Ex-Numis Success Rate Question

CT Colleagues
1. I write to ask if members could share their success rates with the Ex-Numis provenance finding service.

2. Over the last few years I’ve submitted 29 coins to them and received one provenance, which was to a 1987 NFA sale. My submissions are usually in the USD 1 000 to USD 5 000 value range. I’ve sent them a wide range of types, from striated electrum to a late Roman solidus.

3. Is this rate of finding provenance, 1 in 29, typical?

Thanks for your consideration

 

Log in or Sign up to hide this ad.

Would you mind sharing where did you buy the majority of your coins from? Looking for provenances for EL coins or owl tets that came out of the recent hoards will turn out negative results.

Exnumis seems to not be working (at all?) for very old auctions. I also never had any success with fractions or in general greek coins less than 4g of weight. 1 in 29 is rather bad but I wouldn't expect a success rate over 10%.

 

@pprp, my coins are purchased from a wide range of sources. Some of the submissions are from CNG, NAC, Roma and the like, while some are from dealers at shows. My purchases span the last 30 years.
D

 

I have submitted 26 coins thus far and recieved 5 successul provenances. Your results seem about typical, or slightly below average on ex-numis from what I've been able to gather.

 

I have not submitted coins to ex-Numis.

FYI: 75 minute video by Dr. Jonas Emmanuel Flueck discussing ex-Numis and how it works.

 

Another factor is photography: are you taking pictures yourself or using auction house photos? Ex-Numis operates on image recognition and even slight differences in photos can cause positive or negative results.

It's found pedigrees for me on coins that sold recently at auction houses where I know they send everything through Ex-Numis.

As for an overall success rate, for me, it's found 23 pedigrees on ~120 submissions. Of these, the earliest were 1907, 1910, a couple in the 1920s, and the bulk of the rest from 1960-1990.

It is also far from perfect and has missed some very obvious pedigrees and some others which I've stumbled across. But, it's certainly worth the effort in my eyes.

 

Scarborough, it seems Faire. An old Sage (perhaps it was Rosemary, or maybe even Ancient Joe?) once said, "Thyme may yield satisfactory results, though events may be s..parsley occurring!"...or something like that...it was in Old English...

...apologies to Paul & Art...oh heck, apologies to everyone (It popped into my mind & I just couldn't let this one go )

 

I tried ex-numis a few times without any luck at all, usually with pictures from venues like CNG that I'd assume would be the best for finding potential provenances. If submissions were cheaper or I had enough coins in my backlog to justify a premium subscription I might submit more but at ~$5 each it's not quite worth it to me at the moment.

 

It would be interesting to see a test to see how they handle the same coin shot by several different photographers, in several different ways and using images input to their system directly as opposed to scanned from old paper catalogs or plaster cast image plates from the early days. I would not expect them to recognize a coin from a Dattari-Savio pencil rubbing but could they identify a cast fake imaged in one catalog from its original in another? Certainly this could be set up but that would be a considerable expense for users to do. Have the providers of this service published any such tests? Has anyone received a positive report that they could tell was wrong or are all the failures just missing things are were out there but not identical enough for their system to handle?

 

I tend to only run those coins through that look as if they “should” have a pedigree, so my success rate is somewhat higher. That said, ex numis has missed a very large number of pedigrees, particularly to older sales, that I have found manually. I think it is a valuable first step, but I don’t treat a lack of a hit as definitive.

 

I have submitted 7-8 coins with the same criteria as @Cicero12.
Here is the one coin with a provenance that was identified by Ex-Numis:


Kyrene AR didrachm 290-280 BC 6.84 gm, 20 mm
O: laureate head Apollo right
R: silphium plant type III, wreath lower left, jerboa? lower right
BMC 262c (not illustrated)
Provenance: M&M FPL 215, lot 2 09/01/1961

The irregular flan edge probably helped with the identification. I was pleased with the result as I have never looked through old M&M FPLs on my own. I do like the feature that I can add any known provenances to my submission

One of my major collecting interests is finding lost pedigrees and I have spent many hours looking for matches. So, when I submit a coin to Ex-Numis, I have already searched for the coin in hundreds of old catalogs and published collections.

 

I wasn't aware of this service, though I'd probably not use it, since the vast majority of my coins are not great rarities. In fact I could probably count the very rare coins on one hand.

If they are attempting to link images of a coin with archived images, I see two problems. One, as already mentioned, is the quality of the image being submitted by the coin's owner. Lighting, angle and resolution can vary, unless one has a sophisticated set up. Two, archived images can be notoriously fuzzy, lacking good resolution. I think, even using "state of the art" technology, such linking is quite problematic.

I have one archaic Athenian tetradrachm, from Harlan Berk, that has a provenance going back to a M&M auction in 1946. Perhaps this service would work for super rare coins, with only a handful of known examples - I don't know....

 

I just signed up and they give you 25 credits, essentially 1 provenance from the 70s- current, for free. If it's pre1970s then it's 50 credits.
Interesting process. But with all the folks here that utilize it I figure why not check it out.
The coin already has a provenance to the early 2000s, but I'll be interested to see if it garners any previous results.

 

I agree, and, in a perfect world, it'd be possible to browse near-matches to find possible hits rather than assuming the algorithm's parameters are set perfectly.

Take this aureus for example. The first image is from the sale where I bought it; the second I took:




The auction house's image is more than acceptable in my eyes. Especially knowing that Ex-Numis uses the shape of the coin, I'm surprised to see any difference in efficacy.

However, I sent both images in and it found a match for my photograph (Ex. Munzen & Medallien AG Auction 28 lot 327 (June 19, 1964)) but no matches for the auction image.

Then, through some catalog browsing, I later found it as Sir Arthur John Evans, Ars Classica XVII, Oct 13, 1934, lot 1365.

So, it's ultimately not perfect (unsurprising for an AI-based system) but better than nothing and often gives a decent hint for where to start looking.

 

If you sent the images at the same time maybe they processed only one as they realized it was the same coin?! Otherwise it's pretty weird. Did you try finding the provenance in acsearch (paid option), in theory they have the Naville catalogs.

 

I wouldn't bother submitting coins from one of the three sources you mentioned. I had 0% success rate and for good reasons that unfortunately became clear to me too late. The 0% didn't change even after checking manually nearly 10 thousand catalogs and FPLs.

 

Yes, 1 in 29 is my typical exNumis success rate for high quality Roman Republican, and that too to auctions I can easily access anyways such as NFA or Sternberg. Good concept, not great image matching in my view, not great database in my view, tho one can't prove either as the extent of their database is unpublished

 

Thank you for the video link, the presentation was very informative, and long.

 

So, having viewed the video, I must say that the technology of digital recognition, coupled with archive digital images, is an interesting approach to establishing provenance of a heretofore coin with little or no provenance.

The presenter does talk about the issue of import and export restrictions of certain ancient coins. There is labyrinth of regulations and laws, stretching over decades, both for the US and internationally that were established to protect against the illegal export of antiquities, including coins.

Now, enter this new technology. My questions are these:

Given the nature of these laws and how they were written and established, would they allow provenance documentation using digital technology to establish that a given coin meets the criteria of so many years prior to exportation to qualify for legal importation? In other words, if a squabble between a government authority, the seller and the buyer occurs, would proof of provenance based on digital image links be accepted by the authority? Has this ever been tested? Would it be definitive and legally accepted, or would the regulation need to be rewritten to accommodate this technology, provided that governmental and international authorities "buy in"? Would there need to be a test case?

So much of law is based on precedent, it seems to me that such a case, at least for the US, would be necessary.

Now, I think it would be nice to have provenance of a coin going back to the early time of the old collections and catalogs. It would certainly make the coin more saleable. So I might actually try using their service once or twice to see the results. With their sleuthing, who knows what might come out?

 

John Nebel wrote an article for The Numismatist in 2017 where he discussed the challenges associated with getting some coins he purchased through customs. An auction provenance was sufficient for one of them. Given the unique nature of ancient coins, I don't see why a record from a catalog wouldn't be accepted. You can read about his fight with CBP here: https://blog.money.org/coin-collecting/head-to-head