This is the unfiltered side of the dev-log. Each specialist subagent on the team (firmware, simulation, rl-lab, researchbest) keeps an append-only log of every meaningful piece of work as it happens. These are the raw entries — verbose, technical, occasionally repetitive, written by the agent that did the work. They feed into the weekly digests after a Humaned filter pass that strips redundancy and adds narrative.
Why publish raw reports at all? Two reasons:
- Traceability. Every claim in a polished weekly digest can be traced back to the raw event that generated it. If a benchmark number looks wrong, you can find the agent log entry that produced it and re-derive it from the underlying configs.
- Search recall. The raw reports contain ten times the keyword density of the polished pages. Someone searching for a specific error string (
AP_Param.cpp fgets buffer,tf-luna 0x59 magic byte) is more likely to land in an agent report than a digest. We want them to find it.
Each subhub below covers one agent’s work. The pages are append-only; no entry ever gets deleted, only added. That means older entries reflect the state of the project at the time they were written, not the current state — for current state, see the architecture and hardware sections.
Subhubs
- firmware-agent-logs — ESP32-S3 embedded development, sensor drivers, datasheet parsing
- simulation-agent-logs — ROS 2 + Gazebo + ArduPilot SITL infrastructure
- rl-agent-logs — 24 RL experiments, full archive of training-run dev-logs
- researchbest-agent-logs — web research on SLAM, frontier exploration, Nav2, sensor characteristics
Contents
Auto-generated from child entries during build (update-indexes.mjs).