When nothing lights

Start with the symptom

This is for the newer board, the one with the PWM pad. On the earlier board there is no pad and no PWM pin, and its 12 V goes straight to the display, so skip the first check below and start at the 12 V.

Almost every fault here is one of seven things, and each has one check that settles it. They are in the order to look. Work from the symptom rather than from the wiring, because rewiring a chain that is already right is how a working display becomes a broken one.

Why the pad is first

The 12 V does not reach the display directly. It goes through a switch that turns on only while the PWM pin is high, and the pad beside that pin is what holds it high when nothing else does. The pad ships open. So a module with nothing on PWM is dark, with a perfect supply, perfect wiring and a sketch that has loaded its registers correctly. That is the commonest first report, which is why it comes before anything with a meter.

Two ways to find out, both quick. With everything off, a continuity test across the two pads of the pad, marked SB1: a beep means it is shorted, silence means open. Or, with the 12 V on, touch a wire from PWM to 3V3/5V; a digit that lights at once had an open pad and a floating PWM pin. With the pad open and nothing driving PWM, the pin usually reads close to zero volts against GND, and with it shorted or driven high it reads your logic rail.

Then pick one: solder the pad shut, or drive PWM from a pin. Not both. On a chain, the same rule holds for the whole chain, because PWM is one net across every module.

The two measurements worth taking next

Twelve volts, against your board's ground. Black probe on a GND pin of the microcontroller, red probe on the module's 12V pin. About twelve volts there means both the supply and the shared ground are good. Reading it against the supply's own negative lead instead proves only that the adapter works, which was never in doubt.

Continuity across the join. With everything switched off, a continuity test from the first module's DATAOUT pin to the second module's DATAIN pin. That single link is the one connection in a chain that is not duplicated anywhere, so it is the one whose failure has a symptom of its own: the near module counts and the far one stays dark.

The 3V3/5V pin gets the same test as the 12V one. It feeds only the shift register, so a missing logic rail looks like a dark digit or segments that ignore your sketch, and one reading settles it.

The faults that are not electrical

Two of them are in the sketch, and they have shapes you can recognise.

A flicker while the number changes is the latch being pulsed between bytes rather than once at the end. Move the two digitalWrite(LAT, ...) calls outside the shiftOut loop.

Digits in the wrong order is the array being filled the wrong way round. buf[0] goes out first and therefore travels furthest, so it is the leftmost digit.

Neither of those is worth a meter, and neither is fixed by moving a wire. The wrong bit order in the figure is a third that looks like a fault in the board and is a single word in the sketch.

When it is genuinely the module

Rare, and it has a specific look: one segment that never lights whatever you send, on one module, while everything else behaves. With PWM high, send 0xFF: all seven bars and the point. Any bar still dark is that segment's own line.

Swap the module with another from the box before concluding anything. If the fault moves with the board it is the board; if it stays in the same position on the bench, it is the wiring or the joint feeding it.