When to calculate and when to just cut metal

Deciding whether analysis or a prototype answers the question faster.

Calculation earns its keep when a mistake would be expensive or irreversible to discover any other way, and cutting metal earns its keep when the answer is cheap to find out directly and correcting it afterwards costs almost nothing, which makes the real decision less about which method is more rigorous and more about which one gets to a trustworthy answer fastest for this particular part.

The short version

Every design question can, in principle, be settled by calculation, a prototype, or some mixture of the two, and treating one of those as always the serious approach and the other as always the shortcut misses the actual trade being made. A calculation is fast to run once the model is right and can be wrong in ways that are invisible until the part is built, since it only ever answers the question it was actually set up to answer and stays blind to everything left out of that setup, while a physical trial is slower to arrange and slower to iterate but tells the truth about everything at once, including the things nobody thought to model in the first place. Both routes can mislead in their own particular way, a calculation through a wrong assumption carried confidently to a wrong conclusion, a prototype through a lucky first attempt mistaken for a reliable one, and neither method is inherently the careful one just because it looks more like arithmetic or more like carpentry.

A useful habit is treating both as forms of the same underlying question, which is simply how a person can find out, as cheaply and as reliably as possible, whether a part will actually do what it is meant to do. Framed that way, the choice of method stops being about which discipline feels more rigorous and becomes a genuinely practical comparison between two different routes to the same piece of knowledge, one paid for in thinking time and one paid for in bench time.

The sauce-and-soufflé comparison

A cook adjusting a pot of soup tastes it, decides it needs more salt, adds a pinch, and tastes again, a loop that costs seconds and forgives being wrong by a wide margin, because a soup slightly over-salted this minute is still a soup that can be thinned or balanced the next. The same cook baking a soufflé does not adjust by taste part-way through, since the batter has already gone into the oven by the time any mistake would show itself, and the sensible move is to weigh the ingredients precisely beforehand rather than discover the error only once the door is opened. A bracket that can be filed a little thinner after a quick test fit behaves like the soup, forgiving and cheap to correct by trial, while a part that gets sealed inside a welded assembly or committed to an expensive casting behaves like the soufflé, unforgiving enough that a calculation done properly beforehand is the only trial that actually gets a fair chance to be right.

Why the cost of being wrong decides more than the cost of finding out

The instinct to reach for a calculation whenever a problem looks technical, and for a prototype whenever it looks fiddly, gets the actual logic backwards more often than it helps, since what should decide the method is how expensive a wrong answer would be to discover late rather than how the problem happens to look on first glance. A rough bracket holding a light sensor in place can simply be cut, tried, and re-cut in twenty minutes if it is wrong, so a calculation that would itself take twenty minutes to set up correctly buys nothing over just cutting the metal and finding out directly. A pressure fitting rated to hold a gas line, by contrast, is exactly the kind of part where a wrong guess discovered by trial is discovered as a failure, at which point the calculation that would have caught it in advance was always the cheaper route, however long it took to work through by hand.

The number that matters here

A prototype that takes twenty minutes to cut and test against one that would need a two-hour calculation to check properly is an easy call in the prototype's favour, and a great many decisions on a lab bench sit exactly there, cheap enough to try that a calculation would only slow the answer down. The calculation starts winning once trial and correction would cost hours rather than minutes, or once a wrong first attempt would be discovered only after it had already been welded, bonded or built into something bigger, at which point the twenty-minute habit that served every small bracket well becomes the exact habit that causes the expensive part to fail.

What follows from this

Vincenti's What Engineers Know and How They Know It makes the case that engineering knowledge grows through this same alternation, a working method tried until it fails informing the calculation that eventually replaces guesswork for that specific class of problem, and a lab bench runs on a compressed version of the same cycle every single day. The practical habit worth keeping is asking, before reaching for either a pencil or a saw, what a wrong answer would actually cost to discover late, since that single question sorts the vast majority of design decisions correctly without needing to weigh how technical or how simple the part happens to look.

Where this stops being true

Neither method is complete on its own for a part whose failure would be genuinely dangerous or expensive, since a calculation checks only the specific failure it was built to check while a single successful prototype checks only the specific conditions it happened to be tested under, and a part that matters enough deserves both, a calculation to rule out the failures that are easy to predict and a prototype to catch the ones that were never modelled at all. Cutting metal first is also a poor habit to fall into for anything produced in quantity, since a design flaw caught after fifty identical parts have already been machined costs fifty times what it would have cost to catch on paper, a multiplier a calculation never has to pay because it exists entirely before the first cut is made. The right balance genuinely does shift with the part, and a lab that trusts either method as a blanket rule, always calculate or always just try it, ends up paying for that rule's blind spot sooner or later.

My key error with this

I mounted the work to a plate without ever calculating whether the plate was stiff enough for what was about to happen to it, because the plate looked substantial, it was metal, and the question felt like the sort of thing you check when something goes wrong rather than beforehand. Under the cutting load it deflected, quietly and progressively rather than in any single obvious moment, so by the time I noticed that the geometry was wandering I had already lost the part and the material in it. Working out the thickness I actually needed took a few minutes with a pencil once I finally did it, which is what makes the whole episode annoying, since the calculation was never expensive, it simply never got done. What replaced the belief is that the decision between calculating and just cutting metal should be made on how much a wrong answer costs rather than on how confident the part looks, and that anything holding work against a cutting force has earned five minutes of arithmetic before it earns any trust.