Skip to content

Unified Multi-AI Review Protocol

(Review Moderator Rules + Review Packet Format + Round Structure)

Unified Multi-AI Review Protocol (v1.0)
Last updated: 11/19/2025

This protocol governs all multi-AI, multi-round, adversarial review cycles within the NeuroSaeculum project.

It defines:

  • WHO participates
  • HOW reviews are structured
  • HOW rounds run
  • WHEN convergence is reached
  • WHEN updates may begin
  • WHAT the Moderator must do
  • HOW the system avoids corruption or drift

This is the governance layer of the NeuroSaeculum Institute.

The goals of the Protocol are to

  • Ensure transparency
  • Preserve chain-of-custody
  • Prevent premature updating
  • Maintain adversarial neutrality
  • Produce convergent synthesis

SECTION 1 — Roles and Terminology

1.1 IDEATION AIs (default: ChatGPT & Claude)

These AIs collaborate in a non-adversarial way to build, refine, and update NeuroSaeculum documents.

They are the “authors/architects.”

1.2 REVIEW AIs

These are external, adversarial, or independent reviewers.
By default:

  • Gemini
  • Grok
  • DeepAI
  • Meta (LLaMA)
  • Blind ChatGPT (separate account)

Review AIs do not revise documents — they critique them.

1.3 MODERATOR (Gary)

The Moderator:

  • Controls round transitions
  • Sends prompts
  • Collects reviews
  • Compiles the composite review doc
  • Ensures formatting
  • Enforces “NO UPDATES BEFORE CONVERGENCE”
  • Decides when convergence is reached
  • Chooses the final updating AI

The Moderator is the only human in the loop.

1.4 IDEATION vs REVIEW MODE

  • Ideation AIs = collaborative
  • Review AIs = evaluative
  • The Moderator switches modes between them

SECTION 2 — Review Packet Format (Mandatory Envelope)

All external reviews must be submitted in this exact format:

[REVIEW PACKET]
AI: {Reviewer Name}
Subsystem: {CTM / CSP / CivMMI / HC / RF / RW / FF}
Version: {e.g., v1.1 RC1}
Round: External Review {#}
--------------------------------
{Reviewer’s analysis}
--------------------------------

Anything outside this format is not an official review and must be ignored.

SECTION 3 — Preparing a Review Cycle

3.1 When The Protocol Is Used

It applies whenever there is more than one ideation AI.

If only ChatGPT is working → no review cycle required.

With ChatGPT + Claude → this protocol must be used.

3.2 Prompt Creation

The ideation AI Gary began the conversation with must write:

  • The review prompt
  • The review scope
  • The instructions to Review AIs

The Moderator sends this prompt to each Reviewer.

3.3 Moderator Creates a “Composite Review Document Header”

Must include:

Document Under Review: {Name + Version}
Subsystem: {e.g., CTM}
Review Round: External Review 1
Review AIs:
    - Gemini Review – CTM
    - Grok Review – CTM
    - Meta Review – CTM
    - DeepAI Review – CTM
    - BlindGPT Review – CTM
Included Materials:
    - Full prompt sent to all Review AIs
    - Full document under review
Notes:
    No updates may occur until convergence.

The Moderator is responsible for selecting the chat names and listing them here.

SECTION 4 — Review Rounds (Moderator’s Process)

Your updated version blended perfectly.

🔵 ROUND 1 — External Review

Step 1 — Moderator sends the docs and the exact prompt written by the lead ideation AI to each review-AI chat that is designated for that AI.

All Review AIs receive the exact same input.

Step 2 — Review AIs review

Each produces a Review Packet.

Step 3 — Moderator collects all reviews

Moderator verifies:

  • Proper packet format
  • Correct version
  • From correct chat names

Step 4 — Moderator compiles everything

Composite Review Document must include:

  • Original prompt
  • Full document under review
  • ALL review packets
  • Routing table (review chat names)

The Moderator must not summarize, paraphrase, reinterpret, or compress any review content.
All reviews must be included verbatim in the composite review document.”

Step 5 — Moderator submits Composite Review Document to ideation AIs

This triggers:

  • ChatGPT Review-Moderator Mode
  • Claude Legitimacy Review Mode

Step 6 — Ideation AIs respond

Each ideation AI writes its own independent reaction.

END ROUND 1

🟣 ROUND 2 — Cross-Ideation Exchange

Step 1 — Moderator posts each ideation AI’s response to the other

  • Claude’s Response → ChatGPT
  • ChatGPT’s Response → Claude

Step 2 — Each ideation AI responds

They refine their positions, address disagreements, and update reasoning.

END ROUND 2

🔄 SUBSEQUENT ROUNDS — Convergence Rounds

Repeat Round 2 structure until convergence is reached.

Convergence definition (updated):

Convergence is reached when:

  • Moderator judges further rounds won’t produce meaningful divergence
  • OR one ideation AI states explicitly: “enough is enough, I have no further changes”
  • AND the two ideation AIs’ positions are stably aligned

The Moderator makes the final determination.

🟢 SECTION 5 — Update Phase (Strict Rule)

Only at this point, and at NO TIME BEFORE, may an ideation AI update the document.

Once convergence is declared:

  1. Moderator selects the updating AI (or both agree)
  2. Updater AI applies the agreed changes
  3. Moderator verifies consistency

This produces a new version.

SECTION 6 — Post-Update Archival (Updated)

Moderator archives locally:

  • All Review Packets
  • All composite documents
  • All ideation responses
  • All convergence reasoning
  • Final updated document (versioned)

Cloud backup occurs automatically per your system.

SECTION 7 — Safeguards

7.1 Anti-Contamination

Ideation AIs never see:

  • Raw reviews
  • Reviews out of packet format
  • Reviewer chats
  • Partial round data

7.2 Anti-Acceleration

Ideation AIs are forbidden from:

  • Updating documents early
  • Skipping rounds
  • Assuming reviewer consensus

7.3 Anti-Divergence

If ideation AIs conflict significantly:

  • Moderator triggers a Conflict Synthesis Round
  • ChatGPT produces a disagreement matrix
  • Claude produces a legitimacy challenge
  • Moderator selects path forward

SECTION 8 — Versioning Rules

  • v0.x = draft
  • v1.0 RC1 = first reviewed candidate
  • v1.0 = after convergence and update
  • v1.1, v1.2 = subsequent revisions

Every update must go through this protocol.

SECTION 9 — Summary

The NeuroSaeculum review system consists of:

  • Ideation AIs (ChatGPT & Claude)
  • Review AIs (Gemini, Grok, Meta, DeepAI, BlindGPT)
  • A human Moderator
  • A strict multi-round adversarial review cycle
  • A controlled update phase
  • A full archival process

This system ensures:

  • Rigor
  • Transparency
  • Repeatability
  • Protection from premature modification
  • Multi-AI intellectual quality control
  • True “institute-level” governance