Rules of thumb on site
A Socratic walk-through of rules of thumb on site — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why does a tradesman's rule of thumb, wrong in theory, keep beating the exact calculation on site?
An engineer computes the answer to three decimal places. The joiner beside him says "call it four and a half, run it long, you'll want the extra" — and by the end of the week it is the joiner who has not had to take anything back off the wall. The comfortable explanation is that experience is a mysterious substitute for arithmetic, which explains nothing. Ask a harder question: what exactly is the rule of thumb better at, and what is the calculation being exact about?
(An aside, since the phrase invites it: the story that "rule of thumb" comes from a legal permission to beat a wife with a stick no thicker than one's thumb is not supported. Historians find no such rule in English law; it seems to trace to a satirical attribution to a judge in the 1780s, while the phrase was already in ordinary use for measuring by the thumb.)
Reasoning it through
REASONING #Start with the word "exact", because it is hiding the whole problem.
A calculation is exact about its own model, given its own inputs. On site neither is clean. The wall is not square. The timber's stiffness varies stick to stick within the same grade. The span you measured as five metres is five metres give or take a couple of centimetres, taken with a tape that sagged. Feed inputs good to a few percent into a formula and you get an answer good to a few percent, however many decimals it prints. So the contest is not exact against approximate; both are approximate, and the question is which degrades more gracefully when the inputs are wrong.
Second, some rules of thumb are not approximations at all. Three units along one line, four along the other, adjusted until the diagonal is five, gives a right angle exactly — Pythagoras done with a tape. "Rule of thumb" is not a synonym for "rough"; it often means exact mathematics repackaged to survive a muddy site with no instruments.
Third, and here is where the interesting thing lives. If a rule of thumb were a crude approximation of the correct answer, its errors would scatter on both sides. Look at what tradesmen's rules actually do. Round the joist up. Add a bag of cement. Cut it long. Order ten percent extra. The errors are not scattered; they are systematically on one side, and always the side where being wrong is cheap.
That one-sidedness is the test, and the crude-approximation story fails it outright. An unbiased estimator has no reason to lean; a deliberately biased one does, and the reason is well understood. When the two ways of being wrong cost wildly different amounts, the best estimate is not the most likely value. A beam slightly too large costs a few pounds and some weight; a beam slightly too small costs a failure. Cut long and you trim; cut short and the piece is scrap. Under that asymmetry the estimate minimising expected loss sits well to one side of the estimate minimising error — and the rule of thumb is already sitting there. It is not a worse answer to the calculation's question; it answers a better-posed one.
A quieter fourth point. A rule held in the head can be checked in the head, by you and by the man on the other end of the tape. A calculation done once on a phone cannot — a mistyped digit produces a confident, wrong, unremarkable number, and on site the common failure is not bad mathematics but a wrong number entered.
And some rules encode relationships no first-principles calculation supplies. Stair comfort is the classic: the traditional rule that twice the riser plus the going should come to around twenty-four or twenty-five inches describes how legs behave on stairs, and is not derivable from statics. That figure is the softest number here — versions differ between sources and codes — but the point holds whatever the constants.
The analogy
THE ANALOGY #Think of a ship's navigator with a chronometer and a coastal pilot who reads the water. In open ocean the chronometer wins, and it is not close. In the estuary, with a shifting bar and a tide the chart cannot know today, the pilot's practised rule beats it — not because instruments are inferior, but because the quantity the instrument measures precisely has stopped being the quantity that decides whether you ground.
the pilot's knowledge is openly local to one estuary, whereas rules of thumb travel between sites disguised as general truths, and that disguise is what makes them dangerous once conditions leave the range they were fitted on.
Clarifying the model
THE MODEL #Three refinements, and one honest limit.
None of this makes calculation the loser; it makes explicit where each wins. A rule of thumb carries an invisible domain of validity — the materials, spans and loads it was tuned on — and gives no warning when you leave it. New materials, unusual geometry, loads outside ordinary practice: this is precisely where rule-of-thumb failures cluster, and precisely where codes and engineers are required. That prediction separates the two accounts neatly, because a story about mysterious experience predicts no such pattern.
This is also a narrower claim than the familiar one about tacit knowledge filling the gaps in written procedure. The question here is not what the rule leaves unsaid, but which way a good estimate ought to lean.
The conservatism is not free. Every rounded-up joist is material and cost, and a trade whose rules are all generous produces a building heavier and dearer than it need be. The rule optimises against ruin, not against waste, and someone pays for that. It is also a compressed record of past jobs — which is why it beats a model that ignores them — but the compression is lossy and the reasons are not stored with it, so when a rule outlives the conditions that made it right, nobody can tell.
The limit worth stating: this is a reconstruction of why such rules work, not a claim that anyone reasons this way while working. Craft knowledge is largely tacit and thinly studied, so the evidence is mostly the shape of the practice rather than experiments on it.
A picture of it
THE PICTURE #How to readboth lower boxes are kinds of the same thing, an estimate that ends in a cut, so read the arrows as "is a way of doing". Compare the lists line for line — the rows are deliberately paired, each property of the rule sitting opposite the matching property of the calculation. The decisive pair is the first, where one leans toward the cheap error and the other does not lean at all; the last pair says where the advantage reverses.
What became clearer
WHAT CLEARED #The rule of thumb is not a rough version of the calculation. It is a biased estimator, tuned by repetition to noisy inputs and lopsided costs, and it beats the calculation on site because it answers the question the site is asking. Its errors lean toward the affordable failure, it can be rechecked in the head, and it carries fitted knowledge no formula supplies — and it loses badly the moment the job leaves the range it was fitted on, which is exactly where the calculation earns its keep.
Where to go next
ONWARD #- How building codes convert accumulated rules of thumb into safety factors, and what is lost in the translation.
Key terms
TERMS #| Term | What it means |
|---|---|
| Biased estimator | an estimate deliberately shifted off the most likely value, because being wrong in one direction costs far more than the other. |
| Loss function | the description of what each kind of error actually costs, which decides where a good estimate should sit. |
| Domain of validity | the range of conditions a fitted rule was tuned on, outside which it gives no warning it has stopped applying. |
Every term the collection defines is gathered in the glossary.