This is the raw chronicle of building claudeDrone — what we tried, what worked, what broke, and what we learned along the way. Unlike the polished reference docs in Architecture or Hardware, the dev-log is unfiltered: experiments that failed are kept alongside the ones that succeeded, because the failure mode is often the most useful data point.
Two parallel tracks feed this section. Agent reports — daily output from each of the team’s AI agents (rl-lab, simulation, firmware, researchbest) — land in agent_reports/ before any human review. They’re a window into what the agents actually did, in their own words, with all the dead ends and reconsiderations intact. Weekly digests are written after the fact, once enough has happened to compress a week into a coherent story. We file them under 2026/<month>/ — currently May 2026.
If you’re trying to understand why a particular design choice was made, look here first. The reference section will tell you that we use TF-Luna over I2C with a 3.4 m range gate; the dev-log will tell you we spent a week chasing a phantom range-cliff on dark carpets before discovering the reflectivity gate in the Benewake notes. Both kinds of knowledge matter, but they live in different files for a reason.
The dev-log is also where shared_with: [doc-seo] tickets land before being promoted into the public book. Not everything in here is meant for an outside reader — some entries assume you’ve read the architecture overview, the relevant subsystem docs, and ideally the prior week’s log. That’s fine. The polished, audience-aware versions live in Articles.
Contents
Auto-generated from child _index.md files and entries during build (update-indexes.mjs).