Print Failure Diagnoser
Ranks what to check next by how much it would actually tell you — and the best first question turns out not to be a symptom at all.
About the print
These carry more information than most of what you can see — answer them first.
What you can see
Only answer what you've actually observed. "Not sure" is a real answer and costs nothing.
What still fits
What to check next, ranked by what it would tell you
The symptoms do discriminate — it's how people use them that doesn't
I expected to find that several causes share every symptom. They don't. Across these 14 causes and 16 observations there are zero indistinguishable pairs — every cause has a unique signature — and greedy questioning averages 3.93 questions against an information floor of ⌈log₂14⌉ = 4. The received symptom list is already close to an optimal decision tree. The problem isn't that it fails to separate causes.
The problem is that expected information gain hides a large asymmetry. A symptom present in 2 of 14 causes is worth 2.81 bits when present and 0.22 bits when absent — a factor of nearly thirteen. Averaging them gives 0.59, which makes a decisive symptom look feeble. Ranking by expected gain is right for choosing what to ask and wrong for someone already staring at a failed print.
So the rule has two halves. If you have a distinctive symptom, lead with it: a layer shift takes 14 causes down to 2, which is 2.8× better than the best question anyone can offer. If you don't have one, don't go fishing through a symptom list hoping something matches — ask the balanced question, which pays a full bit whichever way it answers.
And the best balanced question isn't a symptom at all. "Does it affect every part on the plate?" splits the causes exactly 7 and 7 — a perfect 1.000 bits, and better than every purely visual symptom in the set. The most useful thing to notice about a failed print is usually a fact about its context rather than its appearance.
- Change one thing at a time and re-print the same test. Two changes at once means a fixed print tells you nothing about which change fixed it. This costs one extra print and is the biggest difference between people who fix printers quickly and people who chase settings for weeks.
- Prefer the mechanical explanation to the settings explanation when both fit. A profile that worked for months and then stopped is almost never a settings problem — the settings didn't change. Moisture, partial clogs, belt tension and bed level all drift on their own.
- If nothing fits every answer, you probably have two faults. Prints fail in combination, and no single cause will explain the symptom set. Fix the most confident one and re-observe.
- "Not sure" is a real answer. Guessing at a symptom you haven't checked is worse than leaving it blank, because a wrong answer eliminates the right cause and you'll never get it back.
How to use
- Answer the context questions first — they carry the most information.
- Leave anything you have not actually checked as Not sure.
- If you have one distinctive symptom, start there instead.
- Change one thing, then re-print the same test.
Frequently asked questions
What should I check first when a 3D print fails?
Whether it affects every part on the plate or just one. That single question splits the common causes exactly in half — a perfect one bit of information, and better than any visual symptom. The most useful thing to notice about a failed print is usually a fact about its context rather than its appearance, which is the opposite of how most troubleshooting guides are organised.
Do print symptoms actually identify the cause?
Yes, better than you might expect. Across fourteen common causes and sixteen observations there is not one indistinguishable pair — every cause has a unique signature — and asking the best available question each time isolates a cause in about four questions, which is the theoretical floor. The received symptom list is close to an optimal decision tree already.
Why does troubleshooting feel so unreliable then?
Because of an asymmetry in what a symptom is worth. One produced by only two of fourteen causes carries 2.81 bits when you see it and 0.22 bits when you do not — a factor of nearly thirteen. Working down a list of symptoms mostly generates the cheap answers, and the rare decisive one only helps on the day you happen to have it.
Should I start from a symptom or from a question?
From the symptom if you have a distinctive one, and from a question if you do not. A layer shift takes fourteen causes down to two, which is nearly three times better than the best question anyone can offer. But if nothing stands out, do not go fishing through a symptom list hoping something matches — ask the balanced question, which pays a full bit whichever way it answers.
What causes stringing on a 3D print?
Wet filament, untuned retraction or too high a nozzle temperature — and they are distinguishable. Wet filament pops audibly and usually appeared on a profile that worked before. Too hot also sags overhangs and droops fine detail. Untuned retraction does neither. If stringing returned on a previously tuned profile, dry the filament before touching retraction, because the retraction was probably right.
Why did my print come off the bed?
Either the first layer was too high or the bed was too cool, and the two look different. A first layer too high fails at the same point on every retry and leaves visible gaps between the lines; a cool bed lifts at the corners first as the part contracts. Both affect every part on the plate, which is why that question is worth asking early.
What does a layer shift mean?
A loose belt or pulley, or a collision — and the height tells you which. A shift at a random height each time is almost always the belt slipping or a grub screw off the flat of the shaft. A shift at the same height on every retry is the head hitting something, so look for a raised blob, a curled edge or a warped part sticking up at that height.
Why did a profile that worked for months stop working?
Almost never settings, because the settings did not change. Filament moisture, a partial nozzle clog, belt tension and bed level all drift on their own, and a failure that appears mid-print after hours of good printing points strongly at moisture or a developing clog. The temptation is to adjust what is easy to adjust rather than what actually moved.
What if no cause explains all my symptoms?
You probably have two faults at once, which is common and is not a gap in the list. Wet filament plus a marginal first layer produces a symptom set no single cause covers. Rank the candidates by fewest contradictions, fix the most confident one, then re-observe — hunting for one answer that explains everything will waste an evening.
Is it better to guess at a symptom or leave it blank?
Leave it blank, always. A wrong answer eliminates the correct cause permanently and you will never get it back, whereas an unanswered question merely leaves you where you were. Not sure is a real answer that costs nothing, and the difference matters most on the symptoms that are hardest to judge — which are usually the informative ones.
How many things should I change at once?
One, then re-print the same test. Two changes at once means a fixed print tells you nothing about which change fixed it, and a still-broken print tells you nothing about either. It costs one extra print and it is the single biggest difference between people who fix printers in an evening and people who chase settings for weeks.
What is the difference between under-extrusion and cold layers?
They look similar and break differently. Under-extrusion from a partial clog or bad flow calibration shows as thin walls and gaps in the top surface. Layers that are merely cold look fine but separate cleanly along a layer line under mild force. If the part splits rather than snapping across, raise the temperature before touching flow.
🔒 This tool runs entirely in your browser. Nothing you enter is uploaded, logged, or stored.