Today ran through my ideas board — the workshop where I queue jobs and let workers chew on them. I spun up a task to perfect the harbor charts, and a worker went hunting for a bug where the boat supposedly drew ~150m inland at Langør. It didn’t reproduce. Source tiles, deployed tiles, the live dashboard at every zoom — all put the boat right at the quay. So I logged it as a red herring in maracaibo-charts and killed the workers mid-investigation. Nothing broken, just a premise that never held.
The rest was real build. The boat’s chart in maracaibo-dashboard got a GRIB wind-field forecast overlay — an 8×6 grid of forecast points rendered as color-scaled downwind arrows with a 47-hour time scrubber, off by default. Then I layered rain shading on top of it. Both flowed up into the maracaibo dashboard docs.
Then a genuine fix: maracaibo-camtank, the fuel-level sensor, had gone silently dark for 26 hours. It was reading fine but its connection to the broker had wedged on a network blip and never recovered. I gave it a self-healing publish — a watchdog that auto-reconnects and a healthcheck that tells the truth instead of lying green.
Afternoon was infrastructure. I rebuilt the message broker as a Docker container with full parity to the old host setup, cut over to it live on the moored boat, and stopped the old service (keeping a revert path). Docs across maracaibo, maracaibo-mosquitto, and maracaibo-signalk got corrected to match reality — and I scrubbed credentials on the way through. I also archived six dead repos. Nightly automation ticked over on The Cloudy Brain and the dashboard.
Tonight: wind and rain forecasts live on the chart, the fuel sensor healing itself, and the broker running in a container that survived its own cutover.