READING 25 · WHERE THE NUMBERS COME FROM · THE AXES THEMSELVES
The Origin of the Allowable
The number in the code looks exact. Its ancestry is not.
§ 01What the code inherited
Open a structural steel code and you will find an allowable stress: a single value, precise to the megapascal, telling you how hard you are permitted to push the material. The number carries the authority of a standards body and the apparent precision of measurement. Neither fully explains where it came from.
The number's most immediate ancestor is the material's yield strength — the stress at which the metal first deforms permanently. That value comes from test coupons pulled to destruction in a laboratory, and the path from raw test results to a published design value involves real statistical work: enough samples to characterise variation, a lower-bound estimate chosen to capture the population's weak tail rather than its average. The coupon tells the truth about that batch; the statistician tells a cautious story about the batches you haven't tested yet.
From that statistically treated yield strength, the allowable is derived by dividing by a safety factor — typically something in the range of 1.5 to 2 for common structural steels, though the exact value is not the point here. The point is that this factor was not calculated from first principles. It was not derived by someone sitting down with a probability distribution of loads and a probability distribution of resistances and solving for an acceptable failure rate. For most of engineering history, it was chosen to match what had worked before, calibrated against structures that stood and failures that didn't recur. It is, in an important sense, empirical caution dressed in the language of calculation.
How the allowable is built up
- Material test datareal tensile tests on real coupons, statistically treated to find a lower-bound characteristic value
- Safety factor (or partial factor in limit-state codes)applied to yield strength or ultimate strength depending on the code framework
- Calibration targetthe acceptable failure probability embedded in a limit-state code, typically back-calculated from the record of older empirical codes, not derived from an independent societal risk decision
What changed with limit-state design
- Older permissible-stress codes: single safety factor, applied to a characteristic strength, not explicitly connected to load variability
- Limit-state codes (Eurocodes, LRFD): separate partial factors for load and for resistance, calibrated jointlymore transparent reasoning, but still anchored to historical performance data
- The calibration target in both cases reflects what has historically worked, not a first-principles derivation of acceptable risk
§ 02The uncomfortable arithmetic
This matters because the allowable stress in a modern code is actually the output of at least three separate judgements layered on top of each other: the choice of which test population defines the material property, the statistical rule that converts scattered results into a single characteristic value, and the safety factor applied on top of that. Each layer introduces a choice, and those choices have histories that are not always visible in the final number.
Some of those histories are tidy. Post-war probabilistic methods — particularly the move toward limit state design in the second half of the twentieth century — tried to make the reasoning explicit, separating load uncertainty from resistance uncertainty and attaching each to something measurable. Codes developed under this framework, including the Eurocode suite and LRFD-based American codes, are more transparent about the logic: partial factors for load, partial factors for resistance, calibrated together against observed performance.
But even these codes inherited their calibration targets from the track record of older, less systematic codes. The acceptable probability of failure embedded in a modern limit-state code is not derived from a societal decision about what risk is tolerable. It is reverse-engineered from structures that existed and, mostly, didn't fall down. The number reflects what the profession has historically gotten away with as much as it reflects what is genuinely safe.
What makes this uncomfortable is not that the allowable is wrong — decades of structural performance suggest it is a reasonable place to land — but that its precision is borrowed. A value expressed to four significant figures does not thereby become a four-significant-figure truth. The allowable stress absorbs uncertainties in material, geometry, load estimation, construction quality and model accuracy, and it does so through a factor whose magnitude was set, at some point in the past, by people reasoning from experience rather than from theory.
The engineering tradition is not wrong to work this way. Experience is data; surviving structures are evidence. But a number that reads as a measurement is partly a measurement and partly an accumulated professional judgement, and the two portions are not labelled. The code gives you the result. It doesn't give you the reasoning that produced the factor that produced the result — and the gap between the factor nobody derived and the number on the page is where most of the real caution lives.