This page does not name, rank or score any company, and it does not repeat a single published score as if it were a fact about a window. It describes how those numbers are manufactured, what each one actually measures, and what it structurally cannot reach. Every arithmetic step in this article is shown so you can redo it; every number taken from a platform is attributed to that platform. Do not copy any number out of this article into a purchase order, a specification or a drawing.
One more disclosure, made deliberately, because it is the subject of this article: this website is operated by a manufacturer that sells sliding windows and doors. A United States federal trade regulation rule requires a business to disclose clearly when a website it controls claims to rate or rank a category that includes its own products. We make that disclosure in full, and we then decline to rate anybody, including ourselves.
Search behaviour has a grammar, and it is worth reading the grammar before reading the results. The phrase you arrived with joins two words that belong to two different activities. Best is a comparison: somebody has to decide what the units are, line the candidates up, and pick the largest. Ratings is a measurement: somebody has to decide what instrument to hold against the object, where to hold it, and when to read it.
Almost every page that answers this search does the first one and quietly assumes the second. It produces ten rows in an order, attaches a number to each row, and lets you infer that the number was read off the product. It rarely was. In this trade the number was read off a company profile, a platform account, or a list maintained by whoever wrote the list.
There is a real, legally consequential, publicly searchable rating system for fenestration. It has been running for decades, it is maintained by a standards body, and it determines whether a product may be labelled and sold in several large markets. Its search key is not a company name. That single fact is the whole argument of this article, and everything below is an expansion of it.
Before comparing scores, separate the instruments. A score printed next to a sliding window company can come from any of five unrelated processes, and they do not measure each other.
| What people mean by rating | Who produces it | What the unit actually is | What it can never see | Can it go into a purchase order |
|---|---|---|---|---|
| Consumer review stars | Strangers who bought something, filtered and averaged by a platform | One experience, usually one household | Anything that happens after the review is written | No |
| Business bureau letter grade | A private nonprofit, working from complaints filed with it | Points for answering complaints plus age of file | The product entirely; customer reviews are excluded by policy | No |
| Editorial brand index | The website publishing the ranked list | A number the publisher chose, sometimes left blank | Its own method, which is usually not printed | No |
| Marketplace row score | A business to business platform | Activity on that row, including reply speed | Whether the row describes a factory or a reseller | No |
| Product performance rating | A rating council, working to a published standard | One tested or simulated configuration, one row | Configurations that were never submitted | Yes, and it is the only one that can |
Read the last column from top to bottom. Four of the five cannot be written into an order even in principle, because none of them describes an object you are buying. The fifth can, and almost nobody searches for it, because it is not published under a company name.
The National Fenestration Rating Council publishes a Certified Products Directory. Its own information page describes it plainly: it lists product lines and individual products selected by the manufacturer or responsible party for which certification authorization has been granted, and it can be searched by the public. This is not a marketing artifact. Ratings printed on a compliant temporary or permanent label, or listed on a compliant label certificate, are treated by energy codes and by the market as certified values.
Now look at how you actually search it, and what each door answers.
| Search the directory offers | What it answers | What it does not answer | Why that matters to a buyer |
|---|---|---|---|
| By manufacturer | Which rows this manufacturer currently holds | Nothing whatsoever about the manufacturer itself | The name is an index into rows, not a verdict |
| By product type | All currently certified rows of that type | Whether any row matches your opening | A type is not a configuration |
| By product directory number | One configuration, exactly | Anything outside that single row | This is the number that belongs on a submittal |
| Label verification | Whether a number on a label was ever certified | Whether the unit in front of you is that unit | Historical labels still return a result |
| Current certification only | What is authorized today | Rows that have lapsed are silently absent | Absence is a status, not a ranking |
The internal structure matters even more than the search page. Each certified configuration is represented by a row in a certification authorization report, and each row carries its own product number. Joining the product line number to the product number produces the directory number. Certification bodies note that identifying the correct row can be a challenge when a report contains hundreds, or even thousands, of product numbers, and that energy performance ratings vary with product configuration: coating, tint and reinforcement each move the value.
So here is the comparison that decides this entire article. In the only publicly searchable fenestration rating directory, one company occupies tens, hundreds or thousands of rows, and every row is a different number. In the search result you arrived from, one company occupies exactly one line with exactly one number.
This is not a criticism of review platforms. It is a statement about resolution. A telescope that resolves a crater is not better than one that resolves a galaxy; it is pointed at something else. The difficulty begins when a buyer takes a one number per company instrument and asks it a one number per configuration question.
Suppose two due diligences are carried out honestly, by two competent people, one month apart, with no fake reviews anywhere in the picture. They can still return different numbers for the same company, because the operators are not computing the same function.
| Operator | What it computes, described from its own documentation | Is the denominator visible | Can the score move with no new review | A detail worth remembering |
|---|---|---|---|---|
| Trustpilot | A TrustScore from three documented factors, recency, frequency and a Bayesian average, explicitly not a simple average | Count is shown, weighting is not | Yes, time decay alone moves it | Displayed stars are the TrustScore rounded to the nearest half star |
| Yelp | An average of recommended reviews only, selected by fully automated software | Only the recommended subset counts | Yes, the software runs daily and reclassifies | Unrecommended reviews remain readable and count for nothing |
| Better Business Bureau | A point score on a one hundred point scale mapped to a letter | Points are published, inputs are not | Yes, age weighting and file updates move it daily | Customer reviews are explicitly not used in the letter grade |
| Search engine business profile | Commonly understood as the simple average of displayed reviews | Count is shown | Only through new reviews or removals | The simplest function here, and the one everybody assumes everywhere |
| Business to business marketplace | A composite of activity signals attached to the listing | Almost never | Yes, with every reply and every login | A fast reply is often worth more than a good sash |
When two published counts disagree, the instinct is to assume somebody cheated. Usually nobody did. Two operators computed two different functions over two different populations and printed both answers beside the same logo. In at least one jurisdiction a court has already been asked whether a platform misleads users by showing an average computed over a filtered subset rather than everything submitted, and the court held that the platform did not claim otherwise. The filtering approach survived judicial review.
Trustpilot documents that its Bayesian average behaves as though every new profile already held a small number of neutral reviews. Independent modellers commonly describe this as seven reviews at three and a half stars. Whether the exact figure is seven matters less than the direction of its effect, and that effect can be computed by anybody with a calculator.
The following arithmetic is ours, built on that published mechanism, and we show the inputs so you can ｒｅｐｌａｃｅ them. Assume every real review a company receives is a perfect five, and vary only how many it receives.
| Real reviews received | True average of real reviews | Score before rounding, our arithmetic | Gap from perfect | The kind of business that typically sits here |
|---|---|---|---|---|
| 3 | 5.0 | 3.95 | Large | A factory with a handful of export clients |
| 6 | 5.0 | 4.19 | Clearly visible | A fabricator selling through distributors |
| 12 | 5.0 | 4.45 | About half a star | A small retail installer |
| 30 | 5.0 | 4.72 | Small | A busy regional replacement firm |
| 60 | 5.0 | 4.84 | Negligible | A national operation with a review collection routine |
Two companies whose customers were equally delighted will display scores separated by roughly two thirds of a star, purely because one of them met more countable customers. The gap carries no information whatsoever about either window.
Now place that beside the structure of this industry. A factory sells through distributors, contractors and developers, and may complete a year of work touching only a few dozen legal entities. A local installer drives to a house and asks for a review on the way out. The score function under rates precisely the tier that melts, extrudes and assembles the window. This is not a conspiracy. It is a consumer instrument pointed at a supply chain.
This is the deepest structural problem, and no amount of honest reviewing can repair it.
Trustpilot states that recency weighting gives more influence to recent reviews than to old ones, using exponential decay, so the score reflects current performance. It does not publish the exact decay curve. Bands circulated by independent analysts put a review in its first year near full weight, a review in its second year materially lower, and a review beyond three years at a small fraction of its original influence. Treat those bands as somebody else's model rather than as the operator's own disclosure. The direction is not in dispute.
The business bureau goes further and sets an explicit horizon: its business reviews generally cover a three year reporting period. Yelp's software runs daily, and its own documentation warns that a review can move between recommended and not recommended days, weeks or months after posting, so the visible star can drift without any new opinion being written at all.
Put those three facts beside what actually happens to a window.
| What eventually decides whether the window was good | When it normally first appears | Is that moment inside the scoring window | Could a handover day review see it | Who actually owns it |
|---|---|---|---|---|
| Gasket compression set and hardening | After several seasons | Often outside | No | Shared between product and maintenance |
| Hardware drift needing seasonal adjustment | First hot season | Usually inside | No | Maintenance, not defect |
| Coating chalking and colour shift | Years of exposure | Outside | No | The coating and pretreatment system |
| Insulating glass unit fogging | Any time, statistically later | Frequently outside | No | The edge seal of the glass unit |
| Drain path blocked by site debris or sealant | First heavy season | Usually inside | No | The site, not the factory |
| Spare part availability and colour batch match | Whenever damage happens | Almost always outside | No | The documentation you did or did not demand |
Read down the third column. Most of the things that decide whether this purchase was wise arrive after the third year. Every major consumer scoring operator discounts, truncates or simply stops counting somewhere between three years and never. The weighting shrinks as time passes while the risk grows as time passes. Using one to predict the other is not an approximation, it is an inversion.
And the review itself is usually requested at the moment of maximum goodwill, the day the old unit came out and the new one went in, when the room is clean and the invoice has not yet been forgotten. A sliding window is one of very few building products whose quality is largely invisible on its best day.
Even setting the timing problem aside, consider what a single review can be describing. Nothing in any platform requires the writer to say which one it was.
| What one review might actually cover | Weight in the average | What the author actually experienced | What it licenses you to conclude |
|---|---|---|---|
| A paid sample sash on a desk | One review | Finish, machining, packaging | Almost nothing about production |
| One replacement unit in one house | One review | A van, a crew, a clean handover | Local service quality, nothing more |
| One container holding several hundred units | One review | Freight, packing, customs, count accuracy | Logistics performance only |
| A whole development delivered over two phases | One review | Batch consistency across many months | The one thing a project buyer needs, weighed the same as a sample |
| One email exchange about a missing roller | One review | How quickly one person replied | Almost nothing about the product |
A five line exchange about a missing roller and a two year, two hundred unit delivery carry identical weight in every average described above. United States trade regulation explicitly treats a rating with no written text at all as a consumer evaluation, which means even a bare click of five stars counts as one whole unit of evidence.
The same reflex also migrates between product words without improving. Somebody building a company shortlist one week may be hunting for best entry door company the next, simply because whichever opening is currently leaking has changed. The mechanism does not get smarter when the noun changes.
Some ratings pages do one decent thing: they print the review count beside the score instead of hiding it. On a page of that kind, four replacement window companies were listed with their customer rating and their count side by side. We will not repeat which companies those were and we will not repeat the warranty wording beside them, because a warranty period quoted from a third party is not something we will reproduce. But the shape of such a table is instructive and rewards being read once, carefully.
| Observation available to anybody reading such a table | What it looked like | What it means for a buyer |
|---|---|---|
| The largest count did not carry the best score | The entry with the most reviews sat at the top and the entry with the second most sat at the bottom | Volume and score move independently, so never read one as a proxy for the other |
| Ordering by score was not ordering by count | The two columns sorted differently | A ranking is a decision made by the publisher, not a property of the data |
| The comparison columns described the visit, not the unit | Installation included, coverage terms, range breadth, service | None of the stated criteria was a fabrication or performance number |
| Every entity being rated was a retail replacement business | Each listed company sells an installation alongside the product | This is the next section, and it is the important one |
When a page prints the count, read the count first and the score second. When a page does not print the count, treat the score as a label rather than a measurement, and ask for the denominator in writing before letting it move anything.
Follow the physical route by which a review comes to exist. Somebody has to be standing in front of the customer, in the customer's building, at the end. Somebody has to have a name the customer can type into a search box, in the customer's language, in the customer's jurisdiction. Somebody has to want the review, because somebody's local visibility depends on it.
That somebody is almost never the plant that extruded the profile. The score you can therefore find describes the last mile in high resolution and the first mile barely at all. If you buy factory direct and hire installation separately, you will spend an evening reading installers' reviews in order to judge a factory, and nothing on either page tells you that this is what you are doing.
| Tier | How many reviews it plausibly accumulates | What each review really describes | How much you need it |
|---|---|---|---|
| Local installer or replacement retailer | Hundreds to many thousands | Punctuality, tidiness, politeness, invoice clarity | Genuinely useful, for choosing that installer |
| Regional distributor or dealer | Dozens | Lead times, documentation quality, credit behaviour | Useful for supply reliability |
| System assembler or fabricator | Single digits to dozens | Shop drawings, tolerances, spare part support | Very useful, and almost never available |
| Extruder and component maker | Close to zero | Alloy, temper, dimensional control, die condition | Determines most of the outcome and carries no score at all |
Stars are densest exactly where you need them least, and absent exactly where they would matter most. None of this is concealed. It only requires noticing that a review is written by whoever met the customer, and that the deeper you travel into the supply chain, the fewer customers anybody meets.
Of all the operators discussed here, the business bureau is the most transparent, because it publishes its arithmetic. Its grade runs from the highest letter down to the lowest, assigned from a one hundred point scale, and the point range for each element is public. Here it is, arranged by direction.
| Element | Published range | Adds or deducts | Is any part of it about the product |
|---|---|---|---|
| Complaint volume, weighted by complaint age | 15 to 0 | Adds | No |
| Unanswered complaints | 40 to 0 | Adds | No |
| Unresolved complaints | 30 to 0 | Adds | No |
| Complaint resolution delayed | 5 or 0 | Adds | No |
| Time in business | 10 to 0 | Adds | No, it is a proxy for being on file |
| Failure to address a complaint pattern | 0 to minus 31 | Deducts | No |
| Type of business | 0 to minus 41 | Deducts | No |
| Transparent business practices | 0 or minus 5 | Deducts | No |
| Failure to honour mediation or arbitration | 0 to minus 41 | Deducts | No |
| Competency licensing | 0 or minus 41 | Deducts | No |
| Government action, per action | 0 to minus 34 | Deducts | No |
| Advertising review, per incident | 0 to minus 41 | Deducts | No |
| Trademark infringement | 0 or minus 41 | Deducts | No |
Add together every element that can only add points: complaint volume, unanswered complaints, unresolved complaints, resolution delay and time in business. Their published maxima sum to one hundred, which is the whole scale. Our arithmetic on that published table gives ninety points tied to complaint behaviour and ten points tied to being on file. Every point a business can earn is earned either by attracting no complaints or by having existed for a while. A perfect letter therefore certifies something real and genuinely worth knowing: this entity answers letters. It does not certify a sash.
Three further details from the same source deserve to be more widely known. The bureau states in writing that customer reviews are not used in calculating the letter grade, so the stars and the letter printed on one profile are two unrelated measurements sharing a page. Where reliable information about time in business cannot be obtained, the bureau treats the business as having started when the bureau opened its file, so the clock for a foreign factory begins whenever somebody first created the record. And the organisation does not endorse any product, service or business, stating that a rating is not a guarantee of reliability or performance; a no rating designation is explicitly not a grade and not negative, though buyers routinely read it as a low score.
Without naming anyone, and without repeating a single figure from those pages, here are four checks anybody can run in about a minute. They are not accusations. They are what an instrument looks like when it has been asked to rank things it was never built to rank.
| The check | What you sometimes find | What it tells you |
|---|---|---|
| Look at the score cell of the row ranked first | Sometimes it is blank, carrying a dash where the index should be | The order was not produced by the printed number |
| Scan the score column for identical values | Several companies printing the same value while occupying different ranks | Ties are broken by something unstated, or by nothing |
| Check whether an attention figure sits beside the quality figure | A raw visit or interest count presented in the same visual family | Attention is being quietly substituted for performance |
| Check whether every entry carries a superlative plus a survey | Each row introduced as highest, strongest or fastest across a sample nobody can inspect | Ten rows all winning something is a formatting requirement, not a finding |
This is where the legal frame becomes concrete rather than academic. The United States trade regulation rule on consumer reviews and testimonials, which took full legal effect in the autumn of 2024, bars a business from misrepresenting that a website it controls provides independent reviews or opinions about a category including its own products. It bars buying reviews conditioned on sentiment, bars undisclosed insider reviews, bars suppressing negative reviews by threat or by claiming a partial set is the whole set, and exposes violations to per violation civil penalties adjusted annually. We will not quote a penalty amount here, because the current figure belongs to the regulator rather than to us.
Read that list again slowly, because it doubles as a specification of what a buyer should demand. The regulators prohibited precisely the four manipulations that make an aggregate look better than the underlying experience: purchased positivity, concealed affiliation, selective display and invented authorship. If any of the four is present upstream, no amount of reading skill downstream recovers the truth, because the input was corrupted before the average was taken.
A star attaches to whichever profile was claimed, verified and maintained. A contract attaches to whichever legal person signed. Those two are frequently different entities, and the gap is not fraud. It is ordinary company structure meeting ordinary platform mechanics.
| Where the name appears | Who maintains it | Can it carry a score | What to get in writing |
|---|---|---|---|
| Consumer review profile | Whoever claimed and verified the listing | Yes | Which legal entity owns the profile |
| Trading or exporting entity | Your commercial counterparty | Sometimes | The exact registered name that will appear on the invoice |
| Manufacturing site | Operations | Rarely | Which plant will build your order |
| Platform storefront | A sales team | Yes | Treat it as a channel, not as a factory |
| Exhibition stand | Whoever rented the space | No | Verify hall and booth number before travelling |
| Declaration or certificate of conformity | The responsible party named on it | No | Read the name on the document, not the brochure |
There is also the question of which plant. A factory operating two production centres can build your order in either, while certificates, schedules and plant qualifications each attach to something specific. Which address is on the file and which plant will press my profiles are two separate questions, and nobody volunteers the second unprompted.
If this article argues that a published number should always be traced back to what it measures, then our own numbers have to be traced too. Here is the whole list, each item named for what it actually is.
| What we publish | What it actually measures | What it does not license you to conclude |
|---|---|---|
| Thermal values 1.3, 1.6 and 1.7 | System values under the boundary conditions stated with them | That any is a whole window value for your opening, or a value from a rating council programme |
| 240 hour salt spray, one series only | Coating and substrate corrosion behaviour of that one series | Anything about other series, or about your project exposure class |
| 5000 newton shear, one series only | The composite bond between thermal strip and aluminium | Structural performance, or air, water or wind resistance |
| Load figures 200, 220 and 400 kilograms | Three separate measurements on three separate products | That they form one descending hierarchy; sorting them in a column is the main way our catalogue gets misread |
| A cycle count quoted for a handle | A bench test of one component | A service life for the window |
| Frame wall width across the range | Depth going into the wall, per series | That depth rises with track count; it does not, and ours is one place to verify that |
| Alloy designation across the range | Which alloy and temper we use everywhere | That it separates one of our products from another; being common, it cannot |
| Two production centres | Where your order may physically be built | Which one yours will use unless you ask and record the answer |
| A sliding window range containing one product | A side press window | That we quietly also build the ordinary two track slider most people mean by that word; we do not, and we will say so rather than substitute another product |
| Several category pages with nothing on them | The honest state of our catalogue | Availability; ask before building a schedule on any category that returns nothing |
| Project photographs and descriptions | Work we have done, visible on our Projects page | That those buildings are still performing; photographs are taken once |
| Nothing at all in the consumer ratings pages people find first | The truth: we do not have a score there | That we are worse; we are simply not the kind of business that accumulates tens of thousands of consumer experiences |
Two further admissions in the same spirit. First, the search reflex that brings people here also sends some of them looking for a very specific small opening, and a phrase such as sliding door for washroom will not land on a dedicated page in our catalogue. That is a real gap and we would rather name it. Second, projects often mix product families, so a single review line about one shipment may be describing behaviour quite unlike a slider, including something like folding sliding glass doors. One star then covers two machines sharing no hardware, no track logic and no tolerance budget.
The most useful position a factory can take is to admit publicly that it does not have a score. What it has instead is drawings, schedules, batch records, test evidence for named items, and a named person who answers. Those are the things that survive contact with a real project.
Everything above is subtraction. Here is the addition, and it is the only part of this article worth copying. If an aggregate star cannot carry your risk, build the instrument that can and apply it identically to every quotation you receive. Leave the weight column blank on the first pass, then fill it in once two or three responses have arrived and you can see which rows actually separated them.
The unit being scored below is not the company. It is the order line, which is also the unit that still exists in year eight.
| Row to score | Full marks looks like | Zero looks like | Weight |
|---|---|---|---|
| Section drawing with dimensions, not a catalogue picture | Every chamber, the water path and the drain exit are dimensioned | A closed outline that hides where water goes | ____ |
| Your opening appears on their drawing | Your millimetres are printed on their drawing | Standard sizes only, for you to adapt later | ____ |
| Glass build written as a stack, spacer included | Glass, gap, gas, spacer type and coating identifier all named | Double glazed, or premium insulated unit | ____ |
| Estimated sash weight with the assumption stated | A number plus which glass build it assumes | Silence, or heavy duty | ____ |
| Load rating with its unit and its referent | Per sash, naming what was measured and under which glass | A kilogram figure with no referent | ____ |
| Maximum sash size under your chosen glass | Given as a stated condition or a curve | One maximum stated as if glass were irrelevant | ____ |
| Hardware platform and groove standard named | Platform, groove geometry and adjustment range in millimetres | European hardware, or German engineering | ____ |
| Configuration and handing, with the convention used | A drawing plus the viewing convention that produced the label | Left or right, with no stated viewpoint | ____ |
| Colour and coating batch rule across phases | One batch commitment, batch numbers recorded per unit | Same colour name, new batch, no record | ____ |
| Packing specification for the voyage you are taking | Corner protection, track guards, desiccant, stacking scheme shown | Export standard packing | ____ |
| Port, delivery term and what each term covers | Named port, named rule, named version, scope in writing | A figure that silently includes something undefined | ____ |
| Who signs the perimeter seal around the opening | A named trade on your side accepting that scope in writing | Nobody, because it sits between two scopes | ____ |
| One named first responder, whoever is ultimately liable | One person who accepts the ticket before apportioning blame | Three contacts spread across two companies | ____ |
| Spare parts list with part numbers in the original shipment | Rollers, gaskets, locks, handles and touch up travelling in your container | Spare parts available on request | ____ |
| Willingness to say no | Declined the wrong application and offered the right alternative | Agreed to everything without asking one hard question | ____ |
The last row deserves a sentence of its own. Every competent supplier you will ever meet has turned work away. A quotation that answers yes to everything without once asking about your wall build up, your exposure class or your tolerance budget has already told you something important, and it is not good news.
Nobody writes about this, and it is the single highest return habit available to a project buyer. A platform star is useless to you in year eight because it was written by a stranger with no identifiers. Write one yourself, with identifiers, at every milestone, into your own file, whether or not any platform accepts it.
| Field to record | Why it earns its place | What breaks without it |
|---|---|---|
| Order number and date | It is the only key every department accepts | Every later question starts with a search that fails |
| Series and frame depth | Identifies the extrusion set | A replacement is quoted from the wrong profile |
| Opening identifier matching the drawing | Ties each unit to a place in the building | Nobody can say which unit failed |
| Glass build written as a stack | Determines weight, acoustics and replacement cost | The new unit deflects differently from its neighbours |
| Colour name and batch number | The number, not the name, governs visual match | The patch reads as a different colour in daylight |
| Arrival condition with photographs | Separates freight damage from product defect | Two contracts argue while you pay for both |
| Who installed and who sealed the perimeter | Most water complaints are not product complaints | A joint failure becomes a claim nobody can settle |
| Date each symptom first appeared | Turns a complaint into a dated record | Wear cannot be distinguished from defect |
| Who responded and how long it took | Response time is the one service metric you can still act on later | You rehire a company you actually waited three weeks for |
| Would you order again, and what would you change | The only question that matters, answered while you still remember | You repeat the same avoidable mistake on the next phase |
Ten fields, filled twice a year, will outperform every published aggregate you can find, because it is the only corpus calibrated to your building, your climate and your supply chain. The most reliable rating of any supplier is the one you wrote down yourself.
Use these on every candidate. The right column matters more than the left, because a plausible sounding sentence carrying no information is worse than silence: silence at least reveals the gap.
| Ask | A real answer contains | Sounds reasonable, carries nothing |
|---|---|---|
| How many consumer reviews do you have, and where | An exact count, a platform name, and an honest note about their real volume | Thousands of satisfied customers |
| How many of your projects appear on that platform | A near zero and a description of how references really work | Our customers are always happy |
| Which legal entity signs, and which entity builds | Two named entities, in writing, before deposit | Same group, dont worry about it |
| Which production centre will my order use | A named site, recorded on the order acknowledgement | Our factory |
| Which standard and version was that value obtained under | Standard number, year, boundary conditions and what the specimen was | Meets international standards |
| Is this configuration listed anywhere searchable | A directory name and either a listing number or a clear no | The system has been tested |
| What exactly is the load figure measuring | Per sash, naming the component and the assumed glass | Heavy duty, four hundred kilograms |
| Which size is the nominal size actually referring to | Frame, opening, or visible glass, each labelled | Standard size |
| How much adjustment remains after installation | A millimetre figure for each adjustable interface | Fully adjustable |
| Which parts can I buy in eight years, and from whom | A part number list split into open market and factory only | Spare parts available |
| Who signs the seal around the opening | A named trade on your side, accepting that scope | The installer will take care of it |
| What would you refuse to build for me | A specific configuration, size or application, with the reason | Nothing, we can do anything |
| Situation | Why the star misleads here | Use instead |
|---|---|---|
| You cannot yet write your own specification | A comparison with no units has nothing to sort on | Fill the fifteen row sheet before contacting anyone |
| The score is published without a count | Three reviews and three thousand produce the same display | Request the denominator in writing, or discard the number |
| You will contract with a different legal entity | The measured object and the liable object are not the same person | Match the reviewed name to the signing name yourself |
| Your dominant risk arrives after year three | The weighting decays fastest exactly where your risk grows fastest | Ask maintenance questions, not satisfaction questions |
| You need a certified value for a specific configuration | No company level score can answer a configuration level question | Search the directory by row, and ask for the row number |
| One household, one installer, local permitting | This is the one case stars were designed for, and it is not this article | Stars are fine here; buy from whoever stands behind the installation |
The last row is written deliberately. For a single family home replacing a few units with a local company that measures, installs and warrants the whole assembly, aggregate review material works reasonably well, because the unit of purchase and the unit of review finally coincide. The argument of this article begins precisely when those two stop coinciding.
Structurally, no. Longevity is decided by gasket behaviour, hardware wear, coating stability, glass unit sealing and drain maintenance, all of which appear over years. Consumer scoring systems either discount older reviews, restrict themselves to a three year reporting period, or reclassify reviews automatically, and most reviews are collected at handover. A number gathered at installation cannot contain information about performance at year five, because nobody had experienced it yet.
Not comparable across platforms, and sometimes not comparable within one. Two operators may use different functions, one of them seeded with a neutral prior that pulls low volume profiles toward the middle, and at least one of them computes its average over a filtered subset. A lower number from one operator can represent a stronger underlying record than a higher number from another. Ask which function produced it and how many observations went in.
You can, and the quality of the answer is informative. A project manufacturer will usually produce very few consumer reviews and will offer project references instead. The useful version of the question is narrower: how many reviews, on which platform, how many of your projects appear there, and can you introduce me to two clients whose buildings are past their third year. Any answer to that last one is more valuable than any score.
Then the score did no work for you, which is normal. Compare rows instead: who will put your opening dimensions on a drawing, who states the glass stack including spacer, who gives a load figure with a referent, who names the first responder. Those fifteen rows separate suppliers reliably, because they are properties of the order rather than properties of a profile.
No, and conflating them causes real damage. A certificate or marking belongs to a responsible party for a defined product, under a named standard and version, and it attaches to the entity that signed it. It never transfers automatically to a different product, a different configuration, or somebody else importing the goods. A rating directory is even narrower: it keys on rows, and one maker can hold hundreds of rows.
Write a record, yes, but not a verdict. On arrival day the only things you can honestly assess are packing, count accuracy, finish and transport damage, and those are worth documenting with photographs precisely because they belong to a different contract from the product. Keep the assessment of performance open, dated and revisable: first hot season, first storm season, then each year.
Nothing in this article is a sales position, and nothing in it should be copied into an order. The platform mechanisms described here were read from each operator's own published documentation; the decay bands are somebody else's model rather than any operator's disclosure, and the regulatory summary is given without penalty figures because those change annually and belong to the regulator.
If you take one thing away, take the arithmetic: score the row, not the company. Then write it down, dated, with the batch number, in a file you will still be able to open in eight years.
