PCB Edge Quality Inspection: Writing a Sampling Plan a Rejection Survives

A rejection is only as strong as the sampling behind it

The scene is familiar. A lot of bare boards arrives. Someone in receiving or quality opens the top package, looks at a handful of boards, finds something on the edges or the surface that does not look right, and the lot gets put on hold. You write to the supplier. The supplier replies that their own outgoing inspection passed the lot, asks which boards you looked at, and suggests the ones you found are not representative.

At that point the conversation is no longer about the boards. It is about whether your look at them counts.

This is the part buyers consistently underestimate. Incoming inspection disputes are rarely won or lost on the finding itself. They are decided by whether the way the sample was obtained was agreed in advance by both sides. If it was not, the supplier has a free and entirely reasonable objection available: you pulled whatever was on top, you looked at an unspecified number of boards, you applied a threshold nobody had written down, and you are now asking them to take back a whole lot on that basis.

The finding can be completely correct and still fail to hold, because the procedure that produced it was invented after the problem appeared. A sampling plan is what makes the finding portable: it turns "we looked and did not like it" into "we executed the agreed procedure and the agreed procedure says this lot does not pass."

Everything below is about putting that procedure in place before you need it.

The four questions a sampling plan has to answer

A usable plan is short. It answers four things, and a plan missing any one of them leaves an opening that a supplier can legitimately use.

How many boards get looked at, and against what base. Not just a count, but what the count is derived from. A number pulled per shipment behaves differently from a number pulled per manufacturing lot, and those two are frequently not the same thing: one shipment can contain several production lots, and one production lot can arrive across several shipments. A plan that says a quantity without saying "per what" is ambiguous exactly when it matters, because the natural reading for the buyer maximizes the sample and the natural reading for the supplier minimizes it.

Where the boards come from within the lot. This is the element most often left out, and the one suppliers push on hardest. Taking the top boards of the top package is convenient and is the weakest possible position, because the top of the stack is not a random draw from the lot. A plan has to state that the sample is drawn across packages and across positions within packages, and it has to say who performs the draw.

What the threshold is, and where the threshold is written. The plan itself does not need to carry acceptance limits. What it needs is an unambiguous pointer to the document that does: the drawing, the specification sheet, the agreed standard, the fabrication notes. The failure mode here is a plan that says a lot fails if defects are found, without saying which document decides what a defect is. That turns every inspection into a negotiation about definitions.

What happens when the sample fails. Does the lot go back entirely, does it get screened one hundred percent, does the supplier get an opportunity to sort, is partial acceptance available? A plan that only produces a pass or fail signal, with no agreed consequence, produces a hold and then a phone call, which is the situation you were trying to avoid.

If you write nothing else, write these four. They cost a paragraph in the purchase documents and they remove most of the ground a dispute stands on.

Why edge and surface findings are the ones that end up contested

Not every kind of finding generates this argument. Electrical results and dimensional measurements tend to settle quickly, because both sides can repeat the measurement and get the same answer. Edge condition and surface appearance are different, for three structural reasons.

They are distributed unevenly across a lot. Edge quality is produced by mechanical operations whose condition changes gradually through a run, and surface condition depends on handling and on position within a stack. A finding that is real on some boards and absent on others is exactly the finding where "your sample was not representative" is a plausible objection rather than a stalling tactic.

They are judged partly by comparison, not purely by number. A great deal of edge and surface assessment comes down to whether what you are looking at is within what the governing document allows, and the governing document often describes rather than dimensions it. Two competent inspectors looking at the same board can land on different sides of a descriptive limit in good faith. When the criterion is partly comparative, the credibility of the sample carries more weight, not less.

They are affected by what happened after the boards left the supplier. Edges get knocked and surfaces get marked in transit and in your own receiving area. A supplier is entitled to ask whether a finding originated with them. The only answer that holds is a sampling and documentation procedure that was already running before anyone had a motive.

The practical consequence: the more your acceptance depends on edge and surface criteria, the more the sampling procedure needs to be written down rather than assumed. This is also why these findings are worth photographing as they are drawn, by position, before the boards move anywhere.

Who pulls the sample

The plan should name a function, not a person, and it should recognize that there are only a few workable arrangements.

Your own receiving or quality staff pull it. The ordinary case. It is the most defensible when the plan specifies the draw method, and the least defensible when it does not, because the supplier's objection lands directly on the part you improvised.

The supplier pulls and documents the sample before shipping. Useful in that it creates a record from the lot while it is still intact, and limited in that it is the supplier's own draw. It does not replace your draw; it gives you something to compare against, which is valuable when a dispute turns on whether a condition existed at shipment.

A third party pulls it at origin. Appropriate when the lot is large, the destination is distant, or the acceptance criteria are the kind that are expensive to argue about after the goods have moved. The thing to settle in advance is not only who the third party is, but whose procedure they execute and who receives their report.

Both sides pull jointly after the dispute starts. This is the fallback, and it is worth noting that it only works if the lot has been preserved. Which leads directly to the part most plans omit.

Preservation: the clause that decides whether a retest is possible at all

Here is the sequence that quietly destroys more claims than any inspection error.

A lot arrives. Receiving pulls a sample, finds something, flags it. The production schedule does not stop for this, so the usable portion of the lot goes into the line. Someone sets aside the boards that looked bad. Two weeks later the supplier asks to see the lot. What exists is a handful of rejected boards, no record of where in the lot they came from, and no remaining population to compare them against.

The claim is now very hard to support, not because it was wrong but because the evidence was consumed.

So the plan needs a preservation element: on a hold, the lot stays segregated and intact, identified by package and position, until the disposition is agreed or an agreed time limit expires. Two details matter in how this is written.

The first is that the time limit has to exist and has to be the buyer's protection, not an open-ended obligation. An indefinite hold is unworkable for anyone with a schedule, so the clause should say what happens if the supplier does not respond within the agreed window — typically that the buyer may proceed with the stated disposition.

The second is that preservation has to cover the boards that passed, not only the ones that failed. The passing population is what establishes that the finding is distributed rather than isolated, and it is the first thing a supplier will ask about.

Retest: whose equipment, whose sample, whose cost

When the supplier contests the finding, a retest follows. Three things get argued about, and all three can be settled in advance in a few lines.

Whose equipment. The supplier will prefer their own, you will prefer yours, and neither is unreasonable. What matters less than people assume is the equipment itself; what matters is that both sides agreed to it before the result existed. The workable arrangements are: the supplier retests and you receive the raw records rather than a summary; a named third party retests; or both retest and a disagreement between them escalates to a third party. Choosing badly in advance beats choosing well after the fact.

Whose sample. This is the one that produces genuine deadlock. If the supplier retests their own retained boards, they are not testing the lot you received. If they retest the boards you rejected, you have shipped away your evidence. The resolution most often used is a split: the drawn sample is divided, each side holds a portion under seal, and a third portion is reserved for a third party if it comes to that. The split and the sealing have to be part of the draw, which is why the draw procedure and the retest procedure cannot be written separately.

Whose cost. Retest cost, return freight and third-party fees are small relative to a lot but large relative to goodwill, and they are the thing that makes a buyer hesitate to reject marginal material. The conventional structures are: cost follows the outcome, meaning the party whose position is not upheld bears it; cost is shared regardless; or each side bears its own and only the third party is allocated. Any of these is fine. Silence is not, because silence means the party with more patience wins.

Destructive sampling: a separate agreement

Some questions about a bare board cannot be answered without cutting it. Anything about internal structure, plating in a hole, or a cross-sectional condition consumes the board that answers it. That consumption is not a detail, and it should not be improvised mid-dispute.

The sacrificed boards have to be accounted for. They are no longer deliverable, which means either the lot is short against the purchase order or the shortage is accepted. Say which in advance.

Who selects them is a decision with consequences. A destructive sample selected from the boards that already look worst answers a different question than one drawn at random, and both are legitimate questions. The plan should state which one it is asking.

Whether coupons substitute for product. Many lots travel with test coupons intended exactly for this. A coupon is far cheaper to sacrifice and is not the same object as the board, and the extent to which a coupon result is accepted as a result about the lot is a matter of prior agreement, not of argument afterward. Settle it when you settle the sampling plan.

Cost and replacement. Destructive work on a contested lot usually also implies a replacement obligation if the finding is upheld, and no replacement if it is not. This is the same cost-follows-outcome structure as retest, applied to consumed material.

Putting it in the purchase documents rather than the quality manual

A sampling plan that lives only in your internal quality system binds you and not your supplier. To be usable in a dispute it has to be referenced in the documents the supplier accepted. In practice that means a short clause in the purchase order or the terms behind it that does three things:

• Points to the plan by document name and revision, rather than restating it. Plans get updated; purchase orders should not have to be reissued because of it, but the revision in force at order placement should be identifiable.
• States the preservation window and the consequence of no response, because this is the element that has a clock attached and the clock has to start somewhere.
• Allocates retest and destructive-sample cost and samples, in one sentence each.

What this is not: an attempt to make the supplier agree in advance that your inspection result is final. No supplier will accept that, and it is not what makes a rejection hold. What makes a rejection hold is that the procedure behind it was mutually known. A supplier who agreed to the procedure has much less room to object to a result the procedure produced.

If you are assembling this for the first time, the natural order is to settle the acceptance criteria first — which document governs what a defect is — then the sampling plan, then the retest and destructive clauses. Doing it in the other order produces a careful procedure pointed at an undefined target. Our note on what an incoming acceptance plan needs to cover deals with the first of those, and the overview of inspection methods covers what each kind of check can and cannot tell you.

Scope note

This article describes commonly used commercial and procedural arrangements between buyers and bare-board suppliers. It states no sampling quantities, no acceptance limits, no defect rates and no costs, and it makes no claims about PCB168's own inspection equipment, laboratory facilities, staffing or capability. Published standards are referred to by name only; nothing here reproduces or interprets their limits. Sampling procedures, acceptance criteria and cost allocation are ultimately governed by the written agreement between the parties and by applicable law. This is not legal advice.

Where to go next on PCB168

• PCB quality inspection methods — what each category of check is able to establish, which is what your acceptance criteria should be anchored to before you write a sampling plan.
• Incoming inspection acceptance plan — which items belong in the plan at all; this article assumes those are already settled.
• Cosmetic defects, repair and disclosure — how repaired and cosmetically marginal material is categorized and disclosed, which is frequently the subject of the finding your sample produces.
• PCB manufacturing process guide — background on where in fabrication edge and surface conditions originate.
• Capabilities and request a quote — if you want acceptance and sampling arrangements discussed as part of the order rather than after it.

FAQ

Do I need a formal statistical sampling scheme, or is a written procedure enough?

A written procedure that is mutually known is the thing that makes a rejection hold. A recognized statistical scheme adds a defensible basis for the quantity and the draw, which matters more as lots get larger and as the cost of a wrong disposition rises. Many buyers reference a published scheme for the structure and then specify the remaining elements — draw location, preservation, retest, cost — themselves, because those are the parts a sampling standard does not address.

The supplier says my sample came from the top of the stack and is not representative. Are they right?

As an objection to the procedure, usually yes, and that is separate from whether your finding is real. The top of a stack is not a random draw, and conditions related to edges and surfaces are known to vary by position. The fix is procedural rather than argumentative: specify the draw across packages and positions, record which positions were drawn, and the objection stops being available.

Who should hold the rejected boards while the dispute runs?

The buyer, segregated and identified, with the passing portion of the lot preserved as well. Returning the rejected boards before the disposition is agreed removes your ability to demonstrate anything. If the supplier needs to examine them, a split sample with each side holding a sealed portion solves this without either party giving up the evidence.

Can the supplier insist on retesting with their own equipment?

They can propose it, and it is not an unreasonable position. The question to settle in advance is not which equipment but what you receive from it: raw records rather than a pass statement, and an agreed escalation to a third party if the two results disagree. Equipment identity matters less than record access and a tiebreaker.

If we cut boards to investigate, who pays for them?

Settle it with the sampling plan, not afterward. The usual structure is that consumed material is replaced if the finding is upheld and borne by the buyer if it is not, with the same logic applied to third-party fees. Also decide whether test coupons can be sacrificed instead of product, and whether a coupon result is accepted as a statement about the lot.

How long should a lot stay on hold?

Long enough for the supplier to respond and no longer. The clause that makes this work is not the duration but the stated consequence of silence — that after the agreed window the buyer may proceed with the stated disposition. Without that, a hold becomes indefinite and the practical pressure falls entirely on the party with a build schedule.

Does any of this change for assembled boards?

The structure carries over, but the stakes and the destructive options change substantially, because a populated board contains components whose value and traceability are separate questions from the board itself. If your scope includes assembly, treat the sampling arrangement for PCBA as a separate agreement rather than an extension of the bare-board one.