LED Facade Acceptance Criteria: Writing a Dead-Pixel and Uniformity Clause That Holds

By the MEKALEO engineering team — 22 years of LED facade production since 2004

Procurement guide · November 2026

LED Facade Acceptance Criteria: Writing a Dead-Pixel and Uniformity Clause That Holds

Most acceptance clauses are written to be signed, not to be enforced. “Dead pixel rate below 0.01%” sounds precise until the first container lands and both sides discover they were measuring different things. This is how to write a clause that survives contact with a real screen.

Reading time 12 min Written for procurement, QA & site engineers Applies to mesh, grille and media facade screens
4
elements every enforceable acceptance clause must contain
100×
how much looser a 0.01% clause is than normal commercial practice
1 pixel
what 3–5 ppm actually allows on a 100 m² P25 screen
The short version

An acceptance clause needs four things: what is counted, how it is measured, under what condition, and what happens when it fails. Remove any one and you have an intention, not a clause.

AQL applies to a lot, not to each module, and not to a percentage of pixels. Quoting “AQL 1.0” without the sampling standard, the inspection level and the defect classes is not a specification.

Pixel clusters and bright (stuck-on) pixels are not statistical defects. They are visible defects, and they need a separate zero-tolerance rule — AQL sampling was never designed to catch them.

The four elements of a clause that holds

An enforceable acceptance criterion fixes four things before production starts: the object being measured, the method used to measure it, the condition under which the measurement is taken, and the consequence of failing. A clause that omits any of the four will be renegotiated on site, at the worst possible moment.

验收条款的四个必备要素 对象、方法、条件、后果四项缺一不可,缺任一项条款就无法执行。 Object what is counted — full pixel, sub-pixel, cluster Method how it is sampled — ISO 2859-1, level, lot Condition test pattern, warm-up, distance, ambient light Consequence rework, concession, reject, who pays A clause missing the fourth element is the one that ends in a meeting. A clause missing the first three is the one that ends in an argument about what “dead” means.
Figure 1. Four elements, in order of how often they are omitted. Consequence is left out most, object second.

Why “dead pixel rate below 0.01%” always ends in a dispute

It is the most common clause in the market and one of the weakest, for five separate reasons.

The denominator is ambiguous. 0.01% of what? Of all pixels, of all modules, of all cabinets, or of the delivered area? On a 100 m² P25 screen the answer changes the permitted defect count by orders of magnitude.

It is roughly a hundred times looser than normal practice. 0.01% is 100 parts per million. Commercial-grade LED screens are typically specified at 3–5 dead pixels per million, and premium grades at 1 per million. A buyer who writes 0.01% believing it is strict has written a clause that permits around twenty to thirty times more defects than the industry norm — and nobody at the table will point this out.

It says nothing about clusters. Three adjacent dead pixels is a visible black spot in the middle of a logo. Three isolated dead pixels across a 160,000-pixel screen is invisible. A percentage treats them as identical.

It does not distinguish dark from bright. A dark pixel disappears into most content. A stuck-on bright pixel stays visible against dark scenes and reads as a fault to every viewer. They are not the same defect.

It does not define the inspection condition. Inspected from 30 cm in a dark room, any screen fails. Inspected at the contractual viewing distance under normal ambient light, most screens pass. The condition is not a detail; it is the clause.

The arithmetic that changes the conversation

A 100 m² P25 screen contains about 160,000 pixels. At the commercial norm of 3–5 per million, the entire screen would be permitted 0.5 to 0.8 dead pixels — in practice, zero or one. That is why the industry samples and classifies rather than applying a pure rate, and it is why a rate alone, written without the sampling rule, tells you nothing about what you will actually accept.

Classify the defects before you count them

Do not use “dead pixel” as a catch-all in a contract. Five distinct conditions behave differently on a facade, are repaired differently, and should be judged differently.

五类像素缺陷的示意与判定 全死像素、子像素失效、常亮像素、像素簇、间歇性缺陷在同一网格中的位置与判定等级。 13 × 7 sample grid — five defect types Full dead pixel Major · counted per screen Dead sub-pixel Minor in most uses, major in close view Stuck-on (bright) pixel Major · most visible at night Cluster — 3+ adjacent Critical · zero tolerance, always Intermittent / flicker Major · retest under load
Figure 2. The same grid, five conditions. Circle marks a single defect; the dashed box is the cluster rule; the dashed dot is intermittent. Only two of the five behave statistically — the other three need explicit rules.
ConditionWhat it isWhy it matters on a facadeSuggested class
Full dead pixelWhole RGB pixel does not lightReads as a black dot; invisible at distance in isolationMajor
Dead sub-pixelOne channel fails; pixel shows wrong colourProduces colour fringing on white content; noticeable close upMinor to major by viewing distance
Stuck-on pixelPixel permanently bright, often white or one colourFar more visible than a dark pixel against night contentMajor
ClusterThree or more adjacent defectsVisible as a blot from normal viewing distance; a logo killerCritical, zero tolerance
IntermittentFlicker or colour shift under load or heatAppears only after hours of operation, so it escapes a short factory checkMajor, retest after burn-in

Putting AQL into the contract without getting it wrong

AQL — acceptable quality limit — is a sampling system, defined in ISO 2859-1. It answers one question: given a lot of a given size, how many defective units may I find in a sample before I reject the lot. It does not answer “how many dead pixels may this screen have”, and using it that way is the second most common mistake after the percentage clause.

Four traps are worth naming in the specification itself:

  • AQL is applied to the lot, not to each module. Applying AQL 1.0 to every cabinet would mean accepting 1% of pixels dead in every cabinet, which is far stricter than the standard intends and would fail almost any batch.
  • AQL 1.0 does not mean 1% dead pixels. It means a 1% rate of non-conforming units in the sample is the acceptance threshold, judged on the defect classes you defined. The number is meaningless without the classes.
  • The inspection level and sampling scheme must be stated. “ISO 2859-1, General Inspection Level II, single sampling, normal inspection” is a usable reference. “AQL 1.0” is not.
  • Sampling cannot see clusters. Defects that matter visually are not randomly distributed, and a sample that happens to miss a cluster has still accepted a screen with a visible blot. Cluster rules and bright-pixel rules must be 100% inspected or excluded from AQL entirely.
Suggested clause wording — pixel defects
Quality acceptance shall follow ISO 2859-1, single sampling plan, General Inspection Level II, normal inspection.Critical defects (AQL 0): pixel clusters of three or more adjacent defective pixels; any defective module; any visible abnormality affecting an area larger than one pixel.Major defects (AQL 1.0): full dead pixels, stuck-on pixels, intermittent or flickering pixels, functional failures of a cabinet or receiving card.Minor defects (AQL 2.5): dead sub-pixels and cosmetic blemishes not visible from the specified viewing position.Clusters and stuck-on pixels shall be inspected at 100%, not by sampling.Inspection shall be performed after a minimum 30-minute warm-up, at three times the nominal viewing distance, under 500-1000 lux ambient illumination, on full white, full black, red, green, blue and 50% grey test patterns.

Adjust the AQL values to the grade you are buying — 0.65 for major defects is reasonable on a premium installation — but keep the structure. The structure is what makes the clause enforceable; the number is what makes it negotiable.

Sample or inspect everything?

For most facade packages the honest answer is a hybrid, and it should be written as such rather than left to the inspector’s judgement on the day.

  • Sample the things that are manufactured in quantity and behave statistically: modules, bars, power supplies, connectors, packaging.
  • Inspect 100% the things whose failure is visual or structural: cluster and bright-pixel checks on every cabinet, torque and sealing on every bracket connection, polarity and labelling on every feed.
  • Burn in 100% before packing. Intermittent faults do not exist at minute five. They appear after hours at temperature, which is precisely why a 48–72 hour burn-in, with records, belongs in the acceptance clause and not only in the supplier’s brochure.

Two practical additions that cost nothing and prevent most disputes: require a defect map for any screen that triggers a rejection, recording cabinet and module coordinates rather than only a count; and require photographs of the test patterns taken at the contractual viewing distance. A count can be argued. A photograph with a timestamp and a cabinet number cannot.

Brightness, uniformity and colour: the numbers people forget to define

Pixel defects are the visible failure. Uniformity and colour are the failure that makes a finished facade look cheap while passing every defect count.

ParameterHow to express itTypical specificationWhere it goes wrong
BrightnessCalibrated nits at a stated colour temperature, not “max”5,000–6,500 nits for sun-facing facades; 2,500–4,000 for shaded elevationsQuoting driver maximum rather than calibrated output
Brightness uniformityMinimum divided by average, measured on a 9- or 13-point grid at full whiteCommonly 85% for outdoor screens, 90% for indoor-facingNo grid defined, so two parties measure different points
Colour consistencyΔE between modules, or SDCM binΔE ≤ 3; 3-step binning or tighterAccepting “same batch” as a colour guarantee
Colour temperatureMeasured CCT plus tolerance±300–500 K across the screenWhite balance set by eye at commissioning
Grey scaleBit depth plus a low-grey visual check14-bit or higher, no banding at 5% greyBand stepped gradients accepted as normal
Viewing angleBrightness retention at a stated off-axis angleDefined per product; state the angle you needMeasuring on-axis only

Values are common commercial specifications, not universal standards. What matters is that each one is expressed with a method and a condition, so both parties measure the same thing.

The condition does most of the work

A brightness figure without a warm-up time is not a measurement. LED output drifts for the first 20–30 minutes, and a screen measured cold reads differently from the same screen measured warm. The same applies to ambient light: a uniformity figure taken at night with no ambient light is not the figure your facade will show at 7 pm in summer. Write the condition, or accept the number you are given.

Factory acceptance and site acceptance are different tests

Most contracts mention “acceptance” once. There are two gates, they catch different failures, and conflating them is how a defect discovered on site gets attributed to transport, installation or the factory with no way to settle it.

Factory acceptance (FAT)Site acceptance (SAT)
WhenAfter burn-in, before packingAfter installation and commissioning, before handover
What it provesThe product left the factory in the specified conditionThe installed system performs in the real environment
Key checksDefect map, brightness and uniformity, colour consistency, control addressing, IP sealing on sample units, burn-in recordsMechanical fixing and alignment, waterproofing at penetrations, insulation and earth, full-screen uniformity in situ, control scenes, night-time visual check
Who attendsBuyer’s inspector or third party; increasingly by live video with dated evidenceMain contractor, consultant, controls integrator
Typical omissionNo written record of which cabinets were inspectedNo comparison against the FAT baseline

The link between them is the baseline. If the FAT records brightness, uniformity and defect positions per cabinet, the SAT has something to compare against. Without it, every site finding becomes a discussion about whether it was always there.

When it fails: four outcomes, and who pays

A clause that specifies acceptance but not the consequence will be resolved by whoever has more leverage. Name the four possible outcomes in advance:

  1. Rework at the supplier’s costFor defects within the supplier’s control and above the agreed limit. Include the re-inspection cost, not just the repair cost — otherwise rejecting a batch is free for the supplier and expensive for you.
  2. Concession with a price adjustmentFor cosmetic or minor deviations that do not affect performance or safety. This is a legitimate outcome and should be available, priced by a formula agreed before production rather than negotiated under pressure.
  3. Rejection of the lotFor critical defects or repeated major failures. State the trigger — for example, two consecutive lots failing — rather than “at the buyer’s discretion”.
  4. Replacement of identified unitsFor isolated defects found at SAT. Require the replacement to be from the same production batch where colour consistency matters, and require the defect map to be updated.

Two commercial details belong in the same clause: retention. Holding 5–10% of the contract value until SAT is signed is normal on facade packages, and it changes behaviour more reliably than any quality paragraph. If your payment schedule is 100% before shipment, you have removed your own acceptance leverage, whatever the specification says.

AQL ends at delivery. Warranty starts there.

These are two different promises and they are routinely merged into one sentence, which is how buyers end up with neither.

AQL describes the condition of the goods at the moment of inspection. It says nothing about what happens in year three, and it is not a failure-rate guarantee over time.

Warranty describes what happens when units fail in service. It needs its own numbers: a failure-rate threshold, a response time, who pays freight, and whether replacement is by module or by unit. A “5-year warranty” that covers parts only, ex-works, with an eight-week replacement lead time is worth materially less than a 3-year warranty with two-week advance replacement.

Both should appear in the contract, with the boundary written explicitly: defects present at delivery are handled under the acceptance clause; failures arising after acceptance are handled under the warranty clause. The warranty comparison covers what 3, 5 and 8 year terms actually include.

Frequently asked questions

What is an acceptable dead pixel rate for a new LED facade screen?

Commercial-grade screens are typically specified at 3–5 dead pixels per million, with premium grades at around 1 per million. On a 100 m² P25 screen of roughly 160,000 pixels, 5 ppm permits less than one full dead pixel across the entire screen. Because that is impractical to administer as a pure rate, contracts normally combine a sampling standard for modules with a 100% visual check for clusters and bright pixels.

Is 0.01% a strict dead pixel specification?

No. 0.01% equals 100 parts per million, which is roughly twenty to thirty times looser than the commercial norm of 3–5 ppm. It is widely copied between contracts because it looks precise. If you intend a strict standard, specify ppm and the defect classes, not a percentage.

What does AQL 1.0 actually mean?

It is the acceptance quality limit used with a sampling plan: lots whose defect rate is at or below 1% have a high probability of acceptance. It applies to the lot, using the sampling scheme and inspection level you specify, and it only means something once the defect classes are defined. It does not mean “1% of pixels may be dead”.

Should clusters be covered by AQL or by a separate rule?

A separate rule, and preferably a zero-tolerance one. Clusters of three or more adjacent defects are visible as blotches and are not randomly distributed across a production lot, so sampling can accept a screen that clearly fails visually. Specify 100% inspection for clusters and stuck-on pixels.

How is brightness uniformity measured?

Most commonly as the minimum reading divided by the average, measured on a defined grid — 9 or 13 points are typical — on a full-white pattern, after warm-up, with the meter at a stated distance and angle. State the grid, the warm-up time and the ambient condition in the clause, otherwise two competent inspectors will produce two different numbers.

Who pays for re-inspection if a batch fails?

The supplier, and the clause should say so explicitly. Re-inspection costs — inspector time, travel, storage, delay — are the mechanism that makes an acceptance limit real. Where the contract is silent, the buyer usually absorbs them, which makes rejection expensive and therefore rare.

Does a passed FAT mean I can skip SAT?

No. FAT proves the product left the factory correctly. SAT proves the installed system works: fixing, alignment, waterproofing at penetrations, earthing, in-situ uniformity and control scenes are all site conditions the factory cannot test. Skipping SAT transfers those risks to the building’s maintenance budget.

Send us your acceptance clause before you sign it

We will tell you which parts are unenforceable, which numbers we can commit to in writing, and what our factory record would look like against each one — including the parts we would fail.

  • Clause review against ISO 2859-1 sampling practice
  • FAT and SAT checklists matched to your specification
  • Defect map and burn-in record format you can hold us to
Request a clause review

Basis of this guide. Sampling references follow ISO 2859-1. The defect classes and inspection conditions described are common commercial practice in facade LED supply, not a published standard; adapt the thresholds to the grade being purchased and the viewing distance of the installation. Values quoted for uniformity and brightness are typical specifications, and should be confirmed against the project’s lighting design.

Project enquiry

Send your drawings. Get a specification you can defend.

Elevation DWG, facade area, closest viewing distance and site country — we reply within 48 hours with a fixture matrix, power and DMX topology, and a quotation. No price-only quotes.

5,000+ projects delivered · 40+ countries · ISO 9001 · CE / UL / RoHS

About the publisher

MEKALEO is the export brand of Guangdong Meidile Technology Lighting Co., Ltd., a manufacturer of outdoor LED facade lighting and LED media facades. Founded in 2004, the company has worked on exterior architectural LED systems for 22 years and runs its own 8,000 m² factory in Guzhen, Zhongshan, China, with in-house SMT, vacuum potting and an IP test chamber. To date: 5,000+ delivered projects across 40+ countries, 100+ patents, ISO 9001, and CE / UL / RoHS certified products.

About us · Talk to an engineer

GET A QUICK QUOTE ↗

Related Articles

Get a Quote for Outdoor Lighting