The print failure triage prompt

You are troubleshooting a failed 3D print with me. Do not give me a generic
checklist.

Printer: {printer, plus any mods that touch the hotend, extruder, or bed}
Filament: {brand, material, roughly how long the spool has been open, dried or not}
Settings: {layer height, nozzle temp, bed temp, speed, retraction distance/speed, cooling}
What happened: {describe the failure, where in the print it shows up}
What changed: {anything different since the last good print: new spool, new
slicer version, moved the printer, changed a setting}

Give me:
1. Your top suspected causes, ranked, and for each one the specific detail
   from my description that points to it. No cause without evidence.
2. One test per cause that confirms or rules it out, ordered cheapest and
   fastest first.
3. The single setting change you would make before anything else, with the
   exact value.

One variable at a time. If you need something I did not tell you, ask
before diagnosing.

Ask a model “why is my print stringing” and you get the same checklist that tops every search result: dry your filament, tune retraction, lower the temp. True, useless, and in no particular order. The problem was never a lack of possible causes. It was ranking them for your specific machine and your specific failure.

The “what changed” line is the load-bearing part of this prompt. Most print failures are not mysteries, they are regressions. Something was working, then something changed, then it stopped working. Force yourself to answer that question and half the time you have diagnosed it before the model replies.

Why it works

• “No cause without evidence” stops the model from dumping the full encyclopedia of print failures on you. It has to tie each suspect to something you actually said.
• Tests ordered by cost means you run the 10-minute temperature tower before the two-hour full reprint, and you check the free thing (is the spool path rubbing?) before buying anything.
• One variable at a time is the discipline everybody abandons at 11pm when they just want the print to work. Having it in the prompt keeps the model from suggesting you change five settings at once, which tells you nothing when the next print succeeds.

What good output looks like

The diagnosis reads like a decent forum answer from someone who actually read your post: “the fact that it appears above the third layer and you opened this spool a month ago points at moisture more than retraction, here is the 30-minute test.” Ranked, specific, testable.

Where it works (and doesn’t)

This wants a model with real 3D printing knowledge in the training data. Claude and the GPT-class models both hold up. Small local models will produce the structure but pad the ranking with boilerplate causes, which defeats the point. And no model can see your printer: if the diagnosis is confidently wrong about something physical, trust your eyes over the ranking.

The prompt does not fix your print. It fixes the part where you change four settings, it works, and you learn nothing.