Ask a fabricator which joint they weld first and the answer is usually the easiest one, or the one at the end they are standing near. Ask which joint has to be welded last and the answer is much more informative, because it names the constraint. Some joint is always inaccessible once another is closed, and that joint decides the whole order.
So the sequence is built backwards. Identify the weld that can only be reached while the assembly is still open, then the one that depends on it, and work back to the first tack. What looks like a plan for welding is really a plan for access, and it is settled at the drawing stage if anyone is paying attention.
A torch needs room for the cup, room for the filler and room for the operator's hand, and all three are larger than the joint. A joint that is reachable with a bare tungsten is not necessarily reachable with a cup fitted, and a joint that can be seen is not necessarily one that can be welded to the root without the arc being blocked.
This is why a mock-up in cardboard or scrap is worth the hour it takes. It shows which joints disappear as the assembly closes, and it does so before any metal is cut. On thin-wall work the problem is worse, because the cup has to be close to the joint and the clearance needed is not negotiable.
The second constraint is where the shrinkage goes. Each weld pulls its own joint and, through the structure, moves everything connected to it. Welding a joint that is far from the datum early means every later weld acts on a part that has already moved, and the errors add.
Welding the joints nearest the datum last, after the outer structure is closed, gives the part a chance to settle before the dimensions that matter are set. It sounds backwards, and it works, because the last welds are then the only ones whose movement affects the critical features.
Most assemblies are better built as sub-assemblies that are welded out, checked, and then joined. A closed shape such as a tank or a muffler body has to be built from the inside out, with the internal parts fixed before the last panel or end cap goes on, because after that there is no access at all.
The order inside a closed shape is fixed by physics rather than preference: baffles, then internal supports, then the final closure. Getting the internal sequence wrong means cutting the part open, which is a repair rather than a build. This is where the titanium fabricated components approach of building and checking sub-assemblies earns its keep.
If a feature has to be machined, the machining comes after the welding that would move it and before the welding that would not. That rule alone settles most of the argument about whether to finish a flange face before or after the runners are joined to it. Weld first, then machine, then weld only what cannot disturb the machined face.
The exception is a feature that has to be machined before assembly for access, such as an internal bore that a torch would obstruct. In that case the machining is protected through the rest of the build, and the protection is part of the plan rather than an afterthought.
A leak test, a dimensional check or a pressure test has to happen at a point in the order where it can still be done and where a failure can still be repaired. Testing after the final closure is normal, but testing a sub-assembly before it is buried inside another is what saves the part.
The same logic applies to inspection of joints that will be hidden. If a joint will be covered by a plate or closed inside a shell, it needs to be inspected before that happens, and that inspection is a step in the order rather than an option. This is the practical reason the stainless welding records note which joints were completed at each stage.
A repair inverts the logic. The part arrives closed, and the access that a build sequence would have preserved is gone. The repair order therefore starts with what has to be cut away to reach the damage, then the repair itself, then the restoration of whatever was removed.
The trap is restoring the removed material before checking that the repair holds. Cutting a window, welding inside it and closing the window in one session feels efficient, but it removes the chance to test. Leaving the window open until the repair has been checked costs a second setup and saves the part.
An order that lives in the fitter's head is not a process. Written as a numbered list of joints, with the access note and the inspection point against each, it becomes something a second person can follow and something a checker can verify. It also becomes the basis for a weld map, which is how a joint can be traced back to who welded it.
The list does not need to be long. On most fabricated parts it is a handful of lines, and the value is in the two or three joints whose position in the order is not obvious. Those are exactly the ones that get welded in the wrong place when nobody wrote it down.
For repeat work a numbered sequence is also what makes a custom part reproducible across batches.
Sequence is a quotation input, because it decides how many setups, how much checking and how much rework the job needs.
Assembly order is not administration. It is the difference between a part that can be built once and a part that can be built repeatedly, and it is cheapest to decide when the drawing is still being read. The other parts range shows assemblies where the order is visible in the finished geometry, the tube work and bent assemblies depend on the same rules, and the fabricated components pages cover closed shapes. Where a job is quoted from a sample, the process begins by working out how the original was assembled, and the welding notes set out what each pass demands from the joint before it is closed.
