Construction CAD, checked.
Also called: Buildability review · CAD design review · DFM check · Assembly evaluation
Checking the developed 3D and 2D CAD against how the thing is actually made: does it assemble, are the clearances right, can it be built.
The CAD looks finished on screen. This activity asks whether it can be made: does it go together, do the clearances hold, can the ceramic be cast and fired to those sections. Fix the features that fight manufacture now, before tooling money commits and the fixes get expensive.
What construction CAD evaluation is
You take the developed 3D model and 2D drawings and read them as a maker would, not as a designer would. A model renders cleanly whether or not the part can be cast, fired, machined or assembled. The screen does not care about draft angles, wall thickness, or whether two parts try to occupy the same 4mm of space. This activity is where you make the geometry answer to the process that will produce it.
It is a deliberately unglamorous read. You are not judging whether the design is good. You are asking four blunt questions of every part and every joint: does it assemble in a sane order, are the clearances real, can it actually be made by the chosen process, and is there any feature that quietly fights manufacture. Anything that fails gets a fix or a flag before the project spends another pound moving toward tooling.
Why this gets skipped
- The model looks done. A clean render reads as finished, so the team moves on. In my experience the gap between “looks made” and “can be made” is exactly where the costly surprises live.
- The maker is not in the room. Buildability is obvious to the person who casts the part and invisible to the person who drew it. If nobody asks the foundry or the board house, nobody catches it.
- It feels like tomorrow’s problem. “We’ll sort it at tooling” is a tempting line. By tooling the change costs ten times what it costs now, in a model you can still edit freely.
The buildability read, worked
The clearest way to show the read is to run it on a real part. Here is the proofing box we ran through the check, the ceramic shell with its low-voltage electronics, so you can see the shape of a good answer rather than a generic template.
Notice the read does two jobs. It confirms what already works, so you do not redesign it out of nervousness, and it isolates the two or three features that genuinely fight manufacture. Fix those, leave the rest alone.
Where teams get it wrong
The same model, read two ways. The first feels efficient and ships a part that cannot be made cheaply. The second costs an afternoon and saves a tooling cycle.
- Model renders cleanly, so the part is declared done.
- Clearances checked at nominal, ignoring firing shrinkage.
- Assembly order never walked through start to finish.
- Sharp corners and thin walls left for the foundry to find.
- Read every part against its actual production process.
- Tolerance bands set for real shrinkage and fit, not zero-gap.
- Assembly sequence walked through, fastener access confirmed.
- Maker consulted, problem features fixed while edits are free.
The screen-only read is unfalsifiable optimism: nothing has been tested against a process, so nothing can fail. The made-against read forces the geometry to earn its place, and tells you exactly which features to change.
How it fits the bigger picture
Construction CAD evaluation is activity 07.10.02 in the framework, inside Stage 07 Engineer. It reads the developed CAD that came before it, and it feeds straight into technology review (07.10.03), where the chosen processes and components get assessed against what the buildability read just confirmed or flagged.
What it can do
It catches the features that fight manufacture while they are still free to fix, in a model you can edit. It confirms the parts assemble in a sane order, the clearances survive the real process, and the ceramic can be cast and fired to the drawn sections. It turns a clean render into a part somebody can actually make.
What it can’t do
It can’t prove the chosen processes are the right ones; that is technology review’s job, next. And it isn’t a substitute for a physical prototype. The read finds what the model can show you. Some failures only appear when the first ceramic shell comes out of the kiln, which is why the firing trial still has to happen.
See the full 10-stage process →
Try it yourself
Take your developed CAD and open it next to a list of the processes that will make each part. For every part, ask the four questions: does it assemble, are the clearances real, can this process make it, does any feature fight the process. Walk the assembly order start to finish. Send the awkward parts to whoever will actually make them and ask one question: “can you make this as drawn, and what would you change?”
Want a structured first pass before you commit to tooling? Start the Free Sprint → and the GPT will help you frame the buildability questions.
Your buildability checklist
Project notes: the corner that would have cracked
▸ From the notebook · optional reading
Reading the proofing box CAD with Dan and Anna Hartley in Stockport, and the two flagged features that would have cost a wasted tooling cycle if the model had been signed off as drawn.
3 min read · click to open
The developed CAD looked finished, and the £149 box was on a tight bill of materials with no room for a wasted tooling cycle. Dan wanted to send the shell drawings to Stoke-on-Trent and the board files to Manchester the same week. I asked for a buildability read first: “Half a day. We read it as the foundry will, before anyone quotes tooling.”
What the read turned up
The cavity corner. Where the electronics cavity met the shell wall, the model had a crisp internal corner. Clean on screen, a crack risk in firing. Ceramic does not like sharp internal corners and a wall that thick around them. We radiused it. Two minutes in the model.
The fixing bosses. The base plate bosses sat close to the heated zone, in a plastic grade that would have softened over time at 26°C of sustained running plus the element’s local heat. We moved them out to a cooler ring and respecified the grade.
The shrinkage fit. The lid-to-rim fit was modelled at nominal. Ceramic shrinks roughly 12% in firing and not perfectly evenly, so a zero-gap model fit becomes a jammed or sloppy real fit. We set proper tolerance bands and checked the wood band still wrapped both extremes.
What it would have cost to skip
Every one of those was a free edit at this stage. Signed off as drawn, the cavity corner alone would likely have shown up as cracked shells out of the first production firing, which means a remade mould, a second firing trial, and weeks lost while everyone argued about whose fault it was. We pushed the change into the model instead, where it cost an afternoon.
Dan’s line afterwards stuck with me: “That render looked more finished than the thing we can actually build.” That gap is the whole reason this activity exists. The clean model and the makeable model are not the same model until somebody checks.
— Engineer stage, project notes, 2026
&mdash