The short answer
A Gerber package describes geometry, not intent. It says where copper is; it does not say what the copper is for. Almost every defect found at DFM review comes from that gap — geometry that is internally consistent as a drawing and wrong as a board.
Twelve defects account for most of what comes back. Four of them are worth understanding in detail, because they are the ones a designer fixes in ten minutes and a fabricator cannot fix at all.
The measure of a good package is not whether it opens. It is whether it can be built without a single question being asked.
Why files fail at review rather than at fabrication
A fabricator's process is robust to a lot of things. It is not robust to ambiguity, because every ambiguous feature has to be resolved by someone, and the person resolving it is reading your drawing rather than your design.
That is the cost structure. A defect caught at review costs an email. The same defect caught after the drill costs panels. Caught after assembly it costs a recall, and the exchange rate between those three is roughly an order of magnitude per stage.
The twelve
| Defect | What it does | Who fixes it | Caught at |
|---|---|---|---|
| Drill file and pad locations disagree | Holes off centre, broken annular ring | Designer | Review |
| Missing or mismatched drill table | Wrong tool sizes, wrong plating | Designer | Review |
| Mask opening too large | Solder bridges between fine-pitch pads | Designer | Review or assembly |
| Mask opening too small | Covered pad edges, poor wetting | Designer | Assembly |
| Silkscreen over pads | Contamination, tombstones on small parts | Designer | Assembly |
| Annular ring below process minimum | Breakout after drill tolerance | Designer or fabricator | Review |
| Line width or spacing below the process tier quoted | Price changes, or the board cannot be built | Designer | Quotation |
| Board outline not closed, or on the wrong layer | Routing path undefined, panel cannot be made | Designer | Review |
| Layer naming inconsistent with the stackup | Layers swapped, inner layers mirrored | Designer | Review |
| Copper too close to the board edge | Burrs, exposed copper after routing | Designer | Review |
| No tooling strip or fiducials on the panel | Assembly cannot place, or places badly | Designer or fabricator | Assembly |
| Impedance called out without a stackup | Target unachievable at the requested geometry | Both | Quotation |
Mask opening versus pad
This is the most common defect in fine-pitch designs, and it fails in both directions.
The mask opening is normally larger than the pad, by a small expansion that used to be generous and now has to be tight. Too much expansion and the mask web between adjacent pads disappears, so the stencil prints paste across the gap and the parts bridge. Too little and the mask overlaps the pad edge, reducing the wettable area and leaving a shadow that traps flux.
The rule is not a number, it is a relationship: the web between openings has to survive the mask registration tolerance of the process, and the opening has to clear the pad by more than the same tolerance. On 0.4 mm pitch parts, that budget is small enough that a designer working from a 2015-era library will fail it.
Annular ring
Annular ring is the copper width from the hole wall to the pad edge. It is not the pad diameter, and it is not a fixed number, because the drilled hole moves: drill tolerance, registration tolerance and material movement all shift the hole inside the pad.
A design that leaves a thin ring on the nominal drawing is asking the fabricator to hold tolerances tighter than the process allows, and the answer will be either a price increase or a breakout. Where the ring genuinely cannot be increased, the honest move is to say so and accept a documented deviation, not to leave the drawing looking compliant.
Layer naming, polarity and mirroring
A package is a stack of files with names, and the names are the only thing telling the CAM operator what each file is. Rename one file inconsistently with the stackup drawing and the inner layers can be pressed in the wrong order — a defect that is invisible until the board is cross-sectioned.
The related failure is polarity. Planes are often drawn as negatives; a positive plane file imported as a negative becomes an almost-solid copper layer, which passes every visual check and shorts everything.
Both defects are cheap to prevent and expensive to find. Send a stackup drawing that names the layers in the same order and the same words as the files.
Units, origin and the drill table
A Gerber file has no units of its own; the units are declared in the file header and often wrong. A package in mils read as millimetres produces a board 25 times too large — obvious. A package in inches read as mils produces a board 1,000 times too small — also obvious. The dangerous case is a mixed package, where the copper is metric and the drill file is imperial: everything looks plausible until the holes land in the wrong place.
State the units, state the origin, and make the drill table a separate document that the drill file can be checked against. A drill table that is embedded in a drawing as an image cannot be checked by a machine, and machine-checkable is the whole point.
If a package needs an explanation to be understood, the explanation belongs in the package. Notes on the fabrication drawing are read by a human; the Gerber is read by a machine. Where the two disagree, the machine wins, and it does not ask.
What to send, in one list
Gerbers per layer with a name that matches the stackup, a drill file with a separate drill table, the board outline on its own layer, a stackup drawing with layer order and dielectric thickness, an impedance table if controlled impedance applies, the panelisation drawing with fiducials and tooling strip, and a note stating units and origin.
None of that is unusual, and all of it removes a question.
What we can and cannot do
We run a manufacturability check on every package and report what we find before quoting, which is why the review in the DFM section exists as a step rather than a courtesy.
What we cannot do is know what you intended. We can tell you that the mask expansion will bridge at 0.4 mm pitch; we cannot tell you whether the part is 0.4 mm pitch because that is a design decision. The vocabulary problem behind several of these defects is covered in the twelve terms page, the stackup questions in the reference builds, and the review service itself on the DFM review page.