Quality assurance: the work done before the first part exists

A claim about conditions, not a verdict on parts

The distinction that matters practically is what the evidence is about. Control evidence concerns an object that already exists: this bar, this batch, this shift. Assurance evidence concerns the conditions under which objects will be made: the fixture holds the part the same way every time, the oven reaches temperature across the whole tray, the operator can perform the operation unsupervised, the gauge can resolve the tolerance. Conditions can be proven once and monitored cheaply thereafter. Parts have to be judged one by one. That asymmetry is the entire economic argument for putting money in early.

The readiness list that earns its keep

A short list beats a long one. Is the drawing revision formally released and is that the revision on the floor? Has the process been run at production settings and shown to hold the characteristics that matter? Are the gauges specified, available and proven able to discriminate at the tolerance in question? Have the named operators been signed off by someone who watched them work, rather than ticked off after a briefing? Is there a written reaction for the failure everyone expects? Five honest answers are worth more than a checklist of forty items answered by the person who wants the launch to happen.

Where assurance quietly stops being work

Decay follows a familiar path. Documents get approved without being read because approval is a queue to be cleared. Training records record attendance rather than competence. A process is proven once at launch and then improved informally, so the evidence describes a plant that no longer exists. The tell is a mismatch between the folder and the floor: settings written on a scrap of card taped to the machine, an operator using a personal gauge, a fixture packed out with shim. None of those are dishonesty. They are what happens when the paperwork stops keeping pace with production and nobody is funded to close the gap.

Accepting somebody else's assurance for a bought part

Most of what a factory ships was partly made elsewhere, so much of your assurance is really an assessment of theirs. The workable question is narrow: for this characteristic, on this part, what will you accept as proof, and what will you verify yourself. Options run from a certificate of conformity, through a supplier's own results for the actual lot, to your own measurement on receipt. Deciding that once per part and writing it down prevents the drift where a certificate is filed unread for years, and it separates cleanly from broader supplier management, which is a purchasing discipline rather than an inspection one.

Why the budget for this gets cut, and how to defend it

Assurance spending is visible, early and easy to postpone; the losses it prevents are late, diffuse and land on somebody else's cost centre. That asymmetry is why the fixture is ordered after launch rather than before. The defence is arithmetic rather than principle: for the specific part, estimate what a containment at the customer would consume in freight, sorting labour, engineering attention and lost orders, then set that against the cost of proving the process first. Do it before the programme is approved, when the number can still change a decision, not afterwards when it merely allocates blame.

Frequently asked questions

Is quality assurance simply documentation?

Documentation is the residue of assurance, not the substance of it. The substance is a set of demonstrations: the process was run and held, the gauge was shown capable of the measurement, the operator was watched doing the job. Records exist so those demonstrations can be pointed at later without repeating them. A folder assembled without the underlying demonstrations proves nothing, which is why auditors ask what the evidence shows rather than whether a document exists.

Who owns assurance in a plant with no quality department?

Usually the manufacturing engineer or the owner, and the important thing is that it is one named person rather than a shared aspiration. Small plants do this well when the same individual signs off the process, the gauge and the operator, because the connections between those three are where problems hide. What fails is spreading the tasks across production, engineering and purchasing without anyone holding the whole picture, since each will assume a gap belongs to somebody else.

How do we tell whether assurance work actually paid off?

Compare what launches cost you now with what they cost before. The signals are concrete: hours spent sorting in the first weeks of production, engineering time consumed after start of production, concessions raised in the launch period, and complaints traced to a cause that was known and untreated. If those fall while the parts get harder, the money is working. If they stay flat, the assurance activity is probably ceremonial and should be redesigned rather than expanded.