OpenTimestamps anchoring on every certificate. Local registry (transparency ledger). Beta deployment with explicit disclosure.
Public registry and full MAIA. Live verification portal. Searchable public ledger. Gemini and open-source attestations integrated. Counter-evidence dispute mechanism. JSON-LD spec published.
Institutional posture. Anthropic or peer-lab integration. Human audit panel active. First Gold certificate issued. First Laureate prize funded ($1M+, zero equity).
License MAIA as the standard verification layer that frontier models honour by default. Zero-deployment-cost integration via the existing prompt card. Structural credibility for whichever lab adopts it first.
Sponsor the first Laureate prize. The most consequential AI-empowered creators of the next decade will be people the existing systems would never find. The Laureate is the highest-leverage mechanism for finding them while they can still be helped.
Five to ten creators willing to be the first verified Master Level cases. Their certificates form the credibility kernel of the protocol. The next thousand cases will be measured against them.
The recognition system the AI era requires does not yet exist. This is the specification for the one that should.