I don't have much time for the angels-dancing-on-a-pinhead style of philosophical argument: at some stage we simply have to make a set of assumptions and work with them until it becomes clear that they need revisiting

Ah, I do agree! Yes, we must at some point refuse to discuss the point of point, for we could do so indefinitely. I would not necessarily agree to put the 'objective' prior to the 'subjective' as you did. Let me add my ramblings on this topic to yours:

'Science', 'Religion', etc.

• 'Science' is a 'better' 'index' of 'reality' than 'Religion' or 'Astrology'.
• 'reality' == 'material substrate' of 'existence'
• 'scientific reasoning' -> better understanding of material consequences of actions than other methods

Science will never tell us what to do. Science is amoral. The horrific experiments of the Nazi's were plenty scientific, we reject them for different, more important reasons.

'Objective' 'reality'? Abstractions all the way down!

• We agree on a shared bit of 'reality'. 'Reality' is defined in the social field, into which we are born.
• The sharing of 'reality' is always a violent action. It proceeds to subjugate, to impose an order, and a particular one at that.
• It is useful, we do it. It was already done. There is no other way to proceed
• One must at some point accomplish a leap of faith. Establish, arbitrarily, a point of reference, a ground for our circuit. Except this was always already done, or we would have no meaning, and therefore not be able to discuss it.

The subjective itself is part of the imposed order, the objective/subjective division as well. When/where was there not such a division??? Some examples from kcurie?? Emotion in ancient Greek texts? External events imposed on the individuals. "Rage descended upon him", is the "descended upon" more than just figurative?

Stories as modelling, modelling as stories

• Models are 'tools to think with', not 'information', or 'facts'.
• Ad hoc models are plenty useful, and may be as good as we can do in some cases.
• One must maintain at least a minimal narrative of context to assert 'meaning'.
• We proceed by abstractions, not by reductionism. Reducing everything to the 'smallest parts', or the most 'general description' is not the most useful way to proceed. It is in fact quite useless!

Can we derive 'how' a transistor 'works' from quantum-crystalline properties of silicon and dopants, Maxwell's equations and some thermodynamics? Indeed, we can. (With the caveat that one cannot obtain analytic solutions to the resulting differential equations, and a few other things...) We do, as a tool for learning the abstraction hierarchy of our knowledge, but day to day, not so much. Even for this derivation, we approximate, the exercise is illustrative, not practical. It enriches our thinking of electronic devises in a difficult to quantify way. Direct, practical advantages? Well, the engineer should know, our knowledge is but a model. As such, there are explicit limits of operation beyond which the model yields wrong values. We must know the hierarchy of models to remember the limits, to not foolishly predict 1000V output from an op-amp with a 5V source. Yet people make theses mistakes, and not rarely. Never confuse the model with 'what really is', and yes, all we have are models, all the way down, and they have limits. parameterise

To design a circuit:

1. Transistors are on/off switches. "This one goes on, current through there, voltage up here, that one turns off" etc.
2. We calculate some things, an operation point of the non-linear approximate equations describing our transistors quite well, as long as they are operated in not too extreme conditions. Then we linearise about this point. Linearise, because we have no useful abstractions for thinking about the behaviour of non-linear systems.
3. Simulate with a computer, use the computational power of our PCs to refine the results. Here, we include those pesky 'parasitics' that where previously ignored. All those capacitances and inductances and resistances that are always there but to complex to be usefully considered manually. Sometimes the equations help to...
4. But wait!! All this assumes a lumped element circuit... Maybe our frequency is high, we are approaching that limit where a signal wavelength becomes comparable to the physical component size... More sophisticated tools for this, 3-d continuum modelling. We put in the info, the computer spits out results, input to output relations, dynamic effects... Do we even know how to represent these to ourselves? To categorise, index, organise results in to something with which we can usably think? This is the challenge, a numerical result is very little without.
5. ...

Even at the lowest, most basic level, there is but abstraction. A model. Equations describing motion, processes, and events.

Where do we draw the line then? When is something 'objective fact'. There is such a point, to be sure, when 'hard' 'reality' asserts itself... For those who would doubt, I encourage you to challenge gravity's effects on you... Then come visit me, and we shall submit your hypothesis to experimental verification. I know the site, I see it from my balcony:

But you don't get a parachute...

'Hard' 'reality'. It is like 'porn'. I may not be able to define it, but I know it when I 'see' it, or fall to my death.