Can someone tell me what a "Compugrade" slab is and why they're worth so much more than the coin?

Apples and Oranges. There is a premium for slab collectors, not coin collectors, supposedly. As I said before, somebody will pay a premium for any junk, and call it “scarce.” This is not numismatics.

 

Log in or Sign up to hide this ad.

One man's trash is another's treasure? It is interesting, I have heard of NGC black holders, PCGS Regency, Rattlers, and Doily, but this is a first.

 

Why have a problem are you better than the slab collector?

Slapping down 2 Cents for that one.

 

No, but it isn’t numismatics. Anybody can collect anything. Am I better? That is kind of a provocative statement. I’m just a coin collector. There are sites for other curio collectors.

 

Coins and coins in slabs are definitely numismatics. They aren't a curio.
No need to beat anyone down because they have a collection of slabs.

 

It is a piece of short-lived plastic, not a coin. Hey, if we can debate the validity of Dan Carr’s work as numismatic, we can certainly debate the validity of a piece of plastic, made by a basement slabber, who didn’t last long in the business.

 

And this is the fourth time we have heard your position. Got it.

 

Compugrade certainly wasn’t a basement slabber. They didn’t last long as the technology was not fully developed (computers in 1991 were no where near what they are in 2019) and was too expensive. Plus there were issues with computer grading, specifically toned coins. They did have a legitimate idea with patents (one that even PCGS was working on). They also were covered and had ads in major numismatic magazines.


http://www.coingrading.com/compgrade1.html

 

So, it is the 4th time we’ve heard your position. Didn’t know this was a contest.

 

I was reading that they would take multiple digital photos before running it thru their program.
In your examples how close to a correct grade did they get with the tech they had?

 

I will admit the grade is close. For the OP’s coin, I would have graded it 63.

 

I’ve only had 62s, 63s, and 64s (to various decimal points). They all seemed appropriate (many would stay the same now while a few may have gone a point up or down). And the higher decimals were slightly better (i.e. a 63.8 looked nicer than a 63.2). There were some inconsistencies, as one would expect when going to 1/10 points (but that could be due to my interpretation and involves some nitpicking within the decimal structure).

I can’t speak to the PL or DMPL as I’ve only seen one picture of each, but it should be noted that the standards for those designations were different in the late 80s/early 90s.

 

This kind of technology would probably have worked well today, with more sophisticated computers and algorithms. I bet PCGS and NGC have considered the option of automated grading.

 

If you take a look at the link, PCGS was also developing this technology in the early 1990s (PCGS Expert).

It probably would work well with untoned Morgans, Peace Dollars, Eagles, and some other coins. Still it will likely cost more than the TPGs want to spend. Plus, if used properly, it could limit crackouts and regrading. That sounds good, but might not be financially wise for the grading companies.

 

I agree. With an untoned coin, probably minimal problems. I doubt that a computer would be able to do special designations, such as PL, or DMPL, where there is a human judgement call necessary. I knew PCGS did some work in the computer grading area in the 90s, but, yes, it was financially impractical. I still wonder if we will see some form of automated, computer generated grading at least as an option in the future, from the “Big Two.”

 

I'm becoming less and less of a believer in the special status of "human judgement". I think we're within shooting distance of trainable AI systems that would render judgments comparable to, and more consistent than, the best human expert graders.

The problem is that nobody wants consistency. They want to buy coins that are undergraded (so they can pay a low price based on the assigned grade), and they want to sell coins that are overgraded (so they can get a high price based on the assigned grade).

Once they've bought an undergraded coin, if they ever want to sell it (or even put it in a registry set), there's a temptation to resubmit it for a higher grade.

So: collectors see bargain-hunting opportunities, dealers see upgrade opportunities, TPGs see resubmission income opportunities. Who, exactly, is going to push for a more consistent system?

 

If you read their patents, it's a vastly different technology than today's neural networks. Back then they were mapping defects against their position on the coin and assigning weights to them. So a gash on the cheek was more of a deduction than a gash in the field behind the head.

Their tech would work as long as you create a map for each type of coin.

A neural network needs a lot of training data - makes it much harder to train for infrequently seeen coins...

 

Of course. But how do humans learn to grade those infrequently-seen coins reliably?

 

You take a bunch of heuristics and apply them to the specific new situation...

Minimal wear on one high point = AU58
Monimal wear on two high points OR slightly more than minimal on one high poont = AU55

Now you can grade a new (to you coin) by learning what the highest points are...


The classic way to collect these heuristics is to have an analyst sit with the expert and ask WHY on every decision.

The problem is this is incredibly annoying to the expert and usually gets less that perfect information. Also leads to short life expectancy for the analyst.

 

...or haul in the topographic scans of a few sample coins and build a precise map of feature height for every ten-micron-square patch of the design.

The problem is that the expert sometimes thinks a decision was made for one reason, when it actually wasn't. Expert explanation is almost always useful, but it isn't 100% reliable, and it may not be 100% necessary.

We all underestimate the power of effectively unlimited patience, focus, and recall. We're living in an interesting time where AI systems are starting to shine a spotlight on that power.