How it works

How verification works

FITD sells verified, not AI. This page says exactly what that word is doing — what gets measured, what the report claims, and what it deliberately does not.


What "verified" means

The definition is the one in the Terms, and it is two things and no others:

  • the geometry is watertight/manifold — it is a closed, valid solid that a slicer or CAM tool can process; and
  • its dimensions match the specification you agreed in the chat.

That is the whole claim. "Verified" does not mean the part is fit for your particular application, that the material you choose is suitable, or that your print settings will produce a good result. Those judgements are yours to make.

The four steps

1. You approve a spec

Your description is read into a spec card — the dimensions, the features, the material, the process. You edit anything that is wrong and approve it. Nothing is generated and no credit is charged until you do, and the approved card is the list of numbers everything afterwards is checked against.

2. The solid is built

A CAD script is written from the approved spec and executed against a geometry kernel in a sandbox. If the script does not produce a closed, valid solid there is nothing to measure, and that is a failure at this step rather than a file you get handed.

3. The kernel measures it

The same kernel that built the solid measures it: the bounding box, the volume, and each dimension the spec stated. This is arithmetic on the solid, not an opinion about a picture of it, which is why it runs on every part on every tier.

4. The report states what it found

Every stated dimension is listed with what you asked for beside what was measured. All inside tolerance is verified. One outside it is best effort, naming the axis. Nothing measurable to compare against is not verified — never a silent pass.

The four verdicts

verifiedClosed, valid solid; every stated dimension inside tolerance.
verified after fixThe same, reached after an automatic correction — which is named in the report.
best effortA closed solid, but at least one measured dimension is outside tolerance. The report says which.
not verifiedNothing could be measured, or the spec stated nothing to measure against. Not a pass.

A worked example

This is the report for the 30 mm control knob for a 6 mm D-shaft — the part the demo on the front page ends on — measured on 2026-09-22. Where it came from is set out below, because a worked example is worth nothing if you cannot tell what produced it.

DimensionSpecMeasuredResult
Knob diameter30.0 mm29.90 mmPass
Knob height18.0 mm18.0 mmPass
Shaft bore6.0 mm6.0 mmPass
Shaft flat5.50 mm5.50 mmPass

Measured volume 10,811 mm³; thinnest wall 2.4 mm, against an FDM minimum of 1.2 mm. The diameter reads 29.90 rather than 30.00 because the knurl flutes cut fractionally into it — the report says what was measured, not what was hoped for.

What verification does not check

  • Fitness for purpose. Whether this is the right part for your job is your call. A knob measured to the millimetre is still the wrong knob if the shaft is splined.
  • Load, stress and fatigue. Nothing here simulates force. A part can be dimensionally perfect and snap the first time you use it.
  • Print settings and material. Layer height, infill, orientation, supports and filament choice all decide whether a good file prints into a good part, and none of them are visible to the measurement.

The print-readiness checks that do run — minimum wall, smallest hole, unsupported overhang — are advisory. A check that errors is reported as "not checked", never as a pass.

Do not use generated parts in safety-critical applications — load-bearing structural parts, medical devices, aviation components, braking systems — without independent verification by a qualified professional. See the Terms.

Where the worked example came from

This knob was modelled by hand with CadQuery 2.7 — the same geometry kernel FITD's pipeline generates against — from the script kept at scripts/demo/knob.py. It was not produced through FITD's own pipeline, which is why it is shown here, as an illustration of what a report says, and not on an /examples page, which is reserved for parts FITD generated.

The numbers follow the file, never the other way round: the kernel measured this STL at 29.90 × 29.90 × 18.00 mm and 10 811 mm³ (the knurl flutes take the outer diameter fractionally under 30). The 6 mm D-bore (5.5 mm across flats) is blind, 13 mm deep from the underside, leaving a solid top with the indicator slot — a through bore would show the shaft end on top of the knob, which no real knob does.

What a failed report looks like

Not every part passes, and a page that only ever shows passes is a brochure. A real failed report — a genuine generation whose measured dimensions came back outside tolerance — will be published here, with the prompt, the spec and the measurements that did not match.

It is not here yet, and nothing has been mocked up in its place: an invented failure would be exactly the kind of made-up evidence this page exists to argue against. When the failure is generated, it goes up as it came out.

Separately from failing a measurement, some requests are refused outright rather than generated — Prohibited uses lists those categories and explains how warnings, suspension and appeals work, and the Terms set out what FITD does and does not warrant.