<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Dallas Crilley — Architecture writing</title><description>Notes on systems work, architecture decisions, and the tradeoffs behind the code.</description><link>https://dallascrilley.com/</link><language>en-us</language><item><title>Adapting Cloudflare’s $1 Review Factory for RevOps</title><link>https://dallascrilley.com/writing/revops-software-factory/</link><guid isPermaLink="true">https://dallascrilley.com/writing/revops-software-factory/</guid><description>Cloudflare’s code-review factory, adapted to revenue data: coordinator, specialist fusion, risk-scaled compute, and the resilience work needed to make it usable as a demo.</description><category>Agent systems · RevOps architecture</category></item><item><title>Cutting Integration Code by 40× — and Knowing When to Stop</title><link>https://dallascrilley.com/writing/throughline-connectors/</link><guid isPermaLink="true">https://dallascrilley.com/writing/throughline-connectors/</guid><description>A four-method connector contract that cut a new integration from about 2,000 lines to about 50 at the entry point, plus the restraint it took to keep the interface small.</description><category>Throughline · Connector design</category></item><item><title>When the API Returns 500, Did the Charge Post or Not?</title><link>https://dallascrilley.com/writing/meter-idempotent-sync/</link><guid isPermaLink="true">https://dallascrilley.com/writing/meter-idempotent-sync/</guid><description>Preview-before-sync, ghost-addition detection, and DB-level locking in a production MSP billing platform.</description><category>Meter · Billing automation</category></item><item><title>Your Own Work Is Invisible to You</title><link>https://dallascrilley.com/writing/adversarial-review-panels/</link><guid isPermaLink="true">https://dallascrilley.com/writing/adversarial-review-panels/</guid><description>A multi-agent review panel with fresh-context personas, an anchored 0–10 rubric, and a recurrence check that turns the last run into the test suite for the fixes.</description><category>Agent systems · Review architecture</category></item><item><title>The Gate Has to Stand Without the LLM</title><link>https://dallascrilley.com/writing/cohost-eval-gate/</link><guid isPermaLink="true">https://dallascrilley.com/writing/cohost-eval-gate/</guid><description>A quality gate for AI-assisted publishing where deterministic metrics block, LLM evals advise, and a human owns the irreversible call.</description><category>CoHost · Eval architecture</category></item></channel></rss>