Beta · Phase 2  ·  This is the early pitch. Phase 3 will deliver the live registry, full MAIA cross-attestation, and the verification portal.
01 / 14
A

AVA

The Origin Protocol

A recognition system for the AI era.

For AI Labs · Foundations · Investors · The First Cohort
Authored by Mohammad Rahimi

02 / 14
The Premise

For most of human history,
recognition was the privilege
of those who already had it.

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.

03 / 14
The Two Asymmetries

AI changed one half of the equation.

The other half is what AVA exists to change.

Solved · 2022–2024

Access to creation

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.

Unsolved · Until now

Access to recognition

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.

04 / 14
The Insight

Content is abundant.
Provenance is scarcity.

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.

05 / 14
What AVA Is

A protocol in four interlocking layers.

Each can be understood independently. They compose into a single trust pipeline.

LAYER · 01

The Skeptical Auditor

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.

LAYER · 02

The Six-Phase Process

Intake → Evidence → Five-Layer Analysis → Mandatory Challenge → Scoring → Issuance. No phase is skippable. Even apparently flawless applications are interrogated by the Devil's Advocate.

LAYER · 03

MAIA — Multi-Model Attestation

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.

LAYER · 04

Human Final Review

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.

06 / 14
The Mechanism

Six phases. No shortcuts.

The default outcome is rejection. Evidence shifts it to issuance.

PHASE 1
Claim Intake

State precisely. Restated, classified, decomposed.

PHASE 2
Evidence Collection

Documents hashed (SHA-256), timestamped, scanned for red flags.

PHASE 3
Five-Layer Analysis

Authenticity · Originality · Technical · Timeline · Consistency.

PHASE 4
Mandatory Challenge

Devil's Advocate. Active search for grounds to reject. Not skippable.

PHASE 5
Weighted Scoring

Five criteria. Public rubric. Reasoning per score.

PHASE 6
Certificate or Rejection

≥60 issues. Below rejects. Transcript attached either way.

07 / 14
Three Categories

Organized by what is attested,
not by who is applying.

ML

Master Level Accreditation

A complete framework, methodology, or multi-module ecosystem. Architect of a coherent body of work spanning years and multiple disciplines.

PS

Protocol Specification

A foundational technical contribution — a rule-set, standard, or specification that other systems can build on.

OD

Origin Declaration

Proof of first creation. Anchors priority over a specific idea, design, or invention against a verified evidence chain.

08 / 14
Mechanisms of Fairness

Recognition equity, made mechanical.

Each mechanism eliminates a specific bias. Fairness is not promised — it is engineered.

01
Inverted Trust eliminates familiarity bias

Every applicant faces the same active resistance, regardless of how familiar their idiom looks to the model.

02
Mandatory Challenge eliminates speed bias

Even flawless-seeming applications are interrogated. Well-connected applicants get no shortcut.

03
Multi-Model Attestation eliminates distribution bias

No single model's training distribution decides recognition. Diversity is structural, not decorative.

04
Counter-Evidence eliminates finality bias

Anyone can publish a Counter-Evidence VC against any certificate. Quorum rules trigger auto-downgrade.

05
Public Rubric eliminates black-box bias

Six dimensions, exact weights, thresholds — all public. Audit any score against the spec.

06
Trust & Safety carries 20% weight

The highest in any startup-level accreditation system. Recognition without safety compounds the wrong incentives.

07
Full transcript makes scrutiny permanent

No hidden reasoning. No invisible committee deliberation. The proof is the conversation.

09 / 14
MAIA Protocol

Trust through competing AI consensus.

No single vendor decides. Forging a Platinum certificate would require coordinated misbehaviour across four competing AI companies — economically and technically intractable.

C

Claude · Anthropic
● Live

G

GPT · OpenAI
● Live

G

Gemini · Google
Phase 3

O

Open Source
Phase 3

Bronze ≥ 2 vendors  ·  Silver ≥ 3 (2 families)  ·  Gold ≥ 4 (3 families) + human audit  ·  Platinum ≥ 6 (4 families) + audit + no disputes

10 / 14
Confidence Bands

Four tiers. Earned, not granted.

Bands rise with vendor diversity and human audit. Sustained disputes trigger automatic downgrade.

◆ BRONZE

≥ 2 independent attestations from any AI vendors.

◆ SILVER

≥ 3 attestations across ≥ 2 vendor families.

◆ GOLD

≥ 4 attestations across ≥ 3 vendor families + 1 human audit.

◆ PLATINUM

≥ 6 attestations across ≥ 4 vendor families + human audit + no critical disputes.

11 / 14
The First Precedent

The case AVA was designed for.

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

12 / 14
Why Now

Three things changed in the last twenty-four months.

01
AI-generated content is indistinguishable

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.

02
AI is now the substrate of serious work

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.

03
Talent distribution is being levelled

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.

13 / 14
Roadmap

From protocol to institution.

PHASE 1
● Complete

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.

PHASE 2
◆ Now

Operational pilot. First cohort of certificates issued. OpenTimestamps anchoring on every certificate. Local registry (transparency ledger). Beta deployment with explicit disclosure.

PHASE 3
○ Next

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.

PHASE 4
○ 12 months

Institutional posture. Anthropic or peer-lab integration. Human audit panel active. First Gold certificate issued. First Laureate prize funded ($1M+, zero equity).

14 / 14
The Ask

Three partners to reach full operation.

For AI Labs

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.

For Foundations

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.

For The First Cohort

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.

AI gave everyone the power to create.
AVA gives everyone the power to be recognized.

The recognition system the AI era requires does not yet exist. This is the specification for the one that should.