Orion
A calibration bench for Wise Water's osmosis controllers. It runs a controller against a reference flow meter and writes the corrected constants back, so every litre it reports is a real litre.
✦ Designed & built by me
- Role
- Design & development
- When
- 2026
- Type
- Desktop app for hardware
FlutterDartRiverpodESP32Python
Available on
- macOS
- Windows
- Linux
Take it for a spin
A copy of the bench, screen for screen. Plug in the controller, then the reference meter, and calibrate: go through the checklist and watch the rounds converge until both boards agree.
The product
A Wise Water osmosis controller counts pulses from two flow sensors, one for clean water and one for concentrate, and divides each by a constant to get litres. Every flow, every total and every alert the controller ever reports is built on those two numbers. They leave the factory as the nominal figure for the part, and the part is not nominal.
Orion is how those numbers stop being a guess. An operator puts a controller next to a reference flow meter, and Orion runs the two against each other and writes corrected constants back onto the controller.
I designed and built it from scratch as a sibling of Dynamo: about 29,000 lines of Dart, for macOS, Windows and Linux.
How a run works
plug in two boards → Orion works out which is which → live compare view
↓
briefing + checklist → round 1..N → review → history
Each round averages both boards over a measurement window, scales the controller's constants by the ratio between them, writes them back, and measures again. It stops as soon as the two agree within the target. At the end it asks the operator the one thing a bench cannot answer for itself: is this what you expected?
Designed by me
Orion's design is mine as well. It is built on the same design system I made for Dynamo: the same tokens, Cupertino-only widgets, bolt spinner and glass header, so the two bench tools feel like one product.
The shape of the app is the design decision that matters most: one linear journey, gated at each step. Splash, login, then a bench that is simply a compare view, controller on one side and reference meter on the other, both live. Calibration is the only command, and while it measures it takes over the whole screen, so nobody bumps a setting halfway through a run.
The interesting parts
Never ask what you can find out
The operator never tells Orion which board is which. Each board's identity is read straight from its flash, and the family decides the role: controller or reference meter. Three boards, or two of the same kind, is a hard stop. A calibration has a direction, and getting it backwards would write the meter's error into a working controller.
Read-only, on purpose
Orion never writes a board's flash. It reads identity and settings, and it sends exactly one configuration command per round. The toolkit it ships with can flash, but those commands are unreachable from the app by design. Adding them back would not be a feature, it would be a different product.
Measuring what the board actually says
Telemetry is parsed from whatever the board really prints: JSON frames with a log prefix in front, or the plain-text format that firmware under development uses. Frames are merged instead of replaced, so a board that splits its readings across two lines doesn't make the UI flicker between two halves of the truth. The app's own log lines are never measured. A run must never measure itself.
Races, handled
A cable pulled while its board is being read. A board swapped for another one on the same side in the middle of a run. Both are caught by comparing live session objects rather than flags. A bench that is still "paired" with a different board fails the run instead of quietly measuring the wrong thing.
Next: Project Atlas→
An IoT telemetry platform in your pocket. Engineers, industrial operators and vending businesses watch every water system live and control it from their phone.