A buyer-side guide to incoming inspection for screen protectors, glass back covers and lens parts — defect classes, sample sizes, what to measure, and the wording that makes a claim stick.
Almost every dispute in this category comes down to one sentence that was never written: what counts as a defect, and how many are allowed. Suppliers say "we inspect every piece" and mean something different from what the buyer hears. Buyers say "quality was bad" and cannot point to a clause. This guide is about closing that gap before the first shipment — with defect definitions, a sampling plan you can defend, and inspection criteria that survive contact with a real factory.
1. Does Your Supplier Really Inspect Every Piece?
"100% inspection" is one of the most commonly misunderstood claims in the glass accessories trade. In most factories it means every piece passed under a light and was looked at by an operator — not that every piece was measured, tested for adhesion, checked for transmittance, or verified against a drawing.
That is not dishonesty. At volumes measured in hundreds of thousands per month, dimensional measurement of every unit is economically impossible, and the industry does not pretend otherwise. The problem is that the phrase is ambiguous, and ambiguity is expensive when a container arrives with a defect rate you did not expect.
There are really three different activities hiding behind the same words, and a serious supplier will tell you which one they perform:
| Activity | What it actually covers | What it does not cover |
| 100% visual sort | Every unit seen under a lamp; obvious chips, cracks, heavy contamination removed | Dimensions, curvature, coating performance, latent bubbles |
| 100% functional check | Every unit tested against a device or a jig for touch, sensor clearance, fit | Optical performance, adhesion strength, long-term coating wear |
| Sampling inspection | A statistically defined sample measured against every specification, with accept/reject numbers | Nothing about the unsampled units — that is the point of the statistics |
The third one is what you actually want in a purchase order, because it is the only one that produces a number you can act on. "We check everything" gives you no recourse. "Sample 200 pieces per lot, accept at 5 major defects, reject at 6" gives you a decision rule.
For the product side of that specification — the dimensions and options a finished high-definition tempered glass protector is actually built to — the product page is the reference the inspection plan should quote, not the other way round.
2. What Are Critical, Major and Minor Defects?
Defect classification is the foundation of every sampling plan, and it is almost always the part that is missing from a Chinese-English quotation. The three-class convention is standard across consumer goods, but the contents of each class must be written for your product, not copied from a template.
Critical defects create a safety risk or make the product unusable. For a screen protector that means a crack that could propagate, glass that was never strengthened, or a cut-out that completely blocks a sensor. Critical defects are normally accepted at zero, regardless of the AQL applied to the other classes.
Major defects affect function or are obvious to the end customer. A chip at the edge, a bubble in the viewing area, a black border that overlaps the display, a contact angle out of specification, a transmittance reading below the agreed minimum. These are what drive returns and one-star reviews.
Minor defects are cosmetic and do not affect use: a slight colour variation in the printed border, a tiny inclusion near the edge, a scuff on the retail box. Minor defects are where a reasonable AQL keeps the price sane, because rejecting a whole lot over one dust speck is not quality management, it is a negotiating tactic.
| Class | Examples on a glass protector | Typical acceptance |
| Critical | Through-crack; unstrengthened glass; sensor opening fully blocked; sharp edge that cuts | Zero — reject the lot |
| Major | Edge chip above the agreed size; bubble in the active area; border overlap; low transmittance; coating missing | AQL 0.65–1.0 |
| Minor | Slight border colour shift; inclusion at the extreme edge; retail box scuff | AQL 2.5–4.0 |
Notice that every example above is measurable. "Bad edge" is not a defect definition; "edge chip longer than 0.3 mm or deeper than one-third of the glass thickness" is. The whole inspection system depends on this translation, and it is the single highest-value thing a buyer can spend a day on.
3. Which AQL Should You Set?
AQL stands for acceptance quality limit, and it is widely misunderstood as "the percentage of defects we are willing to accept". That reading is not quite right and it leads to bad arguments. AQL is a property of the sampling plan: it describes the quality level that the plan will accept most of the time. It is a statement about risk, not a permission slip.
In practice, most accessory programmes land in a narrow band. 0.65 is used where a defect is expensive to discover downstream — retail programmes with marketplace penalties, or branded lines where the same glass carries a logo. 1.0 is the common default for major defects on consumer accessories. 2.5 suits minor cosmetic issues and high-volume, low-price lines. 4.0 appears in commodity channels where the customer expects a certain amount of variation.
Two cautions are worth more than the numbers. First, do not apply one AQL to everything: a plan that treats a blocked sensor opening the same way as a box scuff is not protecting you. Second, do not tighten the AQL as a negotiating move unless you are prepared to pay for it. Lower AQL means more rejects, which means higher unit cost, longer lead time, or a supplier who quietly stops honouring the plan.
4. How Does the Sample Size Get Decided?
Sampling is not a matter of opinion, and it should not be left to the inspector's judgement on the day. The recognised framework is ISO 2859-1, published in China as GB/T 2828.1 and in the United States as ANSI/ASQ Z1.4. All three describe the same basic mechanics: lot size plus inspection level gives a sample size code letter, and that letter plus AQL gives the accept and reject numbers.
For most accessory shipments, General Inspection Level II is the appropriate starting point — it is the level the standards are built around, and it is what a third-party inspection company will default to. Under that level, the lot size determines the sample:
| Lot size (pieces) | Code letter | Sample size | Practical reading |
| 501 – 1,200 | J | 80 | One carton-level check; fast but coarse |
| 1,201 – 3,200 | K | 125 | Typical trial order |
| 3,201 – 10,000 | L | 200 | Common for a full production lot |
| 10,001 – 35,000 | M | 315 | Container-load quantities |
Two practical details matter as much as the table. The sample must be drawn from across the lot — different cartons, different positions in the pallet — not from the three boxes the factory set aside. And the accept and reject numbers must be read from the actual table in the standard edition named in your contract; they are not something to reproduce from memory or from a blog post, including this one. The value you get from this section is knowing that the table exists and that it must be fixed in writing before production, not after a dispute starts.
5. What Should You Actually Measure?
A sampling plan is only as good as the checklist attached to it. For a glass protector, nine attributes cover most of the risk, and each one maps to a specific complaint you would otherwise hear from a customer three months later.
| Attribute | How to check it on the sample | Complaint it prevents |
| Outline dimensions | Caliper or vision, at least three points per side | Overhang past the case, or a gap at the edge |
| Curvature / sag | Height gauge on a surface plate, or a master fit check | Edge hover and light leak |
| Border width | Optical comparator at four points | Sensor masking, or a visible adhesive line |
| Cut-out position | Vision against the drawing datum | Camera or proximity sensor obstruction |
| Edge chips | 10× loupe, measure length and depth | Cracks starting at the edge after installation |
| Bubbles / white edge | Visual under defined illumination, 24 hours after lamination | Delayed bubbles the customer sees first |
| Transmittance and haze | Spectrophotometer on the finished, laminated part | A protector that looks grey or milky on the device |
| Contact angle | Goniometer, initial and after the agreed wear cycle | Coating that disappears in weeks |
| Functional fit | Install on a real device: auto-brightness, ear-proximity, front camera | A part that measures perfectly and still fails |
The last row is the one most inspection plans omit, and it is the one that catches the most expensive mistakes. A protector can pass every dimensional check and still blind the ambient light sensor. Functional fit is cheap to test — one device, five seconds per sample — and it is impossible to argue with afterwards.
Two of the nine attributes depend on construction rather than on the glass alone. A full-glue HD part and an edge-glue 2.5D part will not return the same transmittance or the same contact angle, so the inspection plan has to name which construction it is inspecting. Otherwise you are comparing a measurement to a specification written for a different product.
6. How Do You Define a Chip or a Bubble?
"No chips" and "no bubbles" are not specifications, because no process achieves literally zero and no inspector can apply an absolute rule consistently. Both defects have to be described with a number, a location and a counting rule.
For chips, three parameters are enough: length, depth as a fraction of glass thickness, and location relative to the edge. A chip of 0.2 mm at the outer edge of a part that will sit under a case is a different risk from the same chip on the inside of a sensor opening. Define both, separately.
For bubbles, the rule needs diameter, count and zone. A common and workable structure is: no bubble exceeding 0.3 mm anywhere; no more than two bubbles between 0.1 and 0.3 mm per unit; none at all within the active display area. Written that way, two inspectors looking at the same part will reach the same conclusion, which is the only test of a good definition.
One more variable is easy to forget: when you look. Lamination defects can take hours to appear, as residual stress relaxes and air migrates along the adhesive interface. Inspecting ten minutes off the press will pass parts that fail tomorrow. Build a 24-hour hold into the process, or accept that a percentage of the lot will develop defects in the customer's hands.
7. What About Tests You Cannot Do on Every Piece?
Some of the most meaningful specifications are destructive. A steel ball drop, a cross-hatch adhesion test, a salt spray run, a 3,000-cycle rub for the oleophobic coating — none of these can be performed on a unit you intend to ship. They belong in a separate plan, and mixing them into the AQL count is a common and costly mistake.
The right structure is two documents. The lot acceptance plan covers non-destructive attributes and produces the accept/reject decision for the shipment. The periodic qualification plan covers destructive tests at a defined frequency — per lot, per shift, per month, or on change of glass substrate — and produces evidence that the process has not drifted. Keep them separate, and state the frequency explicitly, because "we test regularly" is not a frequency.
For the destructive tests themselves, insist on the conditions rather than the result. A drop test without a stated height, support and pass criterion is not repeatable. A contact angle without the wear medium and load is not comparable between suppliers. The number is only meaningful if someone else could reproduce it.
Our own impact line shows what auditable data looks like: the anti-shatter film specification pairs a 64 g steel ball drop with a contact angle measured both before and after 3,000 rubs. Two numbers, two conditions, both repeatable — which is exactly what a periodic qualification plan needs in order to mean anything.
8. Where Do Most Inspections Fail?
Not in the sampling table. In the conditions around it. Five failures account for most of the disputes we see, and all five are cheap to fix before the shipment leaves.
Lighting. Inspecting dark glass under warehouse lighting hides everything. A defined inspection station — a light box or at least a bright, diffuse source at a consistent angle — is the difference between finding a hairline and shipping it.
Time. As above, inspecting immediately after lamination misses delayed defects. The 24-hour hold is not bureaucracy; it is the only way to see what the customer will see.
Sample selection. A sample drawn from the boxes the factory chose is not a sample. It is a presentation. Pull units from sealed cartons at different pallet positions, and note in the report where they came from.
Measurement on the wrong object. Measuring bare glass and calling it the finished part hides the effect of the adhesive and the coating. Transmittance, haze and thickness all change after lamination.
No record. A defect photographed without a scale, a batch number or a date cannot support a claim. By the time the argument starts, the evidence is gone.
9. How Should the Inspection Report Be Written?
A report is not a formality. It is the document you will rely on if the lot has to be reworked, discounted or returned, and it should be complete enough that a third party could review the decision a year later.
| Field | Why it must be there |
| PO / SKU / drawing revision | Ties the report to the exact specification that was agreed |
| Lot size and lot number | Defines what the decision applies to |
| Standard, level and AQL used | Makes the accept/reject numbers reproducible |
| Sample size and sampling locations | Proves the sample was not hand-picked |
| Defect count by class | Shows which clause failed, not just "quality issue" |
| Photographs with a scale | Turns an opinion into evidence |
| Result and disposition | Accept, reject, sort, rework — with responsibility stated |
| Retained samples | The only defence when the two sides remember different goods |
10. What Makes a Claim Enforceable?
Six clauses, all short, and all of them belong in the purchase order rather than in an email thread. Together they turn a quality discussion into a decision procedure.
One: the defect definitions. Critical, major and minor, with measurable thresholds. Without this, nothing downstream is arguable.
Two: the standard, inspection level and AQL values by class, plus the edition of the standard being used. "AQL 1.0" alone is incomplete without the level.
Three: when inspection happens. At the factory before packing, after a 24-hour hold, or on arrival. Each option creates a different burden of proof.
Four: who inspects. Supplier, buyer's representative, or an independent third party. If the buyer can send someone, say so; it changes behaviour before the fact.
Five: the remedy. Sort and replace, price reduction, or return. State the proportion at which each applies, and who pays freight.
Six: retention. How long signed samples are kept by each side. Ninety days is common; whatever you choose, both parties must keep them for the same period.
11. Should You Inspect at Source or on Arrival?
Both, for different reasons, and the choice changes what you can prove. Source inspection is the only point at which you can see the lot before it is packed, influence sorting, and reject without paying freight. Its blind spot is that it happens in the supplier's environment, on the supplier's equipment, with the supplier's people.
Arrival inspection sees the goods after transport and storage — the conditions that actually cause damage. Its blind spot is attribution: once the carton has crossed an ocean, every defect becomes arguable. Was the chip there at packing, or did it happen in the container?
Third-party inspection buys independence and a standard report format, at the cost of a day rate and scheduling lead time. It is most valuable on the first two or three lots of a new programme, when the process is still being proven, and least valuable once the relationship is stable and the data is boring.
Whichever route you choose, the packaging standard has to be checked at the same time. A perfect lot in inadequate packaging becomes a defective lot in transit — carton strength, internal separation, desiccant and drop performance are part of the acceptance decision, not a separate conversation. Our glass component packaging and shipping guide sets out the tests worth requiring before a container is loaded.
12. What Changes for Back Covers and Lens Parts?
The same sampling logic applies, but the defect list is different, and using a protector checklist on a back cover will miss the failures that matter. Three attributes move to the front.
Flatness. A back cover that warps will not sit down on the frame, and the resulting gap is both a dust path and a bonding failure. Measure on a surface plate, and define the maximum gap under a straight edge rather than saying "flat". The requirement is tighter on a slim wireless-charging back cover, because there the coil gap forms part of the electrical specification rather than being a cosmetic concern.
Hole position and wall quality. On a drilled or laser-cut cover, the position of the hole relative to the datum matters more than its diameter, and the wall condition determines whether a crack starts there. Inspect the exit side as well as the entry side; the two are not equivalent.
Colour and coating. Decorated covers need a colour measurement, not a visual judgement, and a ΔE limit agreed against a signed master. Adhesion belongs in the periodic qualification plan, not the lot plan.
Lens parts add a fourth: concentricity and inner diameter. A camera lens ring that is dimensionally correct but off-centre will vignette one corner of the image, and no dimensional check on the outer profile will catch it. Inspect against the camera axis, not the part outline.
13. Five Inspection Mistakes
Writing the plan after the first complaint. By then the lot is gone, the samples are lost and the supplier has no reason to agree to a tighter rule.
Using one AQL for every defect class. It either rejects good lots over cosmetics or accepts bad ones with a blocked sensor.
Accepting "100% inspected" as a specification. It describes an activity, not a criterion, and it transfers no risk to the supplier.
Testing bare glass instead of the finished part. Adhesive and coating change the optical and mechanical result; the coupon is not the product.
Not keeping signed samples. Without a physical master, "the colour is different" is an opinion, and opinions do not survive a freight-forwarder dispute.
14. Frequently Asked Questions
Is AQL 1.0 stricter than AQL 2.5? Yes — a lower AQL accepts fewer defects, so the plan rejects more lots at the same quality level. It also costs more, because more lots are sorted or rejected.
Can I just require zero defects? You can require zero critical defects, and you should. Requiring zero of everything is not a sampling plan; it is an instruction to inspect 100% and reject unpredictably, which usually ends in either a price increase or an unenforced clause.
Do I need a third-party inspection for a trial order? Not necessarily, but the first lot of a new programme is where the process is least stable and the cost of a bad surprise is highest. If you cannot attend, a third party is cheap insurance relative to reworking a container.
How long should samples be kept? Long enough to cover the customer's return window plus your own claim window. Ninety days is a common floor; longer is better for branded programmes.
What if the supplier refuses the sampling clause? That is information. A supplier who will not accept a defined plan is telling you how the first dispute will go, and it is worth knowing before the tooling is paid for.
Setting Up Inspection for a New Order?
Tell us the product, the channel and the defect risk you care about. We will send back a defect classification, a sampling plan and the measurement method for each attribute — written so that both sides can check it the same way.