@User also consider that if a circuit works well with some .nodeset conditions there are probably more equilibrium states for the circuit. Bandgap references typically need a startup circuit that removes the unwanted equilibrium state. These start up circuits must be always active ("static" startup) , whenever the bandgap leaves the "on" state for whatever reason (vcc/gnd noise, coupling, whatever) startup circuit must kick in and pull the circuit back to the "on" state. Historically implementations with a "dynamic" startup, like something pushing the bandgap high on poweron and then doing nothing have produced references that in some misterious situations turn themself off. This should never happen. When a good "static" startup circuitry is added normally the circuit does no more need any .nodeset to converge to the right (and now unique) solution.