Engineering

What a calculation has to show before it can be checked

Published August 26, 2026 · SynthoCore Automation Inc.

A reviewer looking at a calculation does one of two things. They check it, or they trust it. Checking means reproducing the result from what is on the page — following the inputs through the formulas and arriving at the same number. Trusting means accepting the number because the person who wrote it is competent and the answer looks about right. Both happen every day. Only one of them catches an error before it reaches steel.

Whether a reviewer can check your work is not a measure of how careful the reviewer is. It is a property of the document you handed them. A calculation is checkable when a competent engineer can reconstruct every number in it without asking you a single question. If they have to ask — where did this factor come from, what load is this, is that the yield or the ultimate — the calculation is not checkable yet, however correct it happens to be.

The reconstruction test

One test settles it. Hand the calculation to someone who was not in the room, give them a spreadsheet and a pencil, and see whether they can rebuild the result using only what is written down. Not the intent, not the standard you had open, not the assumption you made at your desk — only the marks on the page.

If they can, the calculation is checkable. If at any line they would have to reach for something that is not there, that line is where trust quietly takes over from checking. The reviewer either stops and asks, or — more often, under time pressure — nods and moves on. The second outcome is the dangerous one, because it looks exactly like review and is not. Everything below is a way of passing that test.

A calculation that can only be trusted

Here is a result written the way results get written when someone is moving fast:

Plate thickness required: 12 mm (governs)

That is an assertion. It might be right. A reviewer cannot tell, because there is nothing to follow. Where did 12 come from? What load, what material, and what criterion decided that this thickness “governs” — governs over what? To check this line, the reviewer has to reconstruct the whole calculation from scratch, which means doing the work again, which is not review. So they trust it.

Now a version that feels rigorous but is not:

t_req = P / (φ · Fy · w) = 0.95 mm Use 6 mm (min plate). OK.

This shows a formula and an answer, and it still fails the test. The reviewer can see the shape of the calculation but cannot reproduce the number. What is P? What is φ, and is it the value actually used or a different one? Where does 6 mm come from — a fabrication minimum, a stock size, a standard? “OK” against what threshold? Each of those is a question the reviewer must bring back to you. The calculation still runs on trust; it has only dressed better.

The same calculation, made checkable

Nothing about the engineering changes. What changes is that every number now traces to something on the page:

Given P = 45.0 kN applied tension (load schedule LS-03, case 2) w = 150 mm connection plate width Fy = 350 MPa yield strength (CSA G40.21, grade 350W) φ = 0.90 resistance factor (steel in yielding) Required cross-sectional area A_req = P / (φ · Fy) = 45.0e3 N / (0.90 × 350 N/mm²) = 142.9 mm² Required thickness t_req = A_req / w = 142.9 mm² / 150 mm = 0.95 mm Governing check t_min = 6 mm (fabrication minimum for this plate) 6 mm > 0.95 mm → thickness is set by fabrication, not strength. OK.

A reviewer can now take this apart line by line with a calculator. 45,000 divided by (0.90 × 350) is 142.9. Divided by 150 is 0.95. The units carry through — newtons over newtons per square millimetre leaves square millimetres, divided by millimetres leaves millimetres. The word “governs,” which was doing unexamined work in the first version, is now a stated comparison between two named quantities. And the reviewer learns something the first two versions hid: the strength demand here is trivial, and the thickness is really a fabrication decision. That is a finding. It was invisible until the calculation was made checkable.

What actually changed

Read the two side by side and the difference reduces to a short list. In the checkable version:

  • every input has both a value and a source, so a wrong input is a visible wrong input rather than a mystery;
  • every formula appears in symbols before it is used, so the reviewer checks the method separately from the arithmetic;
  • every substitution is written out, so reproducing the number is reading, not re-deriving;
  • units travel with the numbers and cancel on the page;
  • intermediate results are named — A_req — and then reused, rather than silently recomputed inside the next line where an error would hide;
  • the acceptance criterion and its threshold are stated, so “OK” means something a reviewer can confirm.

None of that is exotic. It is bookkeeping. But it is the bookkeeping that converts a reviewer from a truster into a checker.

Arithmetic is not adequacy

One more distinction, because it catches people who have already tightened up their traces. A reviewer checks two different things, and a calculation can pass the first while giving them no way to check the second.

The first is reproducibility: does the math compute? Given the inputs and the formulas, does the stated number come out? The worked example above is fully reproducible.

The second is adequacy: is the result acceptable? A number is only adequate relative to a criterion — a limit, a code check, a threshold. A calculation that produces “0.95 mm” and stops has told the reviewer what, but not whether. It becomes checkable for adequacy only when it also states what the result is judged against and shows the comparison. The line 6 mm > 0.95 mm → OK is that second check made visible. Leave it out and you have handed the reviewer a reproducible number they still have to trust is good enough.

The tells

You can spot a calculation drifting back toward trust without reworking it. The tells are consistent:

  • a constant with no derivation and no source — a factor of 1.6 that arrives from nowhere;
  • a number with no unit, which cannot be checked because it cannot be shown wrong;
  • a result that appears without the inputs that produced it;
  • the word “conservative” with no statement of conservative relative to what;
  • a pass or a fail with no threshold beside it.

Each one is a spot where the reader has to supply something you know and they do not. Each one turns a reviewer back into a truster.

Why it is worth the extra lines

Writing calculations this way costs a little more time at the desk and returns it at every review afterward. The reviewer stops guessing at intent. The errors that do exist surface as specific wrong lines instead of a vague unease that something is off. And the record you leave behind can be checked by someone who was never in the room — which, given enough time, is everyone: the next engineer, the approver, the person who inherits the file years later. A calculation that can only be trusted depends on you being available and remembering. A calculation that can be checked stands on its own.

If this is the kind of rigour you want in how your engineering work is produced and reviewed, get in touch.