Operations¶
| Page | Covers |
|---|---|
| Deployment | version policy, the firmware pipeline, and the hardware acceptance tests |
| On-device pipeline | the frame budget, the detection gate, why not an ESP32-S2, and what the web interface replaced |
| Documentation pipeline | the docs-only build, local preview, validation and automated GitHub Pages deployment |
| Handoff protocol | the rules any contributor — human or model — works under |
Two build modes, independent by design¶
Normal build Docs-only build
└── ./plxy.sh build └── ./plxy.sh docs-build
ESP-IDF, PowerShell, 16 MB mkdocs-material and nothing else
firmware, a board attached seconds, no toolchain, no hardware
Neither triggers the other. A documentation change never queues behind a
firmware build, and a firmware change never republishes the site unless it
touched docs/. The GitHub Actions paths filters enforce the same split.
The record¶
Three files at the repository root are the running history, and they are maintained by hand rather than generated:
| File | Is |
|---|---|
PROJECT_STATE.md |
what is true now, including what has been tried and failed |
ROADMAP.md |
what is planned |
CHANGELOG.md |
what changed, in order |
Read them before major changes. The handoff protocol says the same thing in rule form.
Before a thesis experiment¶
docs/HARDWARE_SETUP.md §10 lists the measurements to record before the
experiments start — board MAC, sensor part, measured fps, PSRAM free, the
component versions in dependencies.lock. Numbers taken afterwards cannot be
attributed to a configuration.