I have been experimenting with ChatGPT to predict potential dates of manufacture (more accurately stated the date the serial number was assigned) for Model 94’s and found it to be very good at estimating the month and a range of days within the month of likely assignment. It is able to do this with a very small data set of known dates. Nothing trumps the factory letters, but if you want a pretty good idea of a month/day/year while you wait on official word give it a try. Here’s a basic prompt you can paste into ChatGPT to set the baseline, from there you can simply ask, “when do you think serial number xxx was made?” You can also provide it with additional data (e.g. Serial number yyy was assigned on Month Day, Year, and tell it to re-estimate a specific serial number.
—-Here’s the prompt…have fun with AI—-
“I’m researching serial numbers and their assignment dates based on historical production timelines. I have the following confirmed serial number data and corresponding assignment dates. Please use these details to estimate assignment dates for other serial numbers when I provide them, considering that serial numbers were typically assigned only on weekdays, excluding weekends and federal holidays in the United States:
- Serial number 471 assigned November 1894
- Serial number 1368 assigned December 29, 1894
- Serial number 3543 assigned March 1895 (estimated late March)
- Serial number 4762 assigned June 29, 1895
- Serial number 168420 assigned end of 1902
- Serial number 175233 assigned July 1, 1903
- Serial number 186044 assigned June 3, 1903 (estimated)
- Serial number 190079 assigned August 28, 1903
- Serial number 194285 assigned September 28, 1903
- Serial number 199578 assigned December 5, 1903
- Serial number 204565 assigned December 31, 1903
- Serial number 330452 assigned December 19, 1906
- Serial number 354000 assigned May 13, 1907
- Serial number 908373 assigned July 27, 1920
- Serial number 998275 assigned January 19, 1927
When estimating dates, please exclude weekends and federal holidays such as New Year’s Day, Independence Day, Labor Day, Thanksgiving, and Christmas Day. Based on these confirmed data points, please help me estimate assignment dates for other serial numbers I may ask about.”
I’ve also been experimenting with it more and more and find it pretty capable at things I didn’t know it could do. It made a complex chart for me that I replicated in excel, but was a lot of work!
I’m pondering a paid subscription so I can use it without limits and for work stuff….I wonder if we input all of the known dates for several, does it remember and get “trained” for future use with everyone?
Based on my examination of the PR records, your request to “exclude weekends” is not accurate (correct). The Winchester factory worked 6-days per week (Monday – Saturday), minus the holidays you mentioned.
Further, how can this AI accurately predict/calculate dates of manufacture when the annual totals are vastly different from year to year?
Many years ago (while on a research trip to the CFM), I recorded the year ending serial numbers for all of the Winchester models that had PR records. I then used that information to create the “When Was Your Winchester Made” resource look-up tables found here on the WACA website. For some models, I recorded the month ending serial numbers for each year of production. The results of my research clearly showed that there was absolutely nothing close to linear production. To give you an example, how would this AI predict when Model 1894 s/n 725000 was manufactured versus when Model 1894 s/n 825000 was manufactured?
The next issue with the AI is the ability (or lack thereof) to account for all of the Model 55 and Model 64 rifles with serial numbers in the same sequence as the Model 94.
In your list above, the date you have for s/n 175233 is out of sequence with the preceding and following numbers, and it is not close to being accurate. Per my records, s/n 175233 was manufactured in February 1903.
Bert
WACA Historian & Board of Director Member #6571L
Bert,
Thanks for your input and points. The AI is only as good as the data it has to operate on and is by no means perfect. #194285 is mine and I was playing around with it to predict when the serial number was likely assigned. I used your tool to effectively brute force the last number assigned in 1903 to set an upper boundary and then input the other information I had based on users posts to this forum, factory letters available on the internet, and information in Renneberg’s book. I will say that it did predict that 194285 would have been assigned in the October timeframe and if I went back and expanded the production to six days a week, it may have suggested late September / early October. Anyway, I just thought it was fun and a way for folks to experiment with AI and its many potential uses.
I also just picked up s/n 3543 and ordered the factory letter. The AI is projecting a March 1895 date, I’ll update this thread in a few weeks when I receive the official letter.
Alaska94 said
Bert,Thanks for your input and points. The AI is only as good as the data it has to operate on and is by no means perfect. #194285 is mine and I was playing around with it to predict when the serial number was likely assigned. I used your tool to effectively brute force the last number assigned in 1903 to set an upper boundary and then input the other information I had based on users posts to this forum, factory letters available on the internet, and information in Renneberg’s book. I will say that it did predict that 194285 would have been assigned in the October timeframe and if I went back and expanded the production to six days a week, it may have suggested late September / early October. Anyway, I just thought it was fun and a way for folks to experiment with AI and its many potential uses.
I also just picked up s/n 3543 and ordered the factory letter. The AI is projecting a March 1895 date, I’ll update this thread in a few weeks when I receive the official letter.
S/N 194285 – 9/28/1903 (late September)
S/N 3543 – 2/28/1895
I would like to add s/n 3543 to my research survey. Is it a 1st variation, or a 2nd variation? Caliber? Configuration? When you get the CFM letter for it, please let me know what the “Received in warehouse” date is, and the Order #.
Bert
WACA Historian & Board of Director Member #6571L
Bert,
s/n: 3543 is a first variation receiver. It was without a barrel when I got it so I will have to wait on the factory letter for the official word on caliber and other details. I expect it is going to be a 38/55 as the left cartridge guide has the spring loaded plunger, but we’ll see.
I posted the details about 194285 on your survey post, I know it’s not in your current bucket of interest, but figured I would capture it somewhere. Maybe you can help shed some light for me, the letter indicates it was a rifle (I asked the Cody folks to double check for me), but the receiver has a saddle ring. The hole for the saddle ring appears to be factory, or else done by someone that was very skilled. I’m just perplexed by the saddle ring rifle.
Alaska94 said
Bert,s/n: 3543 is a first variation receiver. It was without a barrel when I got it so I will have to wait on the factory letter for the official word on caliber and other details. I expect it is going to be a 38/55 as the left cartridge guide has the spring loaded plunger, but we’ll see.
I posted the details about 194285 on your survey post, I know it’s not in your current bucket of interest, but figured I would capture it somewhere. Maybe you can help shed some light for me, the letter indicates it was a rifle (I asked the Cody folks to double check for me), but the receiver has a saddle ring. The hole for the saddle ring appears to be factory, or else done by someone that was very skilled. I’m just perplexed by the saddle ring rifle.
Based on the serial numbers in the immediate vicinity, s/n 3543 was most likely a 38-55 Standard rifle with a 26-inch octagon barrel, full length magazine. The fact that it is a 1st variation is the important piece of information. Currently, I have documented (587) of the 7,999 serial numbers, and (282) of them are 1st variations.
Bert
WACA Historian & Board of Director Member #6571L
I agree with Bert and see many flaws with trying to use ChatGPT to predict such things. A lot of variables to consider. A quick thought would be if you tried this for different models, one factor would be Labor Day wasn’t a federal holiday until June 28, 1894. Then you have the production during war years skewing things way out of whack. The model 95 prime example, most of the model was produced for the Russian contract musket. I suppose the list goes on and on with other examples I can’t think of.
I have a suspicion that the factory may have even worked on “holidays”. I don’t have any specific proof as such, but I do happen to own a factory drawing that is dated Christmas Eve 1889.
Sincerely,
Maverick
WACA #8783 - Checkout my Reloading Tool Survey!
https://winchestercollector.org/forum/winchester-research-surveys/winchester-reloading-tool-survey/
Alaska94 said
Bert,I posted the details about 194285 on your survey post, I know it’s not in your current bucket of interest, but figured I would capture it somewhere. Maybe you can help shed some light for me, the letter indicates it was a rifle (I asked the Cody folks to double check for me), but the receiver has a saddle ring. The hole for the saddle ring appears to be factory, or else done by someone that was very skilled. I’m just perplexed by the saddle ring rifle.
I cannot shed any light on your Model 1894 without first examining it. Clear detailed pictures of it might reveal some clues. I would also like to see the ledger entry information for serial numbers 194284, 194285, and 194286 to determine if any of them match your gun’s configuration.
Bert
WACA Historian & Board of Director Member #6571L
November 7, 2015

I am assuming AI is making the assumption that production rates were consistent but we know that during many periods they were not.
Mike
TXGunNut said
I am assuming AI is making the assumption that production rates were consistent but we know that during many periods they were not.Mike
That is my primary issue with it… and I do not care for numerical “assumptions”. I much prefer to plug verified numbers & dates into an Excel spreadsheet, and then create the proper formulas that allow it to accurately calculate or extrapolate the information. The results can then be readily converted to bar charts or tables to visually present the information.
Bert
WACA Historian & Board of Director Member #6571L
Not a fan of AI – While I see lots of folks using it these days (instead of their own brainpower), I feel its predisposed to assumptions or predictive models that are only as good as the person making/designing the query and the data its using to generate the reporting, not to mention it can be biased or agenda driven. IMHO, Garbage in, garbage out.
1892takedown @sbcglobal.net ......NRA Endowment Life Member.....WACA Member
"God is great.....beer is good.....and people are crazy"... Billy Currington
Very true, for this purpose, it would just be doing math and averages…no different that becoming a random number generator. Now, having an accurate database programmed into it (like the WACA tables) would be pretty cool…just a single prompt to ask about your rifle and it ask the appropriate follow-ups, etc…..probably one day.
November 7, 2015

I’m convinced a majority of the ad copy and political speech word salads we endure are AI generated and it troubles me they are apparently somewhat effective. I spent the last several days researching a distasteful subject and when I recognized AI I went on to the next source. I suppose if AI had a few hundred verified DOM’s each year for each model they could likely come within a week of the correct date. We’re studying history here. When original source documents are available we should use them.
Guess how many fingers I’m holding up, AI.
Mike
Bert H. said
Alaska94 said
Bert,
I posted the details about 194285 on your survey post, I know it’s not in your current bucket of interest, but figured I would capture it somewhere. Maybe you can help shed some light for me, the letter indicates it was a rifle (I asked the Cody folks to double check for me), but the receiver has a saddle ring. The hole for the saddle ring appears to be factory, or else done by someone that was very skilled. I’m just perplexed by the saddle ring rifle.
I cannot shed any light on your Model 1894 without first examining it. Clear detailed pictures of it might reveal some clues. I would also like to see the ledger entry information for serial numbers 194284, 194285, and 194286 to determine if any of them match your gun’s configuration.
Bert
I started a new thread here <https://winchestercollector.org/forum/winchester-rifles/winchester-1894-rifle-with-saddle-ring/#p154702> on this specific topic and included a link to some pictures. Thanks!
TXGunNut said
I’m convinced a majority of the ad copy and political speech word salads we endure are AI generated and it troubles me they are apparently somewhat effective. I spent the last several days researching a distasteful subject and when I recognized AI I went on to the next source. I suppose if AI had a few hundred verified DOM’s each year for each model they could likely come within a week of the correct date. We’re studying history here. When original source documents are available we should use them.
A great deal of internet content, including videos, posts, and comments on (some) forums and YT videos are all bots and trash posting. AI is really going to muddy the waters in the next few decades….will be interesting to see how far it advances in the years I have left.
1 Guest(s)
