There is a particular silence in a workshop when the final piece doesn't fit. The glue is spread and its open time is running out. The tenon that slid home perfectly an hour ago now stops a quarter inch short. Perhaps the panel beside it swelled, the sequence was wrong, or a clamp blocks the last joint. Every maker who has worked long enough knows that silence. It is the sound of a design meeting reality too late to negotiate.
The root problem is rarely a bad part. Parts are usually verified individually. They are measured, checked against the drawing, and inspected. Failures live in the spaces between them: in tolerance stacks, in assembly sequences, and in unwritten assumptions two components make about each other. Component verification answers whether each piece is right. Only integration testing answers whether the pieces are right together.
Aerospace, shipbuilding, and fine joinery all arrived at the same answer. Assemble everything once without commitment, before assembling it for real. The fit-up test is cheap insurance against expensive irreversibility. What follows is a working method for doing it deliberately. It covers how to dry-assemble without losing the ability to modify, how to decide which interfaces earn scrutiny, and how to document what you find so that discoveries actually become fixes.
Dry Assembly Protocol
A dry assembly is a full rehearsal of the build that uses reversible connections in place of permanent ones. The governing principle is simple: every substitution should preserve geometry while preserving the ability to undo. Cleco fasteners exist for exactly this purpose in aircraft sheet metal work. These spring-loaded temporary pins hold skins and ribs in exact registration through the same holes the rivets will later fill. Welders do the same with tack welds, which are small enough to grind out if the fit-up is wrong.
The temptation is to dry-fit selectively, checking the tricky joint and assuming the rest will work. Resist it. A proper protocol assembles the complete system in the sequence planned for final assembly, using the actual fastener locations and the actual orientation. Sequence matters because many integration failures are not dimensional at all. A part can fit perfectly and still be impossible to install. An earlier component may block the angle the part needs to enter, or a wrench may no longer reach a bolt.
Establish datums before you begin. Choose reference surfaces, such as a flat base, a centreline, or a primary mating face, and measure everything from them rather than from each other. Measurements chained from part to part accumulate error invisibly. Measurements from a common datum reveal where the error actually lives. Mark witness lines across joints in pencil or layout fluid so that parts return to precisely the tested position. Label the orientation of every component whose symmetry invites a mirror-image mistake.
For adhesive and curing processes, the dry assembly doubles as a timed rehearsal. Run the full clamp-up with a stopwatch. If the sequence takes twelve minutes and the adhesive's open time is eight, you have found a process failure that no measurement would reveal. Stage clamps, cauls, and spacers within reach and in order. Furniture makers who build complex carcases often split one ambitious glue-up into several smaller sub-assemblies for this reason. The rehearsal showed that the single operation demanded more speed than anyone could deliver.
Finally, test disassembly as part of the protocol. Can the assembly come apart for future maintenance, repair, or transport? A design that only goes together once embodies a design decision, and that decision should be made consciously. The dry assembly is the last point at which reversibility is free. After it, every change carries the cost of cutting, grinding, drilling out, or starting over. Treat that threshold with the seriousness it deserves.
TakeawayA dry assembly is not a check on parts but a rehearsal of the process. Geometry, sequence, access, and timing all have to pass before anything becomes permanent.
Critical Interface Prioritization
Not every interface justifies the same scrutiny. A complex build may have hundreds of joints, and testing each one with equal rigour wastes effort and dilutes attention on the few that matter. The useful filter crosses two variables. Consequence is what happens if the interface fails to mate. Uncertainty is how confident you actually are that it will mate. Plotted on a simple two-by-two grid, interfaces sort themselves into three groups: those you can trust, those you should watch, and those that need physical verification before anything else proceeds.
Consequence is largely a function of irreversibility and dependency. A welded structural node, a bonded composite seam, or a cast-in anchor bolt cannot be adjusted after the fact. Errors also spread from interfaces that many downstream parts reference, such as a mounting plate, a primary bulkhead, or the squareness of a frame. Any error there carries through the rest of the build. A bolted bracket with slotted holes, by contrast, carries low consequence even if it misses, because the design already allows for correction.
Uncertainty has subtler sources, and experienced engineers learn to recognise them. Interfaces between parts made by different people, shops, or processes are inherently uncertain, because each side carries its own assumptions. So are interfaces involving tolerance stacks, where several individually acceptable deviations add up to an unacceptable one. Materials that move add uncertainty a drawing cannot capture. Wood moves with seasonal humidity, polymers with temperature, and castings with shrinkage. Any interface designed from a dimension you did not measure yourself also belongs in this category.
The 1999 loss of NASA's Mars Climate Orbiter is the textbook example. Ground software supplied by one contractor produced thruster impulse data in pound-force seconds. The navigation team using that data expected newton-seconds. Each software component worked as its authors intended. The failure lived entirely at the interface between two organisations, each confident in its own conventions. The interface was high in both consequence and uncertainty, yet nobody treated it that way, and the spacecraft was lost as it arrived at Mars.
In practice, build an interface register before the dry assembly. List every mating condition, score each one on consequence and uncertainty, and rank them. The top quadrant gets a physical fit-up that is measured and documented. The middle group gets inspection against drawings. The bottom group gets trust. The ranking also tells you what to build first. Prototype or mock up the highest-risk interfaces early, sometimes in cheap materials like plywood, foam, or 3D-printed proxies, long before committing expensive stock to them.
TakeawayRisk lives where consequence meets uncertainty, and uncertainty concentrates wherever two sets of assumptions meet. The boundary between teams is often more dangerous than the boundary between parts.
Issue Documentation Methods
A dry assembly that finds problems but records them poorly has done only half its job. Integration issues are easy to notice in the moment and easy to forget by the time final assembly arrives. This is especially true when the person who noticed is not the person who fixes them. The discipline that closes this gap comes from aerospace quality systems. Every discrepancy becomes a numbered record with an owner, a disposition, and a defined closure condition.
A useful issue record is compact but complete. It captures the interface involved, the expected condition, the measured condition, and a photograph with a scale reference. It also includes a hypothesis about the cause and a proposed resolution. The critical distinction is between observation and interpretation. Record what you measured before recording what you think it means. Interpretations change as understanding develops, but raw measurements stay useful long after the first hypothesis proves wrong.
The Hubble Space Telescope's primary mirror shows why. The reflective null corrector used to test the mirror had been assembled with a lens about 1.3 millimetres out of position. Other tests during manufacture gave results that disagreed with that instrument. Those results were dismissed as errors in less precise equipment rather than logged as unresolved discrepancies. The anomaly was observed and then explained away. The mirror was polished precisely to the wrong shape, and the flaw was discovered only after launch.
Every issue needs an explicit disposition. Manufacturing quality systems use a small vocabulary worth adopting. Use as is applies when the deviation is acceptable. Rework applies when the part can be brought to specification. Repair applies when the part can be made functional but not conforming. Scrap applies when neither is possible. Forcing a choice from this list prevents the most dangerous state an issue can occupy: acknowledged but undecided, and carried silently into final assembly.
Closure must be verified, not declared. An issue is not resolved when someone says it has been fixed. It is resolved when the interface has been re-tested and the measurement recorded. For small shops, a spreadsheet or even a bound notebook is enough, as long as each entry ends with closure evidence. Physical tags on affected parts help too. A red tag that stays on until the log shows closure keeps an open issue visible at the bench, where decisions actually happen.
TakeawayAn anomaly that has been explained away is more dangerous than one that was never seen. The goal of documentation is to make it impossible for a known problem to become a forgotten one.
Integration testing rests on a simple asymmetry: verification is cheap before commitment and expensive after it. Every adhesive, weld, rivet, and pour marks a one-way door. The fit-up test exists to make sure you walk through each door knowing what is on the other side.
The method reduces to a repeatable cycle. Register every interface and rank it by consequence and uncertainty. Rehearse the full assembly with reversible connections, in sequence, against common datums, and with timing for any time-critical process. Record each discrepancy as a measured observation with an owner and a disposition. Resolve and re-test until every record closes with evidence.
Fuller argued that good design anticipates problems rather than reacting to them. The dry assembly is anticipation made physical: a deliberate rehearsal of failure under conditions where failure costs almost nothing. The parts will tell you whether they belong together. The only question is whether you ask them before or after it is too late to listen.