One of our own test boards carries a 1.8 V rail built around an AMS1117 adjustable regulator. The schematic note computes the output correctly from the datasheet’s formula — the arithmetic lands on 1.799 V, and every automatic check is happy. The problem is one of the numbers that went into it.
A value that isn’t in the datasheet
The output-voltage equation is the datasheet’s. One term in it is the adjust-pin bias current, and the value written on the schematic isn’t a figure the datasheet gives. Work the same equation with the tolerance ranges the datasheet does specify, and the real rail could fall anywhere from about 1.72 V to 2.11 V.
The bias-current term is roughly a sixth of the output, so it is not a rounding detail — it moves the rail. Light loading pushes an adjustable regulator toward the top of that range, and this rail feeds two external connectors, so the error doesn’t stay on the board: it reaches whatever is plugged in.
Why the checks stayed quiet
Nothing here is wrong in the schematic, the footprint or the netlist. ERC and DRC pass because there is nothing for them to catch — the calculation looks right. The gap is between a number written on a schematic and the datasheet it claims to come from, and no connectivity check reads that.
Probe did. It pulled the resistor values from the BOM, the tolerances and currents from the datasheet, and the load from the schematic, worked the real bound, and cited each figure to its source — including the one that had no source.
A calculation written on a schematic is a claim. It needs the same citation as any other number on the board.