Several different models, sliced at different times, all fail at roughly the same height above the plate. Different shapes, same failure point.
Repeatability is the diagnosis
Random faults do not repeat. When several unrelated prints fail at the same Z position, something at that position is causing it — and the position is either in the machine's vertical travel or in the layer where the models happen to change.
Those are two completely different investigations, and it is worth ruling out the second one first because it costs nothing.
Check the file before the machine
Slice one of the failed models and look at the preview at the failure height. Ask what changes there.
- Does a bridge or a large overhang start? Then the height is a coincidence of geometry, not a machine fault, and the fault is bridging or an overhang problem.
- Do supports begin or end? Support interfaces and the transition to unsupported material are a common failure point.
- Does the cross-section shrink abruptly? A part that goes from a large footprint to a slender column at a specific height concentrates every load at that transition, and small layers cool differently too.
- Does the model have a defect there? Non-manifold geometry in a downloaded file can produce a layer with missing or corrupted paths, and it will do so at the same height every time.
If several different models fail at the same height, this whole branch is unlikely — but if the "different models" are all versions of the same file, check here first.
The machine's vertical travel
If the file is innocent, something physical lives at that height.
A cable chain or loom that snags. The toolhead's umbilical has to move through its whole range. At some point it can catch on a frame member, a fan shroud or its own strain relief, adding a sideways force or physically restraining the head.
Debris or damage on a rail or leadscrew. A dent, a burr or a patch of hardened grease at one point on the Z axis binds the gantry every time it passes.
A frame member the gantry fouls. A cable clip, a screw head, a badly routed wire, or an accessory fitted after the machine was built.
A join in a linear rail or rod. Rare, but machines with segmented Z hardware can have a step at the junction.
The test that finds it in two minutes
Disable the motors, and move the gantry slowly up through the failure height by hand, watching and listening.
You are looking for a change in resistance, a click, a rub, or a cable that goes taut. Do it two or three times, and do it with the toolhead parked at several positions across the X axis — a snag that only occurs when the head is at one end is easy to miss with the head parked in the middle.
Then run the same check with the machine powered and the axis moving under its own control, watching the loom rather than the toolhead. Most people find the fault in this step, having spent the previous week on slicer settings.
A diagnostic print worth having
Print a single tall, thin tower and nothing else — 20 mm square, as tall as your usual failures. It costs little material, it takes the geometry variables out entirely, and it either reproduces the failure at the same height or it does not.
If it reproduces, the machine owns the problem. If a plain tower prints perfectly to full height while real models keep failing at the same place, the height was never the cause and the models share a feature you have not spotted yet.
What to fix when you find it
Re-route the loom with more slack and secure it so it cannot wander into the path. Clean and inspect the rails and the screw along their whole length, not just where you can reach easily. Remove or reposition anything added after the machine left the factory, since accessories are the commonest cause of a new snag on a machine that used to be fine.
Where this is most common
Tall machines with long cable runs, printers with tightly routed looms, and any machine that has been modified. It is also more common on the second half of the Z range, simply because that is where the loom is most extended and where fewer prints ever reach — a machine can be flawless for a year of short prints and fail on the first tall one.
Material makes no difference to this fault, which is itself a useful signal: if the failure follows the height rather than the filament, stop changing profiles.