Using agents to keep sensor data honest
This is a working demonstration of an agent that cross-checks several public sources to watch for drift. For one spot off Monterey it pulls a real buoy thermometer and two satellite maps, lines them up for the same day, and shows exactly how far they agree — so you can watch the readings stay in agreement over time, or catch the moment one starts to drift.
The point isn't the temperature. It's the cross-check: the same agent fetches, quality-checks, aligns, and compares every source the same way every time, and it is careful to flag when sources that look like independent confirmations actually share inputs. Three numbers that nearly match is reassuring only if they were truly independent — and the agent tells you when they weren't.
These are comparability labels: they describe whether the sources were available and lined up for the same day. They do not validate scientific truth, and are not for navigation, safety, regulatory, or operational decisions.
The three signals
A real thermometer on a NOAA buoy, sitting about 1.5 metres underwater off Monterey. This is the only source that physically touches the ocean — everything else is an estimate.
A daily map that blends many satellite passes into one gap-filled picture of the ocean on a ~1 km grid. It is a computed estimate, not a sensor. The settled version also folds in buoy and ship measurements, so close agreement with the buoy is not fully independent; the newest two or three days are an interim version built from satellites alone. The card below names which one you are seeing.
A coarser daily map on a ~25 km grid, also blending satellites and in-water readings. Smoother and broader than MUR, and likewise not an independent check on the buoy.
How it uses them
- Pin one exact question. The water temperature at one point, on the newest day that all three sources have finished reporting.
- Take each source's own value. No averaging, no voting, no picking a winner — just the three numbers as reported.
- Line them up and show every difference. Depth, location, timing, and how each one was made — so a close match and a mismatch both mean something.
- Flag what isn't independent. The two maps partly ingest buoy-type data, so their agreement with the buoy is not three separate confirmations.
No reading loaded yet.