PCB Fabrication Specifications in Conflict: Which Document Wins

Conflicts come in three shapes, and only one of them looks like a conflict

Buyers tend to picture specification conflict as two numbers that visibly contradict. That shape exists, and it is the least dangerous of the three, because it is the only one someone is likely to notice.

Shape 1: direct contradiction

Two documents state the same attribute with different values. The surface finish named in the purchase specification is not the one called out in the drawing notes. The copper weight in the stackup table differs from the copper weight in the spec sheet. The drawing says one thing about a tolerance and the spec sheet says another.

These get caught reasonably often, because any careful engineer reading both documents in the same sitting will trip over them. They get caught inconsistently, though, because catching them depends on one person happening to read both. When the spec sheet comes from procurement and the drawing comes from the design team, often nobody reads both.

Shape 2: silence, filled by default

No document addresses the attribute at all. The package says nothing about whether a particular class of defect is acceptable, nothing about the via protection method, nothing about edge condition, nothing about how a marking is to be applied. The fab needs a value to build, so it uses one — its house default, or the most economical reading of the governing standard, or what it did on your last order.

This shape is more dangerous than direct contradiction for a simple structural reason: there is nothing on the page to notice. You cannot review your way to finding it by comparing documents, because the attribute appears in none of them. It surfaces only when the boards arrive and something about them is not what you pictured, and then the conversation is about whether you ever asked.

Shape 3: same quantity, two different conventions

Both documents address the attribute. Neither is wrong. They express it in ways that are not directly comparable, so two readers extract two different requirements from the same package.

A thickness stated as a finished value and a thickness stated as a pre-process value are the same attribute under two conventions. A tolerance stated as a symmetric band and a tolerance stated as a one-sided limit can describe overlapping but unequal intents. A requirement written as "per the applicable standard" and a requirement written out explicitly in the notes may or may not land in the same place, depending on which revision of the standard the reader has in hand and which class is assumed.

This is the shape that survives review longest. Both documents look correct, a reviewer comparing them sees two statements that are not literally contradictory, and the disagreement only becomes visible when someone has to turn the package into a travel sheet and must pick one convention. Nothing is flagged, because nothing looks flaggable.

Why the fab's own order is not a problem you can fix by picking a better fab

Every fabricator resolves conflicts. They have to; the board cannot be built from an ambiguity. So each one carries an internal precedence habit, usually unwritten, usually sensible on its own terms, and usually different from the next one's.

Some read the fabrication drawing as the governing document and treat the purchase specification as commercial background. Some treat the data files as truth, on the reasoning that the data is what gets built. Some treat the most recent email as the live instruction. Some escalate every discrepancy to a question, which is the best behaviour and also the slowest, and it is the behaviour that makes a supplier look worse during a quote race.

Three consequences follow, and all three are predictable.

Your quotes stop being comparable. If the package is ambiguous and three suppliers resolve it three ways, their prices describe three different products. The cheapest number very often belongs to the supplier whose internal default happened to be the least expensive reading, which is a fact about their defaults, not about their efficiency.

Your second source diverges from your first. Two suppliers can both build exactly to the package, both be fully compliant, and ship boards that differ in a way that matters to you. Nobody breached anything. The package permitted both.

Your own rework claim gets weak. When boards arrive built to a reading you did not intend, the question becomes what the package required, not what you meant. If the package genuinely supported the fab's reading, you are negotiating a favour rather than asserting a right.

The precedence ladder: rank the documents, not the attributes

The fix is small and structural. Instead of chasing every attribute to make all five sources agree — which is real work, never finishes, and breaks again at the next revision — you state once, in the RFQ, the order in which the sources govern.

A workable default ladder, from highest authority to lowest:

1. Written clarifications issued after the package, but only those formally incorporated (see the next section). Most recent governs.
2. The purchase specification, as the buyer's contractual statement of what is being bought.
3. The fabrication drawing and its notes, as the engineering statement of how the part is defined.
4. The data files — Gerber, drill, netlist, stackup — as the geometric definition.
5. The referenced standard, at a named revision and class, as the backstop for anything the four above leave unaddressed.
6. Nothing else. Supplier house defaults govern only where the five above are all silent, and where that happens the supplier must say so rather than choose quietly.

The exact order is yours to set, and reasonable buyers order it differently. A buyer whose drawings are the authoritative artefact and whose spec sheets are boilerplate should put the drawing above the specification. What matters far more than the ordering is that an ordering exists and travels with the package. A declared ladder you would revise next quarter beats an undeclared one indefinitely.

Two provisions make the ladder actually work:

Geometry is a carve-out. Where the data files and a drawing disagree on physical geometry — pad shapes, hole positions, outline — the data files should usually govern regardless of their position in the ladder, because the data is what drives the tooling. Say this explicitly as an exception rather than leaving it to inference.

The ladder does not authorise silent resolution. The top of the clause should require the supplier to report any conflict it resolved using the ladder, in the quote or in the first-article documentation. The ladder exists to keep the job moving, not to make disagreements invisible. A conflict resolved and reported is a correction you can make cheaply; a conflict resolved and unreported is the same correction after tooling.

Email clarifications are the most authoritative and least controlled part of your package

The ladder puts written clarifications at the top, and that is correct — a clarification issued after the package is the most current expression of intent. It is also where most packages quietly fall apart, because a clarification sent as a reply in a thread has none of the properties a governing document needs.

A clarification buried in an email thread is not visible to the next reader. It does not reach the other suppliers who received the same package. It cannot be found again in six months when the reorder goes out, which means the reorder is built from the original package and the clarification silently expires. It frequently has no stated scope, so nobody knows whether it applies to this order only or to the part permanently. And when two threads with two engineers conflict, there is no revision order to resolve them.

So the clause needs a mechanism, not just a rank. Three provisions carry most of the weight:

Nothing governs until it is incorporated. State that clarifications take effect only once reflected in a revised document or a numbered clarification log attached to the package. This sounds bureaucratic and is the single highest-value sentence in the clause, because it converts "we discussed it" into a binary that both sides can check.

One clarification, all candidates. During quoting, every answer goes to every supplier still in the running, in the same words. Answering one supplier's good question privately is how you make their quote the odd one out and then hold it against them.

Every clarification states its scope. This order only, or permanent part definition. Without that line, the reorder question is unanswerable and gets answered by whoever is on duty.

The cheapest working implementation is a numbered clarification log — a short appended sheet, one row per item: number, date, attribute, decision, scope, and the document it will fold into. It does not need a system. It needs to be one artefact that travels with the package instead of living in five inboxes.

Where the cost of an undeclared order actually lands

The argument for spending twenty minutes on a precedence clause is not tidiness. It is that the cost of skipping it does not appear while you are quoting. It appears later, at predictable points, by which time it has grown.

At quote. Hidden, and the most expensive stage to lose. Spreads you read as commercial competitiveness are partly resolution differences, and the undeclared ambiguity systematically rewards whichever supplier took the cheapest reading. You cannot see this in the numbers, and you make the award decision anyway.

At first article. Visible and still cheap. The boards differ from your expectation; you find out whether the package supported the reading. This is the last stage where a correction is a document change rather than a cost event.

At incoming inspection. Expensive, and now a dispute over entitlement rather than a correction. Your incoming inspection and acceptance plan can only reject against stated requirements — if the attribute was never specified, or was specified two ways, the plan has nothing to measure against and the material tends to get accepted under protest.

At reorder. The one that recurs. The original package is reissued, the clarifications that lived in email are gone, and the part changes without any change having been requested. This is the specific failure that the incorporation rule above prevents, and it is why that rule is worth more than the ordering itself.

At second source. Structural. Two compliant suppliers, two different products, and the only honest conclusion is that the package permitted both.

Writing it: a clause, not a program

This does not need a document control system. It needs a short clause in the RFQ, a stated revision and class for every standard you reference, a named owner for conflict decisions, and a clarification log.

Put the ladder in the RFQ, not only in the spec sheet. It has to be visible before the quote, because its job is making quotes comparable. A precedence rule discovered after award has already missed its purpose.

Name the revision and class of every standard you cite. "Per the applicable standard" is Shape 3 in one phrase: it looks like a requirement, resolves differently depending on which revision and class the reader assumes, and nobody notices the gap. An unversioned reference is a blank the supplier fills in.

Name one owner for conflict decisions. A supplier that reports a conflict needs somewhere to send it. If reports route to a shared mailbox, the fast answer wins and the fast answer is often whoever has least context. One named engineer, with a backup, beats a distribution list.

Require conflicts to be reported, not just resolved. Covered above, and worth repeating as the item most often dropped when the clause gets shortened.

Scope it to where it pays. Packages with one source per attribute, simple geometry and a standing relationship can run on habit. The ladder earns its keep where multilayer and HDI builds carry stackup and via requirements across several documents at once, where heavy copper work makes dimensional conventions genuinely ambiguous, where regulated end markets require you to show which document governed, and wherever a second source has to reproduce a first source's result.

It also pays wherever one attribute is already known to be contested. Where a specific parameter has a recurring ownership problem — who declares the build class, who owns the test and inspection definition in a PCBA requirement set — the precedence ladder is what keeps the answer from being re-litigated in each new package, and those parameter-level articles are the right companions to this one: they settle a single attribute, this settles the order of the documents.

If you are preparing a package now, the practical move is to add the ladder and the incorporation rule to your RFQ submission, state revisions on your standards, and send the same clarification to every candidate. For a package that already has several documents in disagreement and a decision due, our engineering team can review the set and identify the conflicts before it goes out — reach us here. Broader document-package guidance for complex builds sits in the HDI PCB manufacturing buyer hub, and a summary of what we can and cannot resolve from a package is on our capabilities page.

FAQ

Which document should rank highest — the purchase specification or the fabrication drawing?

Either works, and the choice should follow which one your organisation actually maintains. If your drawings are the authoritative engineering artefact and your spec sheets are mostly boilerplate, put the drawing on top; if procurement's specification is the negotiated contractual statement, put that on top. Pick based on which one you would defend in a dispute, then write it down. The ordering matters less than its existence.

Should the Gerber data outrank the drawing?

For physical geometry — pad shapes, hole locations, board outline — yes, usually, because the data drives the tooling and a drawing redlined later rarely propagates back into it. For everything the data cannot express, which is most of a specification, the drawing and spec sheet have to govern. The clean way to write this is a general ladder plus an explicit geometry carve-out, so the exception is stated rather than inferred.

Is a precedence clause enforceable if the supplier never acknowledged it?

Treat enforceability as a question for your own contracts people, not for a supplier's quote email. What the clause reliably does, independent of that, is remove the supplier's defence that nobody told them the order, and give every candidate the same resolution rule so their prices describe the same product. The practical value is in alignment before award rather than in leverage after it.

What if the supplier flags a conflict we did not know existed?

That is the system working, and it is worth saying so rather than treating it as friction. Answer it, record it in the clarification log with a scope, fold it into the governing document at the next revision, and send the same answer to every other candidate still quoting. A supplier that reports conflicts is cheaper than one that resolves them quietly, even when it looks slower during a quote race.

Do we need this for simple two-layer boards?

Usually not. Where one document covers each attribute, geometry is simple and the relationship is established, habit handles it. The clause earns its cost once a package spans multiple documents, a standard is referenced without a revision, a second source has to match a first, or a part is reordered long enough after the original that the email thread is gone. The reorder case is the one that catches buyers who correctly judged they did not need it the first time.

We send clarifications by email already. Why is a log better?

Because an email is not findable by the next reader, does not reach the other candidates, has no stated scope, and disappears before the reorder. The log fixes those four properties and nothing else. One numbered sheet attached to the package — number, date, attribute, decision, scope, target document — is enough; the point is that it travels with the package rather than living in individual inboxes.

Sources: 本篇为采购文件治理方法论，不含任何 PCB168 自有能力的数值主张、认证编号、设备或客户案例。文中"三种冲突形态"与"优先级阶梯"为本文提出的整理框架，非行业既有标准条文；提及标准与认证时只按通行做法措辞，不引条款号与条款内数值，并建议读者在自己的文件包中标明所引标准的版本与等级。GSC 需求数据来自站点自有 Search Console 近 90 天导出。

Cover image alt: Off-white cabinet unit on a green epoxy floor with a small control panel and a red-bordered sign at the top left, one of its tinted glass doors open to show a castered metal rack holding a stack of green boards, with a seated person at a monitor in the background

---

Related links

• how a fabrication specification sheet becomes binding on a quote