What breaks when a precision part moves between three vendors

Type II sulfuric anodize adds roughly 0.0002 to 0.0005 inches per surface. Type III hardcoat adds 0.001 to 0.002 inches per surface, sometimes more on aluminum alloys with higher copper content like 2024.

If the CNC vendor machines to nominal, and the anodize vendor doesn't know to undersize, every coated feature grows. On a pin fit at H7/h6 clearance, you've consumed the entire tolerance band before the part hits the inspection table.

The fix is upstream. The CNC vendor needs to know the coating spec before they cut metal, and they need to compensate on the toolpath. That requires the drawing, the coating spec, and the assembly intent all in one place. When CNC and finishing are quoted as two separate POs by two separate buyers, this conversation usually doesn't happen.

Anodize lines rack parts on contact points. Those contact points get masked, scratched, or left bare aluminum depending on the shop's process. If the rack point is on a non-critical surface, fine. If it lands on Datum A, you've just compromised the inspection reference.

We've seen 6061-T6 brackets come back from a finisher with rack marks on the primary mounting face, because the only place the racker could grip the part was that exact surface. The CNC vendor didn't tell the anodize vendor where the datums were. The drawing went into the finisher's queue with no rack guidance.

A correctly orchestrated job ships the part to finishing with a racking diagram that respects the datum scheme. That diagram doesn't exist in most multi-vendor jobs.

Anodize is an electrical insulator. A 1/4-20 tapped hole that gets even partial coating will not accept a screw to torque spec. The hole either needs to be masked completely, or chased after coating, or designed with a generous oversize on the tap drill to account for buildup.

Masking is not free. A typical Class 3B internal thread mask costs $1.50 to $3.00 per hole. On a part with 12 tapped holes, the masking line item is real, and the engineer specifying the part often hasn't included it on the drawing. Either the finisher catches it and the part gets re-quoted, or the finisher doesn't catch it and the part fails inspection.

A 0.005 inch deburr that nobody sees as-machined becomes a visible witness line after Type III hardcoat. The coating doesn't hide voids; it amplifies them. Tool marks, burrs, and chatter that pass QC on a bare aluminum part get rejected after coating because they're now contrasted against a dark grey finish.

Engineering drawings rarely call out a finish requirement that survives coating. "Break sharp edges 0.005 max" doesn't tell the CNC vendor that the customer expects a uniform cosmetic finish post-anodize. The CNC vendor hits the drawing. The finisher hits the drawing. The customer rejects the lot.

For defence, aerospace, and medical work, the inspection package needs material cert tied to the lot, machining inspection records tied to the lot, and process cert from finishing tied to the same lot. When the part splits across two POs and two vendors, the lot identifiers don't always travel.

We've seen a customer try to assemble an inspection package for 50 parts where the heat treat lot, the CNC lot, and the anodize lot were tracked under three different numbering schemes. The parts were fine. The paperwork was unprovable. The job got rejected by the prime's quality audit, and the customer ate three months of rework getting the documentation aligned.

A pocket with an internal corner radius smaller than one third of the pocket depth requires the CNC vendor to take multiple finishing passes with progressively smaller tools, or to switch to a smaller tool and run it deeper than is safe. Either path adds cycle time and tool deflection risk.

When the CNC quote comes in high and the engineer pushes back, the conversation usually defaults to "can you just hit it." The shop says yes, quotes the part, and absorbs the risk. Then a downstream finisher sees chatter marks in the corner and flags it. Now you've got a finger-pointing exercise about whether the corner was machinable, whether the finish was achievable, and who pays for the rework.

The rule that should have been applied at quote time: internal corner radius should be at least one third of the depth of the feature. If it's not, the engineer should know before the part is cut, not after it's anodized.

The pattern in all of these failures is the same. The spec is interpreted in pieces. Each vendor reads the drawing through the narrow window of their own process. Nobody is looking at the part as a finished assembly, because nobody is paid to.

A single PO across the full process forces a single owner of the spec. That owner:

Canada's defence industrial buildout is creating a wave of multi-process precision work for primes who need to hit 70% Canadian content. Most of that work is exactly the kind of part this post describes. CNC aluminum or stainless, secondary finishing, full lot traceability, accelerated timelines. The shops that win it will be the ones who can take a drawing on Monday and ship a fully inspected, certified part four weeks later under a single PO.

If you're a hardware engineer pricing out a multi-process part and you've felt the friction of running it across three vendors, that friction is the whole problem. It is not your project management. It is the structure of the supply chain.