Proving a candidate can build your part, not a part like it

Say what capable means for this part before asking anyone to prove it

A drawing carries dozens of dimensions and almost none of them decide whether the product works. Identify the characteristics that actually govern fit, function, safety and appearance, mark them, and verify against those. Without that step the demonstration becomes a full inspection report on everything, which is expensive, slow and oddly uninformative, because a shop can pass every dimension on a careful one-off and still be unable to hold the two features that matter in continuous running. Telling a candidate which characteristics are critical is not giving away leverage; it is the difference between a proving exercise that tests something and one that generates paper.

Samples made the way production would make them

Parts produced by a skilled technician on a prototype route prove that the geometry is achievable, not that the process is. Ask for material from the grade and supplier intended for production, the tooling or fixturing that will actually be used, the machine you would run on, and an operator rather than an engineer. Where soft tooling is unavoidable at this stage, get that stated in writing along with what will change afterwards. The gap between a demonstration part and a production part is where most launch problems live, and it is far cheaper to name the gap during assessment than to discover it during ramp.

Measurement is the capability nobody thinks to check

A candidate can only control what it can measure, and measurement is routinely the weakest link. Establish what equipment would be used for your critical characteristics, its resolution relative to your tolerance, how parts are fixtured, who performs the check, and whether any test is sent to an accredited external laboratory. Then correlate: have the same parts measured at their site and yours, and compare. Disagreement between two sets of numbers on identical parts is common, and finding it during assessment is a technical conversation. Finding it during a shipment dispute is an argument in which neither side can prove anything.

What a shop says about your tolerances tells you as much as the parts do

Listen to the technical response, not only the samples. A capable manufacturer will ask why a tolerance is where it is, propose a different datum scheme, flag a feature that will fight the process, or warn that a finish requirement conflicts with a material choice. That pushback is evidence of experience, and it usually saves money. Silence is the worrying answer: a shop that accepts every requirement without comment has either not read the specification or intends to deal with the difficult parts by sorting. Ask candidates what they would change about the design, and compare the answers across the shortlist.

Gaps you can close together, and gaps that end the conversation

Assessment usually finds something missing. The useful question is what kind of missing. Absent fixturing, an unfamiliar material, a measurement method they have not used and a documentation practice they have not needed are all closable, given time, a named owner and an agreement on who pays. A core process that does not exist in the building, a tolerance regime the equipment cannot reach whatever the effort, and a sector whose documentation expectations the management team does not recognise are different in kind. Distinguish these plainly rather than optimistically, because a conditional award against an unclosable gap simply relocates the failure to your launch.

Frequently asked questions

Can we accept samples made on different equipment from the production machine?

Only with the difference documented and a plan to close it. Machines vary in rigidity, thermal behaviour, control resolution and repeatability, so results do not transfer automatically between them. If the intended machine is not free, agree in writing which equipment produced the samples, what is expected to change on the production asset, and what evidence will be provided once it is available. What causes trouble is not the substitution itself but nobody recording that it happened.

How many parts do we need before we believe a capability result?

Enough to see variation rather than a best effort, which means parts spanning a genuine production run rather than a handful picked from the middle. Ask for pieces from the start, the middle and the end of the run, including any produced immediately after a setup or a material change, and ask what was scrapped along the way. A small set of perfect parts with no scrap record tells you the shop can select. Variation data is what tells you whether it can control.

What should we do when a manufacturer says our tolerance is unnecessary?

Treat it as a design question and answer it properly. Some tolerances exist because an engineer copied a previous drawing, and loosening them lowers cost and scrap for nothing. Others carry a function nobody documented, and relaxing them causes field failures a year later. Ask the manufacturer what the requirement costs them and what they would propose instead, then have your own engineer confirm the functional basis. Record the outcome on the drawing, because otherwise the same argument runs again at the next source.