Why we use a ReAct loop instead of rules for idle resource detection
Most cloud cost tools work off rules. If CPU stays below 5% for seven days, flag the instance as idle. If a volume has zero read operations for thirty days, flag it as unused. These rules are simple, fast, and easy to audit. We tried this approach first when building Argus. It kept getting things wrong.
This post explains what the problem actually is, what a ReAct loop is, and the specific tradeoffs we accepted by going down this path.