Let's rewind to 2022. A simpler time. I ordered five PCBs with SMD assembly from a reputable short-run prototype house. They arrived. They worked. Compared to the breadboard, cardboard and blu tack "Machine" they were perfect. I deliberately built in lots of options for development and testing alternative voltages and an expansion port. I cracked on with development, gave the design four more years of love, and started to show TMA to the world. This PCB design is tried, tested and trusted.
This week, I needed another five. Same board. Same files. Typically I had procrastinated for longer than I should have on this re-order. Its ok I thought, its only a week to delivery. Plenty of time.
I navigated the usual Byzantine online ordering process—upload Gerbers, cross-reference BOMs, confirm footprints, question my life choices—only to be greeted at the final hurdle with a curt little notification:
"We are unable to offer SMD assembly for this design as your PCB exceeds our maximum machine dimensions."
I beg your pardon?
I re-uploaded the files from 2022. Same result. So I did the sensible thing: I reached out to support, attached the original order number, attached photos of the assembled boards sitting right there on my bench, and politely explained that they had already done this for me.
Every day is another battle
The reply? Vague. Robotic. So devoid of nuance that I'm genuinely convinced it was generated by a language model that has never so much as looked at a stencil printer. Their suggestion?
"We recommend making the board smaller to fit our current machine constraints."
Make. The board. Smaller...
Of course. Why didn't I think of that? Let me just casually shrink four years of routing, component placement, and thermal management like I'm resizing an image. Never mind that the chassis is already dimensionally optimised to within a millimetre of its life. Just... make it smaller. Top-tier advice. Really earning that support ticket.
I escalated. Spoke to a "senior" team. Same hollow, templated nonsense. At this point, I'm fairly certain their entire customer service operation is a single fine-tuned LLM running on a forgotten server in a cupboard—and I'm starting to wonder if they've been told to stop worrying and love the chatbot.
Now, some might suggest splitting the board in two and bridging with headers. And yes, technically, it's possible. But that's a non-trivial re-design, introduces interconnect headaches, consumes precious real estate for header supports, and adds two new failure points—all because their machine constraints have apparently regressed since 2022.
So, I've made the executive decision: I'm taking my Gerbers elsewhere.
This week, I've been hunting for a UK-based manufacturer. And I have some solid leads. Actual people. Real conversation. No chatbots. No "make it smaller" nonsense. Just real people who genuinely want to help..
I'll report back once I've found the one. But for now, if you'll excuse me, I'm working on a new sequencer model...
— Onwards.