Pre-Competition Research Archive

The following posts document the research and strategic thinking behind LogoMesh during and leading up to the UC Berkeley RDI AgentBeats competition (2025–2026). This work laid the foundation for the Contextual Integrity Score (CIS) and the adversarial multi-agent evaluation framework that powers the GitHub App today.

• Your Code Has Amnesia: An Introduction to Contextual Debt

  The original framing: why AI-generated code loses its "why," and what that costs.

• The Unknowable Code: Contextual Debt as a Security Liability

  Why contextual debt is a legal and regulatory risk, not just a technical one.

• The Breakthrough: The Cyber-Sentinel Agent

  Introducing the multi-agent evaluation architecture and the "glass box" approach.

• The Developer Tooling Opportunity

  Market thesis: context, not code, is the new bottleneck in the AI coding era.

• Architecting Resilience: A Proposal for the Contextual Integrity Score

  The formal framework behind CIS — Rationale, Architectural, and Testing Integrity scores.

• Project Plan: AgentBeats Green Phase

  The 6-week build plan that took LogoMesh from concept to 1st-place finish.

• Project Phoenix: The LogoMesh Hard Reset

  Honest accounting of the original codebase failures and the architectural principles we enforced.

← Back to homepage