---
title: "Systems & Craft Track — Superintelligence Buzz"
description: "Inference mechanics, KV-cache geometry, Mixture-of-Experts routing, and self-improving task/verification harnesses."
canonical: "https://superintelligence.buzz/tracks/craft"
---

# Systems & Craft

> Inference mechanics, KV-cache geometry, Mixture-of-Experts routing, and self-improving task/verification harnesses.

## Curated Dispatches

### [JEV-as-a-Judge: Accept When Confident, Escalate When Unsure](https://superintelligence.buzz/essays/jev-as-a-judge-system-one-evaluation)

*By K. R. Sterling · 2026-10-16 · 8 min read*

Routing every production evaluation through a heavyweight autoregressive model burns cluster budgets on redundant monologues. Here is how typed System One evaluators and confidence-routed cascades restore discipline to the inference critical path.

### [The Hidden Geometry of Attention: What Happens When Context Gets Too Long](https://superintelligence.buzz/essays/kv-cache-quiet-truth)

*By Marcus Chen · 2026-10-10 · 5 min read*

Why models boasting two-million-token context windows behave very differently in high-concurrency production than they do in isolated single-user demonstrations.

### [The Architecture Beneath the Hype: Why Sparse Routing and Memory Arithmetic Won the Day](https://superintelligence.buzz/essays/post-scaling-craft)

*By K. R. Sterling · 2026-10-13 · 7 min read*

The era of throwing $100M compute clusters at monolithic dense models has hit an economic brick wall. Here is how modern Mixture-of-Experts and speculative decoding actually work under the hood.

### [The Co-Evolution of Task and Harness: Why Lean Execution Models Are Leaving the Chat Window Behind](https://superintelligence.buzz/essays/the-co-evolution-of-task-and-harness)

*By K. R. Sterling · 2026-10-21 · 9 min read*

For years, foundation models were optimized to produce pleasant conversational prose inside browser windows. The emergence of closed-loop self-scaffolding systems like Ornith-1.5 reveals a different trajectory: lean parameter footprints tuned directly for terminal harnesses, sparse tool use, and verifiable assertions.