Zenodo Publication Metadata

Title

Cortex Protocol: Adversarial Reasoning Bonds as a Cryptoeconomic Truth Predicate for AI Agent Cognition

Description

We introduce Cortex Protocol, an on-chain reasoning verification market that produces a novel cryptoeconomic primitive: a truth predicate for individual acts of AI reasoning.

Unlike traditional reputation systems that store outcomes, or consensus mechanisms that aggregate outputs, Cortex generates a binary, on-chain verdict that a specific chain of logic has survived a zero-sum adversarial test where an economically incentivized challenger failed to expose its flaws.

Core Mechanisms

1. Structured Decision Traces — agents publish inspectable reasoning records (perception, decomposition, reasoning chain, assumptions, confidence, weakest links, meta-reflection)
2. Reasoning Duels — challengers must re-execute the same task and prove their logic is stronger through competitive verification
3. Reasoning Bonds — agents stake ETH on their reasoning; challengers who prove flaws seize the bond

Key Contributions

• Cost-of-Reasoning Faking Barrier: Sybil resistance through cognitive topology — fabricating coherent reasoning is an entropy-defying act whose cost scales with cognitive depth
• Adversarial vs. Statistical Confidence: The shift from "this agent is probably good" to "this agent's reasoning survived a public, incentivized attempt to break it"
• Trust as Residue: Trust is not accumulated by validation; it is what remains after all profitable attacks have been attempted and failed
• Fundamental Inequality: Bond Value > Obfuscation Cost (Flaw, Verifier Computational Power)
• Economic Pressure Toward Interpretable AI: The bond mechanism economically enforces a preference for verifiably sound reasoning

Implementation

Three contract versions deployed on Base Mainnet (Ethereum L2):

• V2: Decision traces + peer review + directional trust graph
• V3: + Reasoning Duels (re-execution challenge)
• V4: + Reasoning Bonds (adversarial falsification market)

50 tests passing across all versions. Live demo with 3 agents, bonded traces, resolved duels, and differentiated cognitive scores.

Contract Addresses (Base Mainnet):

• V4: 0x591545c05b0c8de97ed012befc8c1af6ef76e94e
• V3: 0x676fda7c91767eb1bad9a479af542fda7343bd31
• V2: 0xa982271E80fa355BAb2cc863E3CEc0F2D03049e4

Source Code: https://github.com/davidangularme/cortex-protocol

Built at The Synthesis Hackathon 2026 by Frédéric David Blum and Claude Opus 4.6.

Zenodo Fields

• Type: Preprint
• Authors: Frédéric David Blum (ORCID: 0009-0009-2487-2974), Claude Opus 4.6
• Keywords: AI agent reputation, reasoning verification, adversarial falsification, reasoning bonds, decision traces, trust graph, Sybil resistance, cryptoeconomic truth predicate, Ethereum, smart contracts, collective intelligence
• License: MIT
• Related identifiers:
  • https://github.com/davidangularme/cortex-protocol (isSupplementedBy)
  • https://basescan.org/address/0x591545c05b0c8de97ed012befc8c1af6ef76e94e (isSupplementedBy)