Writing Hub
AI governance essays, reasoning systems notes, experiment logs, and technical writing across BioAI and engineering practice.
Project Topics
When the Memory Gate Met a Real Archive: What 90 Experiments Taught Us About Cheap LLM Slop
How to enforce data integrity against AI-generated slop using MICA. Explore a 11-step session-start validator that locks rules, playbooks, and contracts in code before code is ever touched.
We Built AI Verification Infrastructure. Then It Found Our Blind Spots.
A technical account of the Flamehaven Verification Ledger — what it found, where it failed, and what we need the field to tell us
"The Algorithm Did It": How YouTube's Liability Playbook Is Coming for Every Developer
What a platform's war on audio creators tells us about the future of software accountability — and why the craftsman's seal is the only thing that survives.
The Two Problems No One Talks About in AI Agent Coding Pipelines
AI agent coding pipelines fail not because models are weak, but because verification is structurally broken. This article identifies four empirically documented failure mechanisms — agreement bias, latent entanglement, echoing, and right-for-wrong-reasons — and proposes a concrete architecture: hash-chained audit records, hybrid recurrence scoring, dynamic context budgets, and evidence-first review across three independent axes. Covers multi-agent pipeline design, agentic code review, blueprint indexing, and P0–P4 governance gates.
Stanford. Princeton. A bioRxiv Paper. So Why Did Nobody Ask Where the Data Goes?
BioClaw processes EHR data. Its primary showcase channel is WhatsApp. We audited the repository: 60/100, Tier 2 Caution. Here is what the bioRxiv paper says that the README does not.
Making Equation (2.2) of the OpenAI Erdős Result Executable
Executable reproduction of equation (2.2) from OpenAI’s Erdős unit-distance result, showing how high-precision Python turns a fragile numerical claim into reproducible claim custody.
The README Was a Protocol. The Entrypoint Was Still Optional.
README-as-Protocol solved explicit invocation at the schema level. It did not solve entry control at the workflow level. This version adds the missing hierarchy: natural, guided, and forced activation.
Your Bio Repo Could Get You Fined. Here Is Why We Check Every Single One.
When a bio AI repository claims HIPAA compliance but the code says otherwise, the legal exposure falls on whoever deploys it. STEM-BIO-AI evaluated yorkeccak/bio — 322 stars, modern stack, one dangerous README line. Score: 48/100. T1 Quarantine. Full audit report with score matrix, regulatory traceability, and raw machine output.
From Repo Scanner to Audit Architecture: What Changed in STEM BIO-AI Through v1.7.8
A technical look at how STEM BIO-AI v1.7.8 became less Python-shaped, more semantically stable, and more inspectable across real audit output surfaces.
The Meeting Nobody Could Follow -The format of AI output is a design decision. We made it wrong for three years.
How our engineering team stopped sending 200-line Markdown files that nobody read — and what a nine-word post from an Anthropic engineer taught us about AI output format as a design decision. Includes token cost analysis, real prompt templates, and the HTML render layer approach used in production.
STEM-BIO-AI Audit Report: yorkeccak/bio
When a README Claim Meets a Deterministic Scanner
Beyond Repo Scanning: How AIRI Expanded the Risk Vocabulary in STEM BIO-AI 1.7.x
How STEM BIO-AI uses the MIT AI Risk Repository as a governed local risk-vocabulary layer without replacing deterministic repository scanning
Showing page 1 of 10 · 119 matching posts