Phase 3 · Ship it · L18

The validation set

Three prints and five criteria. Either the profile passes, or it names the step you have to go back to.

⏱ ~3 printer-hours, 20 minutes of you• 📏 Tools: digital calipers• Phase 3

◎ Why this lesson

Your mission says: hit print, walk away, get a part that's dimensionally right and doesn't delaminate. That is a claim, and a claim needs a test. Without one, "the profile is done" means "I got bored of tuning" — and you'll discover the gap on a part you actually needed. The validation set turns done into something with a pass mark. Its second job is diagnostic: each acceptance criterion maps to exactly one Phase 2 step, so a failure tells you where to go rather than leaving you to guess.

Three prints, in this order, and you only sign off when all three pass.

1The three prints

1 · Dimensional test — a calibration cube Measure X, Y and Z with calipers. Target within ±0.15 mm. This is the only one of the three that catches a machine problem surviving from Phase 1, so run it first.
2 · Overhang + bridge + stringing combo One print that exercises temperature, cooling and retraction together. These three interact, so a combined test is more honest than three isolated ones — a temperature that passes alone can fail once the fan is doing real work.
3 · A real part you actually print Non-negotiable. A profile is only good if it's good on your geometry. Test models are chosen to be hard in generic ways; your parts are hard in specific ones — a thin boss, a captive nut, a long unsupported span you happen to need.

⬇ Download · prints 1 and 2

Print 1: calibration-cube-20.stl — a plain 20 mm cube with clean caliper faces (no embossed lettering to sit under the jaws). Mark the X face with a pen before you take it off the plate, while the orientation is still obvious.

Print 2: overhang-bridge-test.stl from Lesson 14, printed together with retraction-spikes.stl from Lesson 13 — overhangs, a bridge and stringing on one plate, exercising temperature, cooling and retraction against each other. Prefer a single combined model? The Printables all-in-one test linked in the references does the same job.

⌁ Elephant foot compensation belongs here, at the end

Print Settings → Advanced → Elephant foot compensation, visible in Advanced or Expert mode. It shrinks the first layer's perimeters to counter the squish that makes the bottom edge bulge; around 0.2 mm works for a 0.4 mm nozzle.1

Tune it last, and only here. It's a top-level setting influenced by every other calibration — flow, retraction, temperatures, Z height. Tuning it before those are settled and then changing them invalidates it immediately.1 It is the natural companion to print 1, since both are about dimensional truth.

2The five criteria, and what each one accuses

This is the table that makes the validation set worth running. Each criterion points at one step.

Read a failure as an accusation, not a verdict.
Criterion	How to check	Failure means → go back to
Top surface uniform no pinholes, no ridges	Low-angle light across the top solid layer	Pinholes → under-extrusion → L11 extrusion multiplier Ridges → over-extrusion → same step, other direction
Corners square no bulge, no post-travel gap	Sight along an external perimeter through a 90° corner	Bulging → advance too low Pinched or gappy after the corner → too high → L12 advance
Seam a fine line not a blob	Find the seam on a curved wall and run a thumbnail down it	L12 advance + L13 retraction, then scarf seams in the print profile
Layer lines even top to bottom	Compare the bottom third against the top third under the same light	Degrading upward → L14 cooling or L15 MVS (on the Core One, suspect chamber heat first)
Snap test granular not a clean layer plane	Break the part. Yes, really	Clean fracture along a layer → L10 temperature, or too much fan in L14

▲ The dependency rule still applies

If validation sends you back to L10 temperature, you are not making one small change. Temperature is step 1 of the calibration order, so everything below it — EM, advance, retraction, cooling, MVS — is void again. Re-run the chain, then re-validate. This is exactly why validation happens at the end and not continuously: an early failure is expensive, and the order exists to make early failures rare.

Note the asymmetry the snap test forces on you. It is the only criterion that requires destroying the part, and it is the only one that tests strength rather than appearance. Skipping it is how people end up with beautiful profiles that produce parts which fail under load — the exact opposite of what a functional print is for.

3What to watch on your machine

Open your machine. The criteria are the same on all three; what fails first is not.

MK3S · dimensional and motion first

Most likely to fail: the dimensional test and the corner criterion, because this is the machine where motion gives out first. Its Y axis is carrying your part.

Check the cube's X against its Y, not just each against nominal. A consistent difference between the two axes points at skew or belt tension — that's Phase 1 leaking through, not a profile fault. Re-run Calibrate XYZ.
Validate at MK3S speeds. If you validated at 45–60 mm/s perimeters, you've certified the profile at 45–60 mm/s perimeters. Don't then print production parts at 90.
Diagonal moiré on the cube's walls is the linearity-correction signature from Lesson 8, not an extrusion problem. Don't chase it with EM.

MK4S · flow and cooling first

Most likely to fail: the layer-lines and top-surface criteria, because with a ~22–28 mm³/s ceiling this machine is fast enough to outrun its own part cooling before it outruns its hotend.

If layer lines degrade in the upper third only, that's a cooling-time problem: layers are returning sooner than the previous one has set. Check L14 before L15.
If the top surface goes matte and thin at a specific speed rather than a specific height, that's L15 MVS — the hotend, not the fan.
Validate at your real production acceleration. Input Shaper lets you run high accel, but corners and seams still degrade first, and the seam criterion is where you'll see it.

Core One · the chamber is part of the test

Most likely to fail: the layer-lines criterion on PLA, and for a reason that isn't in the profile at all.

Record the door and vent state on every validation print, as you did in Phase 1. A profile validated with the door open is only validated with the door open — that's the honest scope of the claim.
PLA degrading steadily upward, with rounded fine detail, is chamber heat approaching PLA's glass transition, not a cooling setting. Vent it and re-run before touching L14.
Run the dimensional test from a cold chamber and after 30 minutes of printing. Thermal expansion of bed, sheet and frame is real on an enclosed machine, and if the two measurements differ materially, that difference is a property of this machine you need on record.
PETG and TPU should do better here than on the open-frame machines. If they don't, suspect the enclosure state was wrong rather than the profile.

4The sign-off sheet

Run this against one material on one machine. Every box, or it doesn't ship.

✓ Validation sign-off — one printer, one material, one date

Phase 1 health check is current for this machine (no hardware change since)
Spool dried, and the date it was dried is recorded
Print 1 · calibration cube: X, Y and Z all within ±0.15 mm
Print 1 · X and Y agree with each other, not just with nominal
Elephant foot compensation set — tuned last, after everything else
Print 2 · overhangs clean, bridges not sagging, no stringing between towers
Print 3 · a real part of my own, at production speeds and acceleration
Criterion 1 · top surface uniform — no pinholes, no ridges
Criterion 2 · corners square — no bulges, no post-travel gaps
Criterion 3 · seam is a fine line, not a blob
Criterion 4 · layer lines even from bottom to top
Criterion 5 · snap test gives a granular fracture, not a clean layer plane
Core One only: door and vent state recorded for every print above
Config bundle re-exported and dated (Lesson 17)
Result written into tuning-log.md — pass or the step it sent me back to

Tick as you go

A partial pass is not a pass. It's a diagnosis: the unticked box names the lesson you go back to, you re-run everything below that step in the order, and you validate again. That loop is the whole method, and it terminates — which trial and error does not.

5Check yourself

From memory. Question 3 interleaves Phase 1.

Question 1 of 3

Corners bulge outward on the validation print. Which step does that accuse?

Question 2 of 3

Why is elephant foot compensation tuned at validation rather than during Phase 2?

Question 3 of 3

Your cube measures 20.00 in X and 20.28 in Y. What is that telling you?

✎ Your task · do this with your tutor

Validate one material on one machine

Pick the machine and material you've carried furthest. Run all three prints and work the sign-off sheet. Then write into tuning-log.md:

The three cube measurements, to two decimal places, and X versus Y explicitly.
Your verdict against each of the five criteria — pass, or the lesson number it sends you back to. Not "looks fine".
A photo of the fracture surface from the snap test. This is the one people skip.
What "a real part I actually print" was, and why that part is a harder test than the calibration cube for your use.
The scope of the claim, in one sentence: "This profile is validated for ___ on ___ at ___ speeds, with ___ enclosure state, as of ___."

→ Paste it back to me. If every criterion passed first time I'll be suspicious, and I'll ask to see the fracture photo before I believe it.

What you can now say

Three prints: dimensional, combo, and a real part of my own. The third is non-negotiable.
Each criterion accuses one step, so a failure is a diagnosis rather than a setback.
The snap test is the only one that measures strength — and the only one that costs a part.
Elephant foot compensation goes last, because every other setting moves it.
A validated profile has a scope: this material, this machine, these speeds, this enclosure state, this date.

📖

Primary source · read this next

Elephant foot compensation — Prusa Knowledge Base

Two minutes, with the ~0.2 mm starting figure for a 0.4 mm nozzle and the reason this setting has to come after everything else. The clearest short statement of a top-level setting in the whole KB.

💬 I'm your tutor — ask me anything. Send photos, especially of fracture surfaces and seams — those two are the hardest to judge from a written description and the easiest to judge from an image. If you're torn between "pass" and "pass, I think", that's worth a conversation. Confused beats stuck.

Linked: Glossary → validation set, elephant foot compensation, calibration order, extrusion multiplier, Pressure Advance, retraction, max volumetric speed, skew, glass transition temperature, tuning log. Keep it close with → The Calibration Card.

§References

Prusa Research. Elephant foot compensation. The Print Settings → Advanced → Elephant foot compensation path, the ~0.2 mm figure for a 0.4 mm nozzle, that Advanced or Expert mode is required, and that it is a top-level setting influenced by every other calibration and so must be tuned after them. Link.
This course. Master tuning guide, §6 — the three-print validation set and the five acceptance criteria. Link.
Printables. All-in-one 3D printer test. The combined overhang, bridge and stringing model used for print 2. Link.

Lesson L18 of the prusa-profile-tuning course · all lessons · The Calibration Card · glossary