When a Step Goes Red
A rebuild that fails announces itself in one way: the step that failed turns pastel red and pulses, and every other pill in the timeline turns gray. Find the red pill and hover it: the message is in its tooltip. Click it and the same sentence is in the banner at the top of its panel, which is where you can also do something about it.
The message lives inside the step
There is no global error list and no summary anywhere in the editor. Nothing counts the failures, nothing lists them, and no toast appears. If you do not look at the timeline, you will not know.
The banner carries the message word for word, and the message is the whole diagnosis. There is no code to look up and no second screen with more detail. Every step's panel renders it from the same place, so a failure never depends on that particular step's panel having remembered to show it. Warnings use the same banner in yellow, and a warning does not stop the rebuild.
Finding the right step first
When several pills are red, open the earliest one. Failures cascade: a fillet that lost its edge will take everything built on top of it with it, and fixing the first usually clears the rest. Work down the timeline, not up.
If a group's pill carries the error tint, the failure is inside it. Click the group to drill in, and the back arrow to come out.
And before anything else, consider undo. A step that was fine a minute ago failed because of the edit you just made, and Ctrl+Z is faster than diagnosis.
A grey step is not a red step
Before you go hunting, check that what you are looking at is a failure at all. A step can also be waiting, and waiting looks nothing like broken: the pill keeps its own colour at half strength inside a dashed ring, the rest of the timeline stays exactly as it was, and everything downstream carries on as though the step were not there.
That is what a step that has not been finished looks like: a revolve whose axis you have not picked, a sweep with no path, a reference plane still asking which face to measure from. Its banner reads
Pick the missing input to finish this step.
and the panel's own rows will be pointing at the field that is empty. Nothing is wrong with the model. The tool asked a question and it is still waiting for the answer, so click the empty field, pick in the viewport, and the step evaluates for the first time.
The distinction matters because it is the difference between "I have a bug" and "I did not finish". A region list is the one exception that is not even a question: leave it empty and it means every region of the sketch.
Family one: geometry refused
The step is asking for something the shape cannot give. The value is too big for the material available, or the elements you picked cannot be resolved together.
Fillet geometry failed — common causes: radius too large, edges interfere, edges not connectable, tangency problem. Try a smaller radius or fewer edges.
Shell geometry failed — common causes: thickness too large for local feature size, or selected faces leave the boundary disconnected. Try a smaller thickness or fewer faces.
Offset failed — try a smaller distance for this geometry
Offset Faces failed — the distance may be larger than a local feature, or the faces cannot be offset together. Try a smaller distance or fewer faces.
These messages name their own causes and end with a suggestion. Take the suggestion literally and in order.
On a fillet or a chamfer, use the number the failure hands you. Under the radius field, after a failed attempt, a small button appears reading Use 4.6 mm. That number is not an estimate: the kernel was asked repeatedly until it found the largest value it would actually accept on those edges. One click takes it. Nothing is shown there before an attempt, and that is deliberate — no maximum is printed under the field, because the only honest answer comes from having tried.
Otherwise, halve the value first. If a 4 mm fillet fails, try 2 mm. If 2 mm works, the problem is size, and the fix is either a smaller radius or more material where it runs.
Then cut the selection down. If a small value still fails, the problem is the selection. Split the step into two, with half the edges in each, and you will find out in one rebuild which edge is the culprit.
Family two: an input that vanished
The step is looking for a face, an edge or a body that is no longer there.
A referenced face/edge/vertex no longer exists
This is what you get after an edit that moved the element far enough that it could not be re-found, or after deleting the step that produced it. A deletion with dependents asks first now and lists them by name, so the common route to this message is an edit rather than a deletion — but the Delete only … button in that dialog is exactly the choice that produces it on purpose.
Two fixes. If the element still exists in some form, open the failed step, clear the pick and pick it again. If the step that made it is gone, the element is gone too, and the failed step has to be re-pointed at something else or deleted.
Three relatives belong here, and each names a different thing that went missing:
The body this sketch was attached to could not be identified
The face was never even looked for: the step that made the body no longer maps to a body at all. Re-attach the sketch to a face that exists.
A body this feature joins no longer exists — pick the bodies to fuse with again
An extrude set to Add keeps a list of the bodies it welds itself to. That list is your answer, so nothing quietly re-answers it when one of them is deleted or consumed by a boolean. Open the step's Fuses with row and pick again.
The face the start offset is measured from could not be found — pick a flat face again
An extrude whose start offset is measured from a face rather than typed as a number. The face has gone, or it is curved, or it lies in the sketch plane itself, where a zero offset cannot be told from a wrong-body match.
The same family covers the case where the inputs still exist but no longer meet:
Intersection is empty (selected bodies do not overlap)
Here nothing is missing. A dimension changed and the two bodies moved apart. Roll back to see where they parted company.
References That Survive an Edit explains why a face that is obviously still on screen can still be reported as gone.
Family three: a value the step will not accept
The geometry is fine, but what you gave the step is outside what it can work with. These messages state the rule rather than suggesting a smaller number.
Profile crosses the revolve axis — profile must lie on one side
Helical sweep does not support face-with-holes profiles in this release
The first is a rule of revolving: move the axis or the profile so they do not cross. The second is a limit of the release: build the outer profile as one sweep and subtract the holes afterwards.
Two more come from the extrude's end conditions, and both mean the pick rather than the number:
Extrude up-to-face needs a flat target face — this one is curved
Stopping at a face means stopping at its plane, and a cylinder or a fillet has no single plane to stop at. Pick a flat face, or use a blind height.
This extrude has no length — the start offset reaches the end condition
The prism came out zero long: a target face lying in the sketch plane, or a start offset that has eaten the whole distance.
A refusal of this kind will not clear by nudging the number. Read what it says the rule is and change the input to satisfy it.
Family four: an upstream step is disabled
A referenced feature is suppressed
Something this step depends on has been turned off. Right-click the disabled pill, the gray one with the cursor that refuses, and choose Enable.
If you disabled the step on purpose to find out what depended on it, this message is the answer you were looking for.
Warnings
A yellow banner does not stop anything, and it never repaints the pill: a step's colour already says what kind of step it is, so a warning shows as a small amber triangle in the pill's corner, with the sentence in its tooltip and in the panel's banner.
The one you will meet most is on a step whose pick was matched by measurement rather than by identity: the model rebuilt, the fillet landed somewhere, and the app is telling you it is not certain that is where you meant. Look at what is highlighted before you carry on, because nothing downstream will complain and neither will the export.
The other worth knowing sits on an extrude aimed at a face:
The target face is smaller than the profile — the extrude stops at its plane, not at its edge
Perfectly legal, and occasionally what you wanted. But if you picked that face expecting the prism to end at its edge, it did not: it ended at the infinite plane through it, and the overhang went past.
Not every refusal arrives as a red step, either. A tool that will not take what you clicked says so where you clicked it: the sketch's Project tool, handed an edge it cannot flatten onto the plane, refuses in a message that names the reason rather than filing a failure in the timeline.
Working through a failure
- Scan the timeline for the red pills and open the topmost one. A pill that is half-strength inside a dashed ring is not one of them: it is waiting for a pick.
- Read the banner and decide which family it is: something refused, something missing, something disallowed, something switched off.
- If the message ends with a suggestion, follow it. On a fillet or a chamfer take the
Use … mmbutton; otherwise halve the value before anything else. - If it names a missing reference, decide whether to re-pick or to re-point.
- Rebuild and look at the timeline again, not at the part. The part looks wrong in the same way whether one step is failing or five.
When you cannot clear it
Some steps cannot be recovered in place: the geometry they were built on has changed shape rather than moved. Delete the failed step and build it again on the part as it now stands.
A red step does not stop you saving or exporting
Neither is blocked, and neither warns you.
Saving is untouched. The document is the recipe, not the result, so a file with a failed step in it saves, closes and reopens like any other, and it comes back failing in the same place. Nothing is lost by leaving a red pill in a document overnight.
Export is the one to think about. A failed step produces no geometry at all, so whatever it would have made is missing from the file you download, silently. Nothing in the export dialog mentions the failure. A cut that failed leaves a part with no hole in it, and that part slices, prints and wastes an evening before you notice.
So look at the timeline before you export, not at the part.
See also
- Reading the Timeline: the gray-and-red display that is the only cue you get
- References That Survive an Edit: why a step loses a face that is still there
- Editing a Part You Have Built: the edits that cause most of these, and how to undo them
- Working with Several Bodies: deleting a feature and what it leaves broken