Tolerance & Fit

Tolerance in off-site construction comes from at least two different places, and confusing them causes real failures. The first is manufacturing tolerance: the range within which the factory can actually make a part. A module, a panel, a pod is made to its nominal size plus or minus some amount, determined by the material, the process and the jigs. A good factory holds tight, repeatable tolerances — far tighter than open-site work — which is one of off-site's genuine advantages. But 'tight' is not 'zero'. Steel grows and shrinks with temperature; timber moves with moisture; concrete shrinks as it cures. The part that leaves the factory is within tolerance, not exact.

The second is setting-out (or site) tolerance: the range within which the site works — the foundations, the slab, the supporting structure, the datum lines — can actually be positioned. This is usually looser than factory tolerance, because site conditions (weather, ground, hand setting-out) are harder to control. And here is the crunch that defines off-site construction: you are bringing a precisely made factory part onto a less-precisely made site. The classic failure is a beautifully accurate module arriving at a foundation that is a couple of centimetres off where the module expects it. The factory did its job; the site did a normal job; and the two tolerances did not talk to each other. The joint between them has to reconcile the difference.

There are more sources beyond these two — the tolerance of *measurement* itself (you cannot even measure a thing perfectly), and the movement of parts *in service* (thermal, moisture, structural deflection, which behave like a live tolerance). The practical lesson is that tolerance is not a single number you look up; it is a budget assembled from several contributors, and the joint has to be designed to cover the worst realistic combination of them. Off-site design treats the interface between the accurate factory and the approximate site as a first-class problem, not an afterthought — because it is exactly where beautifully made parts go to fail if nobody planned for the gap. (The actual tolerance values for any system and any site come from the manufacturer and the engineer; here we are learning the kinds of tolerance and how they combine.)

The reason tolerance is dangerous — rather than merely annoying — is accumulation, also called tolerance stack-up. Line up several parts in a row, each within its own small, perfectly legal tolerance, and the errors do not cancel; in the worst case they add. Put five modules side by side, each allowed to be up to a few millimetres oversize, and the fifth gridline can be off by five times that. It is not that any part was made badly — every one passed inspection — it is that the chain of 'small and legal' became 'large and visible' at the far end. This is why the corner of the building farthest from your setting-out datum is where tolerance problems surface first.

The same logic runs vertically. Stack modules floor on floor and the small oversize of each floor zone accumulates up the building, so the top storey can sit noticeably higher than the drawing says, throwing out cladding joints, lift openings and connections to any site-built core. It runs through assemblies too: a panel within a bay within a structural grid within the building is a chain of nested tolerances, each of which contributes. The naive mental model — 'each part is within a few millimetres, so the building is within a few millimetres' — is simply false, and believing it is how projects get nasty surprises at assembly.

The disciplined response has a few moves. First, do the sum: estimate the worst-case accumulation along the critical chains (the long row, the tall stack, the line to the site-built core) rather than assuming parts are perfect. Second, do not let one joint carry the whole accumulation — distribute the take-up so each joint absorbs a little, rather than asking the last gap to swallow everything. Third, provide adjustment at the right places — a deliberate adjustable joint, a designed tolerance gap at the connection to site-built work, a datum you re-establish rather than one you let drift. Fourth, measure as you go, so a drift is caught at module three, not discovered at module ten. The detailed numbers and the method for any given system belong to the engineer and the manufacturer; the design mindset — chains accumulate, plan for the worst realistic case, distribute and adjust — is yours to carry from the first detail.

A useful distinction in the tolerance literature separates induced deviations from inherent deviations, and it sharpens how you design a joint. Induced deviations are the ones we have been discussing: the differences between where something is *meant* to be and where it actually ends up, caused by the imperfections of manufacture, setting-out and measurement. They are, in principle, about *accuracy* — how close we got to the target — and they are reduced by better process, tighter jigs and careful setting-out, though never to zero. Inherent deviations are different: they are the *movements a part undergoes in use* — thermal expansion and contraction, moisture movement in timber, shrinkage and creep in concrete, structural deflection under load. These are not mistakes; they are physics, and they will happen to a perfectly made part.

Why does the distinction matter to a designer? Because the two demand different provisions in the joint. Induced deviation is covered by a tolerance allowance — a gap and an adjustment range sized to the worst realistic combination of manufacturing and setting-out error. Inherent movement is covered by a movement allowance — a gap and a flexible, resilient detail (a movement joint, a compressible seal) that lets the part grow, shrink and deflect without cracking the junction or tearing the seal. A joint that is sized for fit but not for movement will look perfect on the day of assembly and then fail through a season of thermal cycling; a joint sized for movement but not for fit may never go together in the first place. Good modular joints provide for both, and keep the two budgets conceptually separate so neither gets forgotten.

There is a corollary worth internalising: because some of the deviation is inherent and unavoidable, the answer is never simply 'demand tighter tolerances'. Tighter manufacturing tolerance costs money and has limits, and it does nothing about thermal and moisture movement. The cheaper, more robust strategy is usually to design the joint to be forgiving — to accommodate variation gracefully — rather than to chase a precision the real world will not hold. This is a recurring theme of the whole dimensional discipline: you cannot make imperfection go away, so you design so that it does not matter. The specific split between tolerance and movement, and the numbers for each, are engineering outputs tied to the material and the system; the conceptual separation is a design tool you can use on day one.

All of this converges on a single design act: allocating tolerance — deciding, in advance and on purpose, how much variation each part and each joint is allowed, and where the take-up happens. This is the opposite of the hopeful default, in which every part is drawn at its exact nominal size, every joint is drawn as a clean line, and the drawing quietly assumes perfection it will never get. Allocating tolerance means designing the gap: giving the joint a nominal gap (so parts that are a bit big still fit) plus a tolerance range around it (so the gap can open or close as the real parts vary), and then detailing the joint so that a sealant, a cover trim, an overlap or an adjustable fixing closes that range reliably for weather, fire and acoustics. The fit is held by the designed joint, not by the parts happening to be exact.

There is an order of operations that experienced designers follow. Decide the critical dimensions that must be controlled (the ones a person sees or that other systems depend on) and the non-critical ones that can absorb slack. Choose where the slack lives — ideally at a joint designed to take it, at a junction with a forgiving material, or at a deliberate adjustable connection, and *not* at a prominent sightline or a safety-critical interface. Make work sizes smaller than coordinating sizes so a part never has to be its absolute maximum to fit its space — the gap is built into the sizing, echoing the coordinating-size-versus-work-size idea from lesson 5.1. And detail the joint for the worst case, so that even when every contributor is at its unfavourable extreme, the joint still closes. A joint that only works when everything is perfect is not a joint; it is a wish.

The payoff is robustness: a building that goes together on a normal day with normal parts on a normal site, because the design expected exactly that. The cost of getting it wrong is the off-site nightmare — a module that will not land, a cladding line that steps, a fire or acoustic seal that cannot be closed because the gap it needs is not there — discovered at assembly, where fixing it is slowest and most expensive. So treat the phrase 'it fits on the drawing' as a warning, not a reassurance. The real test is whether it fits when the parts are at the edges of their tolerances and the ground is a little off — and whether you designed the gap that makes that true. As always, the precise tolerances, gaps and movement allowances for a given system and site are determined by the manufacturer and the engineer; the designer's job is to build the thinking in from the start and to leave room for the real world to be itself.