The $3,200 Mistake That Taught Me How to Actually Use Xometry

The first time my mechanic Ray answered the question "why does my brake click when I press it," I expected a one-line answer. Instead he broke open the caliper, spun the wheel, and pointed at the worn pin. "That click is information, not noise," he said. "Your car was talking to you. Most people don't listen until everything screams."

That was in March 2022. He charged me $90, and I thought the lesson ended there. But over the next two years it kept replaying in my head — especially in 2023, when I nearly lost a client for my employer, Greenline Manufacturing, because I ignored a series of increasingly loud "clicks" in a prototype aerospace order.

The whole project cost us $3,200 in avoidable waste before I finally started listening.

The background: a foot in the aerospace door

I'm a senior project engineer at a 12-person precision machine shop outside Milwaukee. We've been around since 2004. We cut decent parts for medical device companies and automotive tier 2 suppliers. Aerospace was the wall we had never gotten through.

Then, in May 2023, a regional aerospace vendor offered us a shot at a small prototype order: forty sensor housings for an attitude indicator test bench. The housing had three parts: an aluminum body, a brass insert, and a liquid silicone rubber seal. Quantities were tiny, but the follow-on potential was real — this was the classic "prove yourself on a nothing order and hope it becomes something" moment.

We didn't have all the capabilities in-house. For the brass insert and the seal features, we chose Xometry. The instant quoting engine let us upload the CAD file and get pricing in minutes. And since Xometry covers CNC, 3D printing, injection molding, and sheet metal, we could route the whole overflow to one platform instead of managing five vendors.

We also looked at the Xometry aerospace 3D printer route for the housing body — industrial FDM in ULTEM or PEEK. At the time, the quote was around $240 per part for a three-unit prototype build (as of May 2023; verify current pricing at xometry.com). But the seal lip tolerance was tighter than an as-printed polymer surface could hold, so we went CNC instead.

The platform wasn't the problem. The files I fed it were.

The first click: the file wasn't as ready as I was

I had spent about three hours on the sensor housing model. Fully defined sketches, proper GD&T, a tidy feature tree. It looked right. So I exported a STEP file and submitted it to Xometry's instant quoting engine.

I want to say I verified that file before sending it. The truth is I didn't. The rendering looked clean, and I took that as evidence of manufacturability. I had skipped the basics of CAD file preparation for manufacturing.

The next day, the quote came back with seventeen surface issues flagged. Seventeen. Unresolved edges, internal voids, a fillet that only existed as a construction line. Visually invisible in my viewer. A nightmare in CAM.

To Xometry's credit, their support team didn't just reject the file. They sent a review with specific notes on what was wrong and how to fix each issue. That document became the seed of our current pre-production checklist. Which made it worse that I didn't apply it the first time.

I resubmitted the file in a rush, fixed only the flagged geometry, and missed the fact that my reference datum was tied to a construction plane rotated six degrees off the part's natural axis. The revised quote came back $780 higher to account for reprogramming. Delivery slipped by nine days. The client was not happy.

That was click one. I ignored it.

The second click: vertical machining center spindle specifications

Once the brass parts arrived correct, we planned to machine the aluminum housings in-house. I'm an engineer. I know my machines. Or so I thought.

Our vertical machining center is a Bridgeport VMC with a 30-taper spindle, 8,000 rpm maximum, 7.5 kW motor. Perfectly fine for the conventional 2D work we do most weeks. Not fine for the slot geometry on this sensor housing — a 0.4 mm end mill with a 3:1 depth-to-diameter ratio.

The client's engineer asked the question during a routine call: "What's your high-speed machining setup?" I quoted the spec sheet. There was a pause. Apparently the slot needed a minimum of 12,000 rpm, and ideally 20,000, plus a balanced holder. ISO 16084 sets out the balance requirements for spindle tooling in high-speed machining. My little BT-30 was nowhere close.

I had never connected the geometry in the drawing to the actual vertical machining center spindle specifications of our floor. The numbers existed in two different documents. I was the person who should have connected them, and I didn't.

We routed the slot features through Xometry's manufacturing network instead. Their design analysis flagged the geometry automatically and sent it to a shop with the right equipment. It cost us six more business days.

Click two. Still waiting for the brake to fail before doing anything.

The third click: liquid silicone rubber injection molding is not a gasket

The seal on the housing was the thing that broke me.

I had spec'd liquid silicone rubber for the seal lip. Easy, I thought. Silicone is silicone. Cut it, place it, squeeze it, done.

Liquid silicone rubber injection molding is not that.

LSR is a thermoset that arrives as two components, A and B. They mix just before injection and cure through a platinum-catalyzed reaction. Depending on the grade, the mixed material has a pot life measured in hours at room temperature — sometimes less. Mold temperature, injection speed, and shot weight all control the final crosslink density and dimensional accuracy. For a seal lip with a 0.5 mm tolerance, you cannot hand-pour this mixture into a cavity and expect it to hold geometry. That's not how thermoset processing works.

I learned that the expensive way. We ordered $400 worth of LSR, tried to hand-cast it, and produced ten housings with leaky, inconsistent seal lips. Every single one failed an air pressure test. The scrap sat on a bench for three days while I tried to figure out what to tell the client.

The most frustrating part of that week: absolutely nothing about the LSR surprise was unavoidable. Xometry, and every major material supplier, publishes LSR design guides. If I had done even basic research, I would have understood that the part should have been produced via proper liquid silicone rubber injection molding with temperature-controlled tooling, not by hand-pouring. We eventually sent it through Xometry's injection molding service, which had the process control we lacked — closed-loop shot weight monitoring, controlled mold temperature, and a proper screw injection unit.

Click three. The loudest one.

When I finally started listening

The day I cleared the scrap LSR parts off the bench, I remembered Ray's brake lesson.

The worn caliper pin didn't break the brakes. It clicked. It gave you information early. The failed housings did the same thing. Every flag, every delay, every failed test — they all pointed to a specific gap. The problem wasn't Xometry, or the client's requirements, or bad luck. The problem was that I had no systematic review process that forced me to look at the part from the manufacturing process's point of view before committing.

That's where our pre-production checklist came from. It's now five pages, but the core is a set of questions:

• Have I re-imported the STEP file in a neutral format and checked for watertight geometry, datum alignment, and hidden construction features?
• Does this drawing have any feature that demands high-speed machining? If so, do our vertical machining center spindle specifications meet the requirement, or do I need to route it to a supplier?
• Is the specified material a thermoset or a thermoplastic? Does the required process match (injection molding vs. curing)? What design requirements depend on that process?
• What did the quoting engine's manufacturability review say? Did I actually read it?

We rolled it out in Q3 2023. In the 18 months since, we've caught 47 potential errors across 13 project quotes. A hole pattern rotated 18 degrees. A thread callout that matched no standard tap. A wall thickness that would have warped in every material we recommended. None of those shipped. That checklist is the best return on a $3,200 mistake I've ever gotten. (Note to self: the full checklist probably deserves its own post eventually.)

Quality is what the client sees

The original aerospace order didn't turn into a big contract. We lost money on it. But the client did give us a small follow-on order in April 2025. When I asked for feedback on our first article, they said:

"This is what we expected from the first submission."

Ouch. But they said it with a smile, and the work is real.

I'm not a materials engineer, so I can't speak to the chemistry of LSR formulations or the thermal modeling of mold cooling circuits. What I can tell you from a project engineering perspective is this: every part that reaches the client speaks for your company. They don't see our internal chaos. They see the part. When the first batch was wrong and the second batch was right, what changed wasn't the platform or the material — it was our willingness to listen to the clicks, fix the root causes, and never ship a part that we hadn't checked against the process that would make it.

This worked for us, but our situation was specific: a small shop, an understanding client, and a strong platform partner. Your mileage may vary if your context is different. And I should add that all of the pricing and process details here were accurate as of 2023–2025; verify current rates and material specs with Xometry and your suppliers before relying on them.

So the next time you're tempted to ignore a red flag on a quote, or dismiss a design review note as an inconvenience, ask yourself what the sound is telling you. Most of the time, it's a worn caliper pin, not a broken brake. Fix it now. Ninety bucks is cheaper than the alternative.