Anyone else skip the assembly workbench and just link parts together?

I ran loose and strategic for about three years on my CNC router build, maybe 90 parts across 30 files, all positioned by hand against a common origin. It worked until I changed one rail spacing and spent a whole evening nudging fifteen parts back into place. That was my breaking point. The manual approach is fine right up until you need to move a parent part and everything welded to it in your head but not in the model.

What I do now is somewhere in between. I still keep my parameters in one spreadsheet file and link it everywhere, same as you, but I group parts that physically bolt together into small clusters. Each cluster gets positioned once. So a bracket with its four bolts is one unit, and the assembly is just clusters at the top level. It's maybe 20 moves instead of 90 and my brain can hold that.

I never fully learned the dedicated assembly tools either, and honestly for printing parts one at a time like you're doing, I question whether you need them. You're not checking for collisions between moving links on screen, you're slicing joints individually. The one thing I'd borrow from proper assemblies is just the habit of constraining each cluster to one reference point so nothing drifts when you edit geometry upstream. The real test for me is a question: if I changed one dimension right now, how many files would I have to reopen? If the answer is more than three or four, my structure has gotten too loose.

Funny, I went the other direction and it stuck. I tried learning the Assembly workbench on a delta printer build a couple years back and bounced off it hard. Felt like I was doing bookkeeping instead of modeling, and half my printed parts needed clearances anyway that made the 'exact' assembled positions kind of fictional. So I went back to loose files but with one change that saved the whole approach for me: every position is an expression bound back to a master spreadsheet, not typed in by hand. Change the arm length in the spreadsheet and the carriage, the idler bracket, and the belt path all follow on recompute. I never position anything by hand anymore, I just never position anything at all. The catch is naming discipline. My first attempt at this collapsed beacuse I had twenty expressions pointing at cells named like Spreadsheet2.Set11.B3. Now every parameter gets a real name and a comment, and I keep a plain text file listing what drives what.

For your robot arm I'd guess the joint angles are the real question, not the part positions. If you bind each joint's placement to a spreadsheet cell too, you can type in angle values and watch the whole arm move in the 3D view. Crude animation, but it found two interference problems on my build that I never whould have caught eyeballing static positions.

For what it's worth, the hand positioning thing gets old fast once you want to check the arm's reach. I built a similar arm last year and just made a throwaway copy where I spun each joint by editing one angle number, wich caught a cable routing collision I never woud've seen otherwise. Took twenty minutes.

Also, don't sleep on naming your origin sketches something obvious. When everthing's 'sketch' and 'sketch001' the manual approach turns into archaeology real quick.

My break came from fit, not structure. I modeled a pan tilt head for a camera and every joint shell printed either sloppy or cracked until I bought a cheap digital caliper and started measuring the first print of each fit, then typing the real numbers back into the spreadsheet. One tool ended three rounds of reprinting. Now the model holds a target fit and the measured print tells me how far off reality runs. Do you measure your printed joint holes after printing, or has it been consistent enought at your sizes that the spreadsheet numbers just hold?

Gotta push back on the throwaway copy trick. I did exactly that on a gantry project, spun the joints around, found my issues, felt smart. Then two weeks later I fixed a real part in the main file and the copy silently went stale. Next time I trusted it, it showed me a clearance that didn't exist anymore. Copies don't get edits, they just get older.

If you want to check motion, I'd put the joint angles in the shared spreadsheet instead and rotate the actual parts off those numbers. Same twenty minutes, but it stays true because there's only one model.

One thing that bit me on a 40-part SCARA build: external links are only as strong as your file names. Renamed one base file three months in and half the tree went red. Now every file gets a number prefix in build order and the name never changes once it's linked, no matter how ugly it gets. Also worth deciding early wich file owns the origin. On mine each joint had its own and aligning them later cost me a weekend.