How to get a part reviewed before you send it to a manufacturer

The most expensive part of a hardware project is the loop you don't see coming: you email a part to a print farm or a machine shop, it looks fine on your screen, and three days later a box shows up with a wall that snapped or a quote that says "this won't print as-is." Now you're re-modelling, re-sending, and re-waiting. The failure wasn't in the part. It was in the handoff.

I build Cadre, a browser CAD tool, and the handoff is the thing I think about most. Here's the honest version of how to review a part before it leaves your hands — and, just as important, how to send someone the review instead of a bare file.

Why sending a file is a lossy handoff

When you export an STL and attach it to an email, the recipient gets a bag of triangles. They don't get the thing you actually care about: whether the part is sound. Is the thinnest wall thick enough to survive? Are there overhangs that'll sag without support? Is the solid even watertight, or does a hairline crack from a boolean quietly corrupt the volume the shop is about to quote against?

You checked some of that in your head. None of it travels with the file. So the shop re-derives it — or doesn't — and you find out which on delivery.

The fix is a two-part habit: run a measured review before you send, and send the review, not just the geometry.

Step 1 — Get the part into a tool that can measure it

You can either model the part in Cadre or drop in an STL or OBJ you already have. There's a real difference between the two paths, and I'd rather be straight about it than sell you the shiny version:

• Modelled in Cadre, the part has a feature tree — named, editable parameters. A fix can be a one-line parameter change.
• Imported as a mesh, there's no feature tree to edit. You still get every measured flag, but the fix is one you make by hand (or back in whatever tool authored the STL).

Either way, the next step is the same.

Step 2 — Run the measured review

Cadre's AI reviewer reads the actual geometry and flags what tends to ruin a part. The important distinction — the one most "AI checks your CAD" tools blur — is which flags carry a real number:

• Thin walls, overhangs, enclosed voids, and watertightness are measured by the kernel. When it calls a wall thin, that's a measured millimetre (at sampling resolution), not a vibe — and a flag the measurement can't actually back is dropped server-side before you ever see it.
• Sharp internal corners and sub-resolution features are the reviewer's judgment call. They're useful, but they don't come with a measurement, and Cadre doesn't dress them up as if they do.

That line — measured versus inferred — is what makes the review something you can hand to a shop without embarrassment. There's a longer write-up on how the grounding works if you want it.

You'll also see the problem, not just read it: toggle the overhang or wall-thickness heatmap and the offending faces light up on the model, so a 62° overhang isn't an abstract number — it's the red patch on the underside of your bracket.

Step 3 — Fix the flagged number

If you modelled the part in Cadre, a suggested fix is a single parameter change you accept in one click ("wall 0.8 → 1.2 mm"), non-destructively, and you can undo it if you don't like it. If you imported a mesh, you get the measured number and the target, and you make the change by hand. Fix it, run the review again, and watch the flag clear.

Step 4 — Send the review, not just the file

This is the part that closes the loop. Once you've run a review, publish it and you get a report-only review link (/r/…). Send that to your print farm, your machinist, or a colleague, and they open it in a browser — no CAD, no account, nothing to install. They see the flags, the measured numbers, and the model each one points to; click a flag and it frames the part it points to, with the face called out.

It's a report they read, not a live session they join. Nobody's editing your model over your shoulder — the link carries the findings, and that's the point: the person quoting your part sees the same checks you did.

If they want the geometry too, export a real STEP, not just a mesh. For parts Cadre authored, the STEP carries exact analytic surfaces where it can prove them — a cylinder comes back as a true cylinder, not a ring of flat strips — with boolean-intersection seams and imported meshes staying faceted. That's a proper handoff for a quoting portal or a colleague in desktop CAD. (More on what a STEP export actually gives you.)

Where a second human comes in

Sometimes the reviewer you want isn't an AI — it's the colleague who's built this part before. In Cadre, a human reviewer's feedback lands in the same inbox as the AI's: they open a shared copy, suggest a parameter change, and you accept or discard it exactly the way you'd handle an AI suggestion. The AI is just the first reviewer who's always available; a teammate is the same mechanic with a name instead of a badge.

To be clear about what this is and isn't: it's asynchronous review — share a link, get suggestions back — not two people dragging the same model around at once. I'd rather it do one thing honestly than promise a live-multiplayer experience the tool doesn't deliver.

The habit, in one line

Model or import → run the measured review → fix the number → send the review link (and a real STEP if they need geometry). It turns a part you hope is right into a part you can show is checked, and it moves the "will this print?" conversation to before you've paid for the print.

You can do the whole loop free, in your browser, right now — Cadre is here, nothing to install.