I’m just going to post the “canvas” created by the Amazon chatbot, and the text where it at last admitted to simple facts that would lead a seller to expect “100%” featured offer status for a very niche FBA-sold product with no competitors, and admitted that the percentage is not accurate enough to ever use less then 5% increments. (look up the Nyquist-Shannon sampling theorem for why)
Note also that it apologized for giving me utterly bogus reasons for the less-than 100% metric, which took about 15 mins of my time, because you have to wait while the AI makes up lies, it does not simply look at actual data and facts unless you cross-examine it into a corner, as I did.
Just curious if these “brief moments” ( gaping holes given the size and complexity of the Amazon grid) could be the causes of the ‘ghost’, ‘phantom’, or, my preference, ZOMBIE listings that pop up every so often.
They deny that they exist but there seem to be way too many instances to all be ‘operator error’ on the part of sellers (holding up hand to admit my flawless inventory tracking is sometimes flawed).
Our friend @lake (who was on the ground, in the front-line trenches when certain decisions were made) can explain the reason why the underlying situation which typically produces this phenomenon - Race Hazard (aka “Race Condition”) errors in Relational Databases - inherently exist, and why there’s really no practical way to counteract (much less prevent) them, better than anyone I’ve ever seen.
The issue is fundamental to “distributed databases” with multiple copies of the same data on multiple servers. There is no possible solution to the need for record locking across the database to insure all of the copies of the inventory count are updated when a sale is made or new inventory is added.
It is further complicated by automatic scaling by modern systems which bring multiple servers on and off line based on transaction load.
This was all known before “Computer Science” was a recognized “Science”. Computer science ignored most of the knowledge learned by programmers AKA Software Engineers, choosing to concentrate on the higher levels of computer software architecture. They assumed the plumbing required no attention. They suggested that some other form of engineering might adopt Software Engineers.
When I started to sell on Amazon, Amazon was still trying to find the cause of Zombie listings. I was among the sellers who worked with Amazon to try to isolate the cause. Needless to say, they could not reliably duplicate the problem, which is typical of race conditions. I shared my insight with a few of their programmers. Shortly later, all attempts to deal with the problem ceased.
I claim no credit for developing this theory. I was part of a team responsible for network strategy for a now defunct computer manufacturer. We marked the area of distributed databases as “off-limits” unless and until a breakthough in approach could be developed. There has been none. All attempts to date have been attempts to band-aid the failure, after it has occurred.
Thank you for that clear & concise explanation of this phenomenon, my friend.
Our Hitchhiker-admiring friend @ZaphodBeeblebrox’s message is one of the rare times I’ve seen Amazon admit that Race Hazards DO exist.
For instance, in a ¼-century of perusing the various iterations of the ASF (“Amazon Seller Forums”), stretching back to the Seller Soapbox Days, the only time I’ve ever seen an Amazonian 'fess up on the subject (albeit w/ the well-known “It’s not Amazon’s fault” disclaimer lurking below the surface) was the “Multiple Exchanges” reply made by ‘Original Glenn’ on 041916*, back in the Q1 2012/Q1 2018 “Age of Jive”:
It’s possible (maybe even probable) that the URL is still hosted somewhere, such as in this or that publicly-accessible archive, but since I’m aware that the problem isn’t practically-solvable, I’ve not bothered to check…
Here is an example of an exchange I had with Co-Pilot, Microsoft’s version of Chatgbt.
I saw an ad on a game I was playing on my phone for a creme that claimed to relieve the pins and needles and cramps associated with neuropathy,
I asked Co-Pilot if the chemicals in the creme worked to relieve the symptoms.
It told me that it did. I looked at the references it provided for its answers. There were 5 sites, Two were sites I had never heard of but upon visiting them, there no evidence that there was any science or even personal experience represented on the site. The other three sites were selling various formulations of the chemicals.
I asked Co-Pilot it it thought its sources were credible.
It told me that the three sites which were selling product might not be, but the other two were credible.
I question it on why it thought those sites were credible, and it changed its opinion.
BTW
I bought the cheapest formulation of this product on Amazon, under a tenth of the cost of the product in the ad, and it worked for me.
AI does not think. Like any new function they introduce on Amazon, the thought process is incomplete, and the function fails to stand the test of the real world or an advocate of the real world, much of the time.
Companies with an executive suite full of Harvard MBAs paid me outrageous consulting fees to tell them how to solve problems in their business. I found them the answers by finding the people in their organization who knew the business and interviewing them and translating their knowledge to what the MBAs could understand.
