PCBA First Article Inspection Report: Which Rows Are Release Criteria and Which Are Just Records
The report everyone files and nobody reads twice
First article inspection is one of the few steps in a PCBA order that almost every buyer asks for and almost no buyer specifies. The purchase order says "FAI required before mass production." The supplier builds a small quantity, measures some things, fills a form, attaches photos, and sends a PDF. Somebody on the buying side opens it, sees no red cells, replies "approved, please proceed," and the file goes into a folder.
Three weeks later the second or third production batch arrives and something is wrong. A connector sits at a slight angle. A silkscreen mark is missing on one variant. A board edge has a burr that snags in the enclosure. The buyer goes back to the FAI report expecting it to settle the argument, and discovers it does not settle anything. The report recorded what that particular sample looked like. It never said that those observations were the standard the rest of the order had to meet.
That gap is not a paperwork accident. It is what happens when a document gets requested by name without anyone defining what the document is for. An FAI report can do three quite different jobs:
- A feasibility record. Proof that the design can be built at all, with the files as submitted.
- A release baseline. A set of measured and observed values that later batches must reproduce within agreed limits.
- A change reference. A frozen snapshot of the process state, used to detect whether anything drifted later.
The same PDF can serve all three, or none of them. Which one you get depends on terms that have to exist before the build, not after the dispute. On PCB168's PCB assembly page the inspection plan is listed as "Agreed per project" and test requirements as "Customer-defined / reviewed" precisely because these are negotiated inputs, not defaults that appear on their own.
Two documents that are easy to confuse
Before going further it helps to separate FAI from the thing it is most often mistaken for.
A first article inspection report is about a *sample from a specific build*, produced under a specific process setup, against a specific revision of your files. Its subject is that build.
An inspection and test plan is about *the whole order*: what gets checked, at what frequency, by what method, against what limits, on every batch. Its subject is the ongoing arrangement.
The FAI report can become the reference point that the inspection plan compares against. But it cannot replace the plan, and the plan does not automatically adopt the FAI numbers as limits. If you only have the FAI, you have a photograph. If you only have the plan, you have rules with no calibrated starting point. Defining the plan itself is a separate exercise, covered in defining inspection and test requirements before you get quoted; this article is about the effect the FAI report has once that plan exists.
Every row in an FAI report has one of three statuses
Open any FAI report and you will find a table. Dimensions, component presence, polarity, solder appearance, markings, cleanliness, maybe electrical results. What the table almost never tells you is the *status* of each row. There are only three possibilities, and mixing them is the origin of most FAI disputes.
Status 1: Binding criterion. The row states a limit, the limit was agreed before the build, and later batches must fall inside it. If a batch falls outside, that is a nonconformance and the buyer can act on it.
Status 2: Recorded observation. The row states what was measured or seen on this sample. No limit was agreed. Later batches can differ and the supplier has broken nothing, because nothing was promised.
Status 3: Reference-only attachment. Photos, screenshots, process settings, material lot identifiers. Useful evidence for diagnosing a later problem. Not a criterion by itself, and not a promise that the same settings will be used again unless a change-control term says so.
Most FAI reports as delivered are almost entirely Status 2 and Status 3. That is not the supplier acting in bad faith. A supplier who writes limits nobody asked for is inventing your requirements, which is worse. The absence of Status 1 rows is the natural result of nobody having specified any.
The practical consequence: when you approve an FAI report, you are not converting Status 2 rows into Status 1 rows. Approval means "this sample is acceptable, proceed." It does not retroactively turn every observation into a guarantee. If you want binding rows, they have to be marked as binding before the build, and the marking has to live in a document the order references.
Which rows are worth making binding
You cannot make everything binding. A report where every row is a limit takes longer to produce, costs more, and invites arguments about measurements that never mattered to your product. The useful question is narrower: which observations, if they drifted on a later batch, would actually hurt you?
Four families usually deserve Status 1, and the reason is always about consequence rather than about how easy it is to measure.
Fit-critical dimensions. Anything that has to mate with something else: board outline, mounting hole positions and diameters, connector positions relative to the edge, overall height including tall components, and edge condition where the board slides into a slot or a gasket. These are the rows where a small drift produces an assembly that will not go together, and where the buyer has no workaround at receiving. Edge condition in particular tends to fall through the cracks because it sits between fabrication and assembly responsibility; if depaneling method affects your fit, the board edge and depaneling acceptance criteria belong in the binding set too.
Build configuration identity. Which BOM line went into which designator, including any approved alternate. This is the row that decides whether a later batch is even the same product. A board with a functionally equivalent but differently sourced part may pass every dimensional and visual check and still fail your qualification, your customer's audit, or your regulatory file.
Named workmanship class and its application. Not just "IPC-A-610" as a word, but the class you are buying and the specific conditions you care about on this design. A generic standard reference with no class and no applicability statement is closer to Status 2 than Status 1, because two competent inspectors can apply it differently to the same joint.
Whatever your customer or regulator will ask you to evidence. If you have to demonstrate traceability, a records deliverable is a criterion, not a courtesy. It needs to be binding in the same way a dimension is.
Rows that usually work fine as Status 2: cosmetic observations with no functional effect, process values you are not qualifying, and measurements taken for information because someone had the fixture set up anyway. Leaving them as records is not sloppiness. It is how you keep the binding set small enough to defend.
The revision question, which decides everything else
An FAI report is only meaningful against a named file set. This is the single most common failure in practice, and it is invisible until it matters.
A PCBA build consumes several documents that revise independently: the BOM, the fabrication data, the placement or centroid data, the assembly drawing, and any test or programming instructions. On PCB168's assembly page the quotation input is listed as "BOM + fabrication + placement data," which is exactly the set that has to be pinned. These files do not revise in lockstep. A BOM gets a new line because a part went end-of-life. An assembly drawing gets a clarifying note. Nobody reissues the whole package for either change.
So an FAI report that says "inspected and conforming" without naming revisions has recorded a conclusion about an unidentified subject. Later, when a batch is disputed, the conversation goes in circles: the buyer says the FAI proved it could be built correctly, the supplier says the FAI was against the file set current at the time, and neither party can point to what that set was.
What to require instead, as a single block on the report's first page:
- Every consumed document by name, revision identifier, and date.
- The quantity built for the FAI, and whether it came off the production route or a separate setup.
- Any deviation, waiver, or engineering question that was open at build time, listed rather than silently resolved.
- The process state that the supplier considers reproducible, distinguished from what was one-off for this sample.
That last point matters more than it looks. If the FAI sample was hand-placed and hand-soldered while the production batches will run the automated route, the FAI has told you the design is buildable and has told you almost nothing about what the batches will look like. That is a legitimate way to run a first article, but you need it stated, because it changes what the report can be used for. The assembly process guide walks through the route stages where this distinction shows up.
The four things "approved" can mean
When a buyer signs or emails approval of an FAI report, at least four different meanings are in circulation. Naming yours removes an entire class of later argument.
Meaning A — Feasibility accepted, nothing frozen. "You can build this. Proceed." No baseline. Every batch is judged only against the inspection plan, if one exists. This is what most unqualified approvals actually amount to.
Meaning B — Sample accepted as representative. "This sample is acceptable, and later batches should look like it." Sounds stronger, and is still weak, because "look like it" has no limits. It gives the buyer a rhetorical position, not an enforceable one.
Meaning C — Marked rows become binding. "The rows flagged as criteria are now the release baseline for this order. Batches outside them are nonconforming." This is the version worth having, and it requires the flagging to have happened before the build.
Meaning D — Configuration frozen under change control. Meaning C, plus: the process state, material sources, and approved alternates recorded in the report cannot change without written notice and re-approval. This is appropriate for regulated or safety-relevant products and costs more, because it constrains the supplier's flexibility on sourcing and routing.
Most disputes are two parties acting in good faith while holding different meanings. The buyer assumed C or D. The supplier delivered A. Neither lied. The PO never said which one was being purchased.
State it in one line on the PO: *"FAI approval under this order constitutes Meaning C: rows marked CRITERION in the agreed FAI template become the release baseline for all subsequent batches."* You can write it in your own words. What matters is that it is written.
Alternates, which is where FAI most often quietly breaks
A first article can pass every check and still stop representing your product, and the usual cause is a component substitution.
The mechanism is ordinary. The FAI is built with the parts on hand. A later batch runs when one of those parts is on allocation. The supplier finds an equivalent, and depending on your terms either flags it or treats it as routine. The new batch is dimensionally identical, visually identical, and electrically passes whatever test was agreed. It is also, from the standpoint of your qualification file, a different product.
Whether this is a problem depends entirely on what you wrote down. Two terms do the work:
Which BOM lines are locked. Not the whole BOM, which is unrealistic and expensive to enforce. The specific lines where substitution is not acceptable without re-approval: parts in the safety path, parts your firmware depends on for timing or behaviour, parts named in a certification, parts whose supplier is specified by your own customer.
What counts as approval for the rest. Who on your side can approve an alternate, what evidence they need, and whether approval applies to one batch or to the order. Without this, you get either a supplier who blocks and waits (schedule damage) or a supplier who proceeds and informs you afterwards (configuration damage).
The FAI report is where the as-built configuration gets captured, so it is also the natural place to record which lines were locked and which alternates were already approved. The mechanics of running the approval loop itself are covered in turnkey PCBA BOM risk and alternate part approval. PCB168's assembly page lists component supply as "Defined in RFQ" for the same reason: who sources what is an input, and substitution authority travels with it.
Who has to prove what when a batch is disputed
This is the part that decides whether the FAI report was worth requesting.
Suppose batch three has a problem. Two burdens of proof exist, and your documents determine which way they point.
If the disputed attribute is a Status 1 row, the buyer's obligation is limited: show the measurement, show the agreed limit, show the deviation. The supplier then has to explain and remedy. The argument is about facts and cost, not about whether a standard existed.
If the attribute is a Status 2 row, the burden inverts completely. The buyer now has to argue that something unstated was nonetheless required — by implication, by industry norm, by the fact that the FAI sample was different. Sometimes this argument succeeds commercially, because suppliers value the relationship. As a contractual position it is weak, and it consumes weeks.
Three supporting terms keep this from becoming a standoff:
Measurement method, not just limit. A dimension with a tolerance but no stated method invites two correct measurements that disagree. Name the method, or at minimum name who re-measures when results conflict.
Document precedence. You will end up with a PO, an FAI report, an inspection plan, a drawing, and a standard reference, and at least two of them will eventually conflict. One line stating which governs saves a long exchange. The same precedence problem, viewed from the receiving dock, is worked through in the incoming inspection acceptance plan.
Named adjudicator and retained samples. Who decides, and against what physical reference. Retaining one approved first article sample on each side is cheap and ends an entire category of argument, because there is something to compare to rather than a PDF to interpret.
A PO clause block you can adapt
Short enough to actually get used. Adjust the bracketed parts to your product.
> First Article Inspection and Release
>
> 1. Supplier shall perform FAI on [quantity] units from [the production route / a setup identified in the report] before releasing subsequent batches.
> 2. The FAI report shall name every consumed document with revision and date: BOM, fabrication data, placement data, assembly drawing, and test or programming instructions.
> 3. Rows marked CRITERION in the agreed FAI template are the release baseline for all subsequent batches under this order. Unmarked rows are records and create no obligation.
> 4. Criterion rows for this order are: [fit-critical dimensions list], [named workmanship class and applicable conditions], [as-built configuration for locked BOM lines], [records to be delivered].
> 5. Measurement method for each criterion row shall be stated. Where results conflict, [named party] re-measures and that result governs.
> 6. BOM lines [list] are locked; substitution requires written approval from [role] before build. Other substitutions require [notification / approval] per [scope].
> 7. Any deviation or open engineering question at FAI build time shall be listed in the report, not resolved silently.
> 8. Both parties retain one approved FAI sample for the duration of [period].
> 9. Order of precedence on conflict: [PO] > [drawing] > [inspection plan] > [FAI report] > [standard reference].
>
> Lead time, test availability, equipment, certifications, capacity, and traceability deliverables are per the written quotation and engineering response.
That last line is not filler. It keeps the clause from implying capability commitments that were never quoted, which is the same discipline PCB168's assembly page applies when it states that no fixed values appear on the page and that these items are confirmed in the written quotation.
What this changes in practice
Two shifts, both small.
At RFQ, the FAI stops being a checkbox and becomes a scoped deliverable. A supplier who has to mark criterion rows and state measurement methods is quoting a defined piece of work. That may cost slightly more than an undefined FAI. It is also the version that has value later, and it makes quotations comparable, which is the same problem discussed in PCBA quote comparison and scope boundaries.
At dispute, the conversation changes shape. Instead of arguing about whether a requirement existed, both sides look at a marked row, a method, and a number. Most of these disputes then resolve in days rather than weeks, not because anyone became more reasonable, but because there is less to interpret.
Neither shift requires a quality system, an audit, or a new standard. It requires deciding, before the build, which rows you intend to rely on. If you are preparing an assembly RFQ now, the quote request form and capabilities overview are the places to attach these terms, and the PCB quality inspection methods guide covers the methods you may want to name in the criterion rows.
FAQ
Is an approved FAI report a binding release baseline?
Only if the order says so. By default, approval usually means the sample was acceptable and production may proceed. Rows in the report are records of what was observed unless they were marked as criteria and agreed before the build. If you need a binding baseline, state it in the PO and mark the rows.
Which revision does an FAI report apply to?
Whichever file set was consumed at build time, which is why the report must name each document with its revision and date. BOM, fabrication data, placement data, and assembly drawings revise independently, so a report that says "conforming" without naming revisions has recorded a conclusion about an unidentified subject.
Do we need every measurement to be a criterion?
No, and it works against you. A large binding set costs more, takes longer, and produces arguments about values that do not affect your product. Make binding only what would hurt if it drifted: fit-critical dimensions, as-built configuration for locked BOM lines, the named workmanship class with its applicable conditions, and records you will have to evidence downstream.
Can a supplier change components after FAI approval?
That depends on your substitution terms, not on the FAI itself. Name the BOM lines that are locked, state who on your side can approve an alternate and what evidence they need, and state whether approval covers one batch or the order. Without those terms the FAI captures the as-built configuration but does not restrict later sourcing.
Who proves a batch is nonconforming?
If the attribute is a marked criterion, the buyer shows the measurement, the agreed limit, and the deviation, and the supplier responds. If it is an unmarked record, the buyer has to argue that an unstated requirement applied, which is a much weaker position. That asymmetry is the practical reason to mark rows in advance.
How is this different from an inspection and test plan?
The FAI report concerns one sample from one build under one process state. The inspection plan concerns what is checked on every batch, at what frequency, by what method. The FAI can become the calibrated reference the plan compares against, but it does not replace the plan, and the plan does not automatically adopt FAI values as limits.
Should the FAI sample come off the production route?
It is better if it does, and it is not always possible. Either is workable as long as the report states which it was. A hand-built sample demonstrates the design is buildable; a production-route sample also tells you something about what batches will look like. The distinction changes what you can rely on the report for.
Related links
Need PCBs for Your Project?
Get an instant quote — prototypes from $2, delivered worldwide.
Get Instant Quote