The board powered on. The current draw looked reasonable. Nothing on it worked. That's a worse starting point than a board that won't power on at all, because 'nothing obviously wrong' rules out the fast diagnoses (a shorted rail, a reversed connector) and leaves you with everything else.
Start at the source, not the symptom
It's tempting to start probing at the microcontroller, since that's where the visible failure is. Start at the regulator instead. Measure input voltage, then output voltage, then move downstream one net at a time. On this board, the 3.3V rail measured 3.3V right at the regulator's output pin, and 1.9V forty millimeters away, at the MCU's supply pin.
That's not a regulator problem. That's a trace problem, or a connection problem between those two points.
The actual fault
A via between layers had a hairline crack, visible only under magnification, invisible to continuity testing at rest because the two halves were still touching under normal conditions, just barely. Load from the MCU pulled enough current through the marginal connection to drop the voltage across it. A multimeter in continuity mode showed a closed circuit. A multimeter measuring voltage under load showed the truth.
The lesson wasn't about the crack. That was a fabrication defect, and it happens. The lesson was the order of operations: measure at the source, measure at the load, and if they disagree, you've found your fault's address even before you know its cause.