The Factor Nobody Chose

§ 01The Illusion of a Derivation

Open almost any structural code and you will find a factor — 2.5, 3.0, 4.0, 1.67 — sitting in the text as though someone once sat down with a probability distribution, a failure database and a sharp pencil and arrived at precisely that number. The presentation is orderly. The origin, more often than not, is not.

Safety factors have a texture that careful engineers learn to recognise. Some are genuinely derived: they emerge from statistical treatment of material scatter, from reliability targets, from the kind of analysis described in limit-state codes where partial factors carry explicit probabilistic meaning. But a large proportion of the factors embedded in working practice today have a different character. They are inherited — passed from one generation of codes to the next, from one industry's practice into another's, from a committee vote that nobody recorded into a number that eventually became unchallengeable because it was old.

This is not a scandal. It is, in many cases, a reasonable response to the epistemic situation engineers have always occupied: you do not know everything about the material, the load, the workmanship or the use, and you need a number to work with today. The factor you borrow from your predecessor embeds, however clumsily, whatever your predecessor learned. The question worth asking is not whether that lineage is tidy but what it actually contains.

The number acquires authority that is not entirely technical

§ 02Where the Numbers Accumulated

The history of the factor-of-three, -four or -five in nineteenth-century iron and early steel construction is largely oral. What survives in the technical literature is a scattering of retrospective justifications: factors that were already in use before anyone tried to derive them, then explained afterward by reference to material variability, overload risk and the limitations of calculation. William Rankine's mid-Victorian writing on applied mechanics is illuminating here not because he derived a single canonical factor but because he catalogued what practitioners were already doing and tried to give it structure. The numbers preceded the structure.

The process was often failure-driven in a way that is almost the inverse of principled derivation. A structure failed; an inquiry attributed the failure to understrength or overload; the community added conservatism to the next generation of practice. What came out was not a factor derived from first principles but a factor inflated from the last disaster. The infamous rash of early railway bridge collapses in Britain contributed to conservatism in bridge loading rules that lasted well into the twentieth century — not because someone calculated exactly how much conservatism was warranted but because failure had made under-conservatism politically and professionally unacceptable.

Pressure vessel practice tells a similar story with a cleaner paper trail. The ASME Boiler and Pressure Vessel Code, first published in 1914 and continuously revised since, has carried safety factors that have shifted over its history as the Code's understanding of material behaviour, inspection quality and stress analysis matured. The older factors were large partly because the uncertainty they covered was genuinely large: stress distributions in complex geometries were not calculable with any precision before finite element methods, and material certificates were not trusted the way a modern mill certificate is. When calculation got better and inspection got better, the factors came down — but they came down through committee deliberation, not from a single derivation that everyone agreed on. The process was negotiated.

That negotiation is important. Standards committees are not a collective of disinterested mathematicians. They contain representatives of industries who have different interests in where the number lands — manufacturers who want thin walls and low cost, insurers and regulators who want margin against unknowns, operators who want longevity. The number that emerges from this process carries, whether acknowledged or not, the fingerprints of all of them.

§ 03What the Number Contains and Doesn't

A derived safety factor has a clear meaning: it is the ratio of the characteristic strength to the design action, set so that the probability of failure over a defined reference period stays below a target value. You can interrogate it. You can ask what distribution was assumed, what return period, what consequence class. Limit-state design in structural engineering — the framework behind the Eurocodes and much of modern structural practice — attempts precisely this. Partial factors on loads and on material resistance are calibrated against explicit reliability targets, and the calibration reports are published.

But even here the inheritance is visible. When the Eurocodes were calibrated, one of the constraints imposed on the calibration was that the resulting designs should not be wildly inconsistent with what earlier national codes had produced. Designers and clients had expectations. Economies of material that went beyond what practice had established were, in some sectors, not politically viable. The probabilistic framework gave the factors a new rationale, but the numbers were shaped partly by continuity with what already existed. The derivation legitimised the inheritance; it did not replace it.

For a large body of working practice — crane ratings, rigging hardware, lifting gear, pipework, fasteners, many categories of pressure equipment — the factors in use today are best understood as accumulated conservatism encoding both what has failed and what has not. The factor on a shackle's working load does not come from a reliability analysis in the Eurocodes sense. It comes from a long history of practice in which a particular ratio between proof load and rated load produced an acceptable incidence of failure — acceptable meaning that the industry could function and the catastrophic incidents stayed rare enough that the number was not forced upward. This is empirical calibration, and it is legitimate. But it is not derivation.

The practical implication is asymmetric. A derived factor can in principle be renegotiated when the inputs change: better material consistency, better inspection, better load characterisation. An inherited factor is stickier. It has become standard practice; departing from it requires demonstrating that nothing was lost, which is hard to prove about events that did not happen. Factors have come down in several industries as practice matured and analysis improved — the trajectory in pressure vessel codes is a clear example — but the friction is real. The number acquires authority that is not entirely technical.

From the notes — The two kinds of factor

Term	What it means here
Derived factor	has an explicit probabilistic meaning; can be interrogated and renegotiated when inputs improve
Inherited / empirical factor	encodes accumulated practice; legitimate but tied to the context that produced it; does not automatically apply outside that context
Failure modes not covered	a factor calibrated for one failure mode (e.g. static overload) may be silent on another (fatigue, dynamic load, environment)

§ 04How to Hold an Inherited Number

None of this means the inherited factor is wrong. For most applications in well-trodden territory, it is probably well-calibrated, in the sense that following it produces structures and systems that fail at acceptable rates. The accumulated failures that shaped it provide a kind of empirical backing that a purely theoretical derivation, working from uncertain inputs, might not match.

What changes when you know the origin is the quality of your reasoning about departure cases. If you are working in territory the inherited factor was not designed to cover — a novel material, an unusual loading pattern, an operating environment with different characteristics from the practice that produced the original number — then you cannot simply import the factor and assume it still carries the same conservatism it carried in its original context. The origin of the allowable matters precisely because it tells you which unknowns the number was buffering against, and which ones it was not.

This is also why the question "is this factor conservative enough?" is harder to answer than it looks. Conservative for what specific failure mode, operating on what assumptions about load and material, tested against what historical population of structures? A factor that is generous on material strength may be silent on fatigue. A factor calibrated for static load may not protect against dynamic loading. The number does not speak; the reasoning behind it does — when you can find it.

Treat inherited factors with respect, not reverence. They earned their place. But they earned it in a specific context, and they describe that context whether or not the text around them admits it.