I think you’re audience here is not the audience you are seeking. I think the audience you are seeking may be more easily found on social media platforms maybe?

I’m sorry you are unhappy. I hope you find what makes you happy. (I mean that sincerely, with zero sarcasm. It sucks when you invest in something you aren’t happy with.) Good luck!

Let's start estimating the probability of a single user receiving repeated faulted machines; we need to define some parameters and assumptions:

Definitions:

Probability of a Faulted Machine (Pf): The chance that any given machine is faulty.

Number of Replacements (n): How many times you've received a replacement, including the original purchase. In this case, it's 4 (original + 3 replacements).

Using the binomial probability formula, the probability of receiving k faulted machines out of n total machines is:

P(X=k) = (n choose k) × Pf^{k} × (1-Pf)^{n-k}

However, in my case, I've received faulty machines every single time, so k = n. Thus, the formula becomes:

P(X=n) = Pf^{n}

If we knew the actual probability Pf (i.e., the fraction of machines that are faulty), we could calculate this directly. Lets assume a high number, say 10% of all machines are faulty (Pf=0.10), and I've received 4 defective machines in a row:

P(X=4) = 0.10^{4} = 0.0001 = 0.01%

This means there's a 0.01% chance of receiving 4 defective machines in a row if 10% of all machines are faulty.

However, there's a nuance: The chance of several users experiencing the same repeated issues suggests that the probability isn't just random bad luck. If multiple users are consistently receiving faulty units, Pf might be higher than estimated, there could be systemic issues in certain batches, or the way the refurbished units are handled might be problematic.

To thoroughly assess the situation, we can estimate now the case for multiple users

New definition:

j: Number of users experiencing the repeated fault.

Scenario:

Four users (j=4) each receiving four faulty machines in a row (n=4).

Assumption:

All users get machines from the same pool of refurbished machines, and the probability of getting a faulty machine remains constant across all picks.

Calculations:

The probability that a single user gets four faulty machines in a row is:

P_{1} = Pf^{4}

The probability that four users each get four faulty machines in a row is the fourth power of P_{1}:

P_{k} = (Pf^{4})^{4} = Pf^{16}

Now, let's solve for Pf in the equation P_{k} = Pf^{16} given a specific P_{k}.

For simplicity, let's use the probability we got for one single user in the initial calculations above, P_{k} = 0.0001 (0.01% chance that all four users would get four faulty machines in a row). Then,

0.0001 = Pf^{16}

To solve for Pf, take the 16th root of both sides:

Pf = 0.0001^{1/16}

Calculating this value, Pf ≈ 0.38 or 38%.

This means that for there to be a 0.01% chance of four users each receiving four faulty machines consecutively from the pool, approximately 38% of the refurbished machines in that pool would need to be faulty, that this a lot!

Of course, we need to keep in mind that this is a simplified model and the actual probability can be influenced by various factors. However, this gives a basic understanding of how high the faulty rate needs to be, in order to observe such a number of repeated faulty machines by several users under the defined assumptions.

Thank you, @trually. I genuinely value your insight. At times, I feel like I’m seeking a form of group therapy here.

I have no intention of broadly criticizing Glowforge on social media or creating negative content about them on platforms like YouTube, FB or IG. My hope was to express my feelings and frustrations in a community where others can truly empathize — those who understand the emotional and financial investment of purchasing such a machine, dedicating time, building expectations, and then facing disappointment.

It becomes even more challenging when I come across comments that seem to dismiss the very real frustrations that some of us face. Regardless of the percentage of users experiencing issues (and I’ve tried to estimate this), the fact remains: there are owners out there who’ve had multiple issues. Their experiences and feelings are valid. Fortunately, these kinds of comments are not the rule but the exception.

I have what is perhaps a weird position. I am either the oldest of the new people or the newest of the oldest. Instead of waiting over 2 years for my machine I waited for six months and paid more that the others did. That first machine died of my own stupidity within the first year and I got it replaced at no cost.

The replacement was indistinguishable from new. However, there had been early on a redesign of the exhaust and as best as I can tell those who had the older exhaust, have the older design in the refurbs they receive. The replacement worked fine for several years until one day it didn’t, and I paid for a refurb, that again was indistinguishable from new but still had the older exhaust. That machine did not even need to be calibrated.

Over the years I have seen complaints often of the same issue repeating over several machines. I suspect that many machines get replaced where the error was not in the machine, and many more (particularly during Covid), when a bad batch of some part (like rubber wheels) did not get caught during production and has to flow through the refurb system to wash out.

By the time the machines get to the refurb desk those issues are well known and double checked before the refurb leaves the plant. In this way my experience has been that the refurbs are more reliable than those brand new, and my two replacements have each worked better than the one before.

But clearly this can’t be the case or they’d be out of business — support costs would eat them up, both financially and in terms of labor allocation. They’ve got something like 100k units out there, if 38% of people had issues with returns we’d see a great deal more of these types of complaints. Right? Seems right to me.

Like you said your model is too basic to the point of being unpublishable. Just too many assumptions even if you’re right. You were right to put that disclaimer on it but I think you understated the number of guesses you’re making.

Edit: I’m also making a lot of guesses. Neither of us knows what we’re talking about with any certainty. Uninformed high five!

And here are the people beyond disappointed with their Formlabs machine and suggesting there should be a class action lawsuit.

Everything has bad reviews. Very few people come to online forums to post “I bought this thing and it works normally and I have nothing to say.” We have no way to know how common your bad experience is, as a fraction of all Glowforge owners. Statistically, half of us are below average.

[Edit] I wrote this before seeing your math post. I’m not going to critique the calculation as I don’t have the expertise, but it seems implausible. Rare events happen, and (in absolute terms) they happen more often with a larger sample size. I don’t see how your observation that there exist four users who received four faulty machines can correctly lead to the conclusion that 38% of refurbs are defective. If they only sold 16, it would be 100%, and if they sold 16 million, it would not take a 100% failure rate for four people to be unlucky. So my conclusion is that any such calculation needs the size of the pool as one of the variables.

Yes, as I mentioned we used to get tech support via this forum. Back then, posting in the Support area (now renamed Community Support) opened a support ticket, and that’s why Rita responded. Now there’s the Discord server where GF staff are often present. There’s also phone support available.

Other perks you’ve mentioned actually are linked to forum participation rather than longevity of ownership. Correlation <> causation, as they say.

Dan has mentioned in the past that the failure rate is in the low single digits. With hundreds of thousands of machines out there, that’s still a significant number of affected customers, and they are the ones most likely to be on social media complaining, which easily leads to a skewed impression of what’s happening.

Well, you are right, not 38% of the machines sold are faulty, and not 30% of ppl are having issues.

What I estimated was that 38% of the pool of the refurbished could be at fault. And we can estimate that number as well:

If we assume that most people have no issues, then only a very few numbers had to get a refurbished one, say, only 1% of them? That makes only 1,000 machines returned. Out of that, only 38% of those would be at fault, that is 380. That is a very low number and totally manageable.

Regarding finances, not all refurbished are for free, I just paid for mine. I don’t think they make their profits from refurbished, but I don’t think they are losing money either. But this is totally an speculation and I could be very wrong.

Actually, the model is simplified, but it is a solid one. The few assumptions are actually in favor of Glowforge, not against it.

The most important premise is that if you are getting a glowforge at random from a pool of refurbished machines if 90% of them were ok, then the probability of you having a bad one should be only 10%. But if you get one faulty one, your second is at fault, and then your third and fourth, are not OK, then the percentage of faulty ones in the pool from where you are receiving your replacement is higher than 10%. And we know how to calculate that percentage.

Even more, if the same situation happens to more than one user, that percentage of faulty machines has to grow even more. And we know how to calculate that percentage as well. And I did.

In the end, either the percentage of faulty machines within the pool of refurbished is high, or the assignments or glowforge are not at random.

Since I believe Glowforge is not sending me faulty machines on purpose, then the only plausible conclusion is that the pool of refurbished from where I’m getting my replacements has a high percentage of faulty machines.

Sure thing, we don’t know the actual numbers, but we can accurately calculate the percentages.

I think that @geek2nurse has it correct. First, by far the most significant number of faulty machines are due to rough handling in shipping. The packing system has improved considerably since the beginning, but idiot-proof is a fantasy. I have yet to see the big image of “Fragile” and “This-Side-Up” paid any attention to. In one case flipping it end over end instead of lifting and carrying it.

In the case of new owners the same “idiot proof” applies. Here I confess to being one of those idiots. I thought magnets were an obvious choice to hold material in place and did not consider what a strong magnetic field would do to spinning fans or other electronic equipment moving quickly through it. Liquids running about on circuit boards affect reliability as well.

For a long time, the black cable would break or mess up if you lifted the lid too far. Now that piece is a lot longer, but many folk still have the original and just don’t lift the lid that far. Other design issues will still exist like the faulty wheels, or bad glue, but the numbers will be a lot less than you are thinking.

Yeah, you are right that every manufacturing company faces similar problems. I’m lucky that my two Formlabs 3D printers (Generation 2 and Generation 3) have been super good to me.

I agree that there is no way to know how probable is for a user to get a faulty new machine.

However, we can get estimates if a random user gets random replacements, and then the replacement is bad, and the new replacement is bad, and the new replacement is bad. That series of dependent events is a chain: what’s the probability of me getting a bad machine after I got a bad machine after getting a bad machine?

Even more, that is not only me, there are at least another two or three ppl I have read their posts that have received chained faulty machines. That adds up to the model.

Because of the previous information, we can calculate the percentage of defective machines within the pool from where these machines were taken and shipped .

Even more, if there are more than one user with the same chain of events (which is true), then

The fact that there are several users with repeated, one after another, faulty machines,

Because, as I mentioned above, it is a chained event. One single user getting faulty machines in series.

Again, it is not 4 people receiving one bad unit. It is four people being unlucky in series. That’s the key difference being overlooked.

We don’t need the size of the pool to get the probabilities and work backward the number percentage needed to match these probabilities. The other possible conclusion would be that the machines are sent on purpose to a certain pool of users. That is just not possible.

A person hit by lightning once is unlucky, getting hit twice is very unlucky. But three or four times? and if there are 4 ppl? definitely they were doing something to increase the chances to a degree of that not being pure luck.

I’m familiar with the bathtub curve. It refers to a new generations of the products, not to the same physical product being better because it ages.

When I got my first new-refurbished I wanted to believe it was better than my new one. But alas, in a few cuts, I got sparks getting close the tube, and I saw this:

Glowforge new-refurbished are not new, advanced machines that fell under the right side of the bathtub curve; they are the same as the “new”; the only difference is that they were faulty once, so, more like at the left, in the faulty side.

There should be some statistical analysis to show if these are superior to new ones. But I agree we have no data.

Meybe we can use to use another distribution curve?

It is all at once, see, I have been with the machine since Jan 2022. I just missed the old good days. I got it.

Well, in QA there is a saying “In God we trust, everything else, we test”

While I get that a single event can be amplified in social media for different purposes, this is not what I’m talking about.

Mine is not a skewed impression. It is repeated (chained) events of receiving bad after bad, after bad, after bad. Not one single user with that experience, but a few. That data renders a way to estimate a percentage of the number of bad items in a given pool of goods. Not opinions, but data.

Now, while the percentage of bad items within the pool could be estimated and it seems to be high, the actual number of physical machines could be low, and there is no way to know it without data from Glowforge. But if I have to guess, I believe it is in the 200 - 300 hundred. (yes, it is a believe, I could be VERY wrong. Not the same with the percentage).

BTW, I’m not proposing any legal action against Glowforge, just for the records!

You can calculate the probability of an event such as getting 4 failures in a row from the failure rate, but the expected number of events depends on the population. For example, with a failure rate of 5%, there is a 1 in 160000 chance of getting 4 bad machines. That means that if Glowforge sold 1 million machines, you would expect just over 6 people to be unlucky enough to get 4 bad ones. If we use your failure rate of 38%, it would only take 200 machines before you’d expect to see a 4-in-a-row event. With a more realistic assumption of 100,000 sold, there would be over 2000 customers who got 4 bad ones, and I don’t think that’s consistent with the evidence. I think your math is wrong.

That’s a question I can answer: 100%. I think this is part of what you are misunderstanding about the statistics. You are conflating the probability of one specific person winning the lottery with the chances that someone will win. There’s a great Feynman quote:

“You know, the most amazing thing happened to me tonight. I was coming here, on the way to the lecture, and I came in through the parking lot. And you won’t believe what happened. I saw a car with the license plate ARW 357. Can you imagine? Of all the millions of license plates in the state, what was the chance that I would see that particular one tonight? Amazing!”

Roy Sullivan was hit by lightning seven times. Unlikely things happen. Random events are also more variable than we intuitively expect.

That’s only one interpretation. A lot of products exhibit failure rates that follow a bathtub curve due to the nature of electronics components and typical defects: faulty components, manufacturing errors, damage in shipping, etc., will cause a lot of them to be dead on arrival or burn out quickly. Once you get through that period, reliability creeps up because you probably got one with good capacitors, non-burnt wiring, etc. And then of course the fundamental impermanence of all things kicks in at the end.

That’s not really relevant. You’re talking about a different sample set. @artexmg is saying the pool of refurbished machines is exhibiting the high failure rate - not the pool of 100,000 sold machines.

True but probably not explanatory to the scenario of why there have been multiple people experiencing 4 or more failures. In fact Dan posted a few years back after (I think) the first one of those popped up. He pointed out that it could be predicted that situation would occur. But that was (presumably) from the same analysis you made of the total number of machines that have shipped (or had at that time).

We all know that refurbished machines had a problem. They were often not manufactured properly. So they have a failure in their past. By definition then, they are also more likely to have more than one thing not up to snuff. (The existence of one failure in a component assembly increases the likelihood that there is another failure as whatever led to the first can often contribute to more.) Then they get refurbed where whatever was broken is supposed to be fixed. In a quality refurb operation, refurb machines are also checked for any known defect vectors that may be latent in the machine being repaired. So for instance, any machine returned would be looked at for the possible need for the replacement of the wheels or the lid cable, or head connectors, or wiring misrouting - all things that have been identified as not rare and thus should be checked in all new (and refurbed) machines leaving the factory.

It is entirely possible and even reasonable that the refurb company may not do that or may not do that for all machines. While refurbed products in general tend to be quite reliable, the results of a specific refurb operation may affect that negatively. We have seen examples of dirty and obviously not “120 point inspected” machines getting received by customers so the refurb operation’s track record does not suggest they are to be entirely trusted to do the job being contracted to them.

You both are potentially correct with regards to the argument you are positing - but those are not arguments about the same thing.

this i agree with 100%. there were most definitely QC problems with the refurb company at least at one point. i know that they were having problems getting enough refurbs at one point to replace machines, and i’m guessing that was around when they were trying to improve that QC quality.

A single careless idiot at your local UPS office would really greatly increase your chances of repeatably getting bad machines. Likewise, a user that had 4 machines in a row go bad with the same or a related issue (from opening the lid too far for example that several symptoms point to the black cable breaking in several different locations) that could point to an issue with the owner. If I had not been able to imagine what the magnets were doing to the print head, I could have been one of those people. My previous machine also had a replaced black cable from opening the lid too much just one time. The replacement was longer than the original, and the one in the replacement machine is longer still. I was flabbergasted when the replacement black cable actually worked after having to play with the connections when they looked good to the eye. It is quite easy to break the new cable in installing it and getting mad that replacing the cable “did not solve the problem”.

It is certainly true that there are weak places in the design that 20-20 hindsight has no doubt sorted out. In the place of the black cable these have been reconfigured several times. In the place of that weak point in the high voltage cable I can only imagine that like making sure the newest version of the black cable is installed, looking at those places known to be a problem would be looked at and replaced if getting worn. This alone will make them better than new as issues not picked up would be addressed.

I studied my latest machine looking for evidence like crud in hard to get at areas, but the only thing that told me it was a refurb was the older design for the exhaust that it would take the newer case as well to make the change.

I suspect but cannot prove that if there is any ranking at all, the machines that came in with the least problems (or none as is often the case) and looked the best will be the first to go out and those like my first machine with the steel sheet glued on to hold the magnets would be the last if at all.

So your math does not contain those points and therefore not close to the facts.

The difference is that I’m not calculating the number of events as you mention, but the probability of a scenario when several users get a chained bad machine from a reduced pool (new-refurbished pool). These two are different probability events, asking for different questions.

One more time, you are trying to answer a different question.

No, it is not 100%. Again, your initial setting is not right. You are confusing the settings, which is someone winning the lottery versus the same person winning the same lottery several times. Worst, the same person is winning in a row, not only in one state but in three more.

The lottery system and Casinos look closely at these kinds of outlier events; oversimplified, but this is how they detect fraud when these extremely low-probability situations happen. In other words, if it is too good to be true …

Interesting quote. Unrelated to the current case but interesting.

The setting to compare would be if he had correctly predicted the numbers of four license plates that he would see the next day and repeated the same fate for four days.

If that were the case, I would say that, with a probability of almost 100%, he performed a magic trick. This is, he “cheated” (in the sense of magicians “cheat”) to increase his chance of success every time he predict the plates.

This is a great example, indeed. Will help clarify the ones you’ve gotten wrong.

First, the probability of being struck by lightning is about 1 in 1,000,000. Extremely low to assign to a particular person, but 100% certainty that the event will happen. By the way, I’m not confused about these two separate scenarios, as you implied above.

Second, the probability of the event the same person being struck seven times, is (1 in 1,000,000) to the power of seven, that is, 0.00,000,000,000,000,000,000,000,000,000,000,000,000,000,001. This is very, very low, practically zero.

However, given that event actually happened, Bayesian statistics tells us that we have to conclude that the odds for Roy Sullivan are way higher than most people due to his work, his habits, and the region where he lived. Given the low probability of simple chance, some are advancing that there are genetic predispositions to be struck.

The other option is that he lied about the events and they didn’t occur as he told.

Finally, considering that there are three more people with the same “luck” as Roy Sullivan, then the odds are incredibly small. This is the odds of point 2 to the power of 4; this is:

0.000 … [followed by 164 zeros ] … 01

This number is so small that it signals an event that cannot happen if these guys have the regular odds of being struck; they have to be chosen from a pool of individuals with a higher probability than the rest of the population of being hit by lightning.

Or that you believe God sent the punishment to them on purpose.

Following the line of reasoning, the curve states that early manufactured machines will have bugs. Then the bugs will be worked out, the design will be updated, and the new machines will be better than the old ones.

We know that the pool of refurbished is composed, by:

“new glowforges” that became defective.

old machines, born early in the curve, but retrofitted (renewed).

Truly new glowforges (I doubt it, but I’ve heard about it, so, say it is possible)

With these three premises, we can see that the bathtub curve predicts that the best performance of the new-refurbished batch would be, at best, as good as the new ones.

I don’t say that it can not happen. I’m saying that if we base our assumptions on the bathtub curve, we cannot conclude that the refurbished machines are better than the new ones.

Someone (@dan84 I think?) told me to ask for a hold of the machine in the warehouse, and I have done that since my second machine. I’m grateful for that pro-tip!

I for one remember playing support response time guessing game. Closest without going over won. Was anywhere from 4 days to 4 weeks around holidays for a SINGLE RESPONSE. Here or by email. I could be wrong but I think the responses are faster now…wether they fix the issue I couldn’t say, but they respond faster.