Roman Hartmann's table; often cited, sometimes maligned, recently described as a "hand grenade" by a very well respected member of OF. I had some time to kill, so I decided to see if there was a better way to map vintage Speedmaster serial numbers to production dates based on extracts I found online. MWO and Speedmaster101 give serial ranges and production date ranges for each sub-reference. But this is a subtly different matter. Is the relationship such a mess that it's pointless to attempt to map serial numbers and production dates directly? The following is what sometimes passes for fun at my place

I found 40 full or partial images of "Extracts of the Archives" (and earlier "Extracts of the Records") online for Speedmasters in the 321-861-1861 family where production date, reference number, movement number, and at least the first part of the serial number could be discerned. At that point, I stopped because I ran out of patience. For those where only partial serial numbers were shown, I replaced unknown digits with zeros. I transformed production date into months since January 1957 and plotted serial number by production month number to produce the scatterplot below.



The interesting thing about this plot is that for at least some references, there seems to be a fairly tight linear relationship between serial number and month number. The exceptions are the very early 29xx models with serials in the teens (green circles, bottom left) and the later 1861s with serials up in the 48s (red triangles, top right). These sit below and above the line respectively (the latter fairly dramatically). Given this, maybe it's possibile to predict production date with reasonable accuracy as a simple, linear function of serial number if we restrict ourselves to Speedies produced in the 60s and 70s. So, I removed the early and late references, leaving only 105.003s, 105.012s, 145.012s, and 145.022s produced between 1964 and 1980 (n=35) and fit a regression line to the data as shown below.



This is not a bad fit. So next I compared the performance of the linear model with Hartmann's table at predicting the production date of the watches in the sample (more on the wisdom of doing this below). Hartmann's table correctly predicted 16 of the 35 production years (46%). The linear model outperformed the table, correctly predicting 24 of the 35 production years (69%). An added advantage of the linear model was that it also enabled prediction of production month. Hartmann's table only gives an estimate to within a calendar year, or sometimes two. The average absolute error of the model in predicting production month was 3.91 months. The maximum was 11.71 months. Not too shabby ... BUT this is almost certainly an overly positive outcome, since the model will of course fit the data set that was used to derive it reasonably well. The real test is whether such an approach can predict production dates for watches that were not in the sample more accurately than the table.

What's the take-home? Well, I wouldn't put too much stock in the specifics of this particular equation, but the approach suggests that at least for Speedies produced in the 60s and 70s the relationship between serial number and production date is at least not a complete mess. In fact, it may be quite predictable. However, there are big caveats around this. For example, 40 watches is a tiny sample, who knows what factors influence the tendency of people to post pictures of their extracts online (indeed there may be a "file-drawer" effect where people keep odd results to themselves), Google played a role by promoting some images to the top of the search, and again, assessing the performance of the model by using the same data set that was used to derive it is not a great way to go about this.

Another possible take away is that this helps to reinforce what Omega probably means by "production". For example, the accepted serial range for 145022-74s is approx 31-39xxxxxx (as an aside, I have one slightly below this range which myself and another knowledgable member believe to be original). The linear model predicts that an extract would give the "production" date of a watch with a 31xxxxxx serial as approx 1971. Yet obviously, 145022-74s were not made until well after this. The implication is that "production" on the extract refers to when the movement was made; which may have little to do with when the watch was assembled, shipped, and sold. Is that what Hartmann meant when he constructed his table? Is that what most people mean when they talk/think about "production" dates? Not sure. I'm also not sure why the relationship between serial and production date would have changed so dramatically when the 1861s were introduced. Maybe Omega changed the way they counted?

So that's how I killed a couple of hours yesterday afternoon. I'm just an enthusiastic novice, so maybe this is all obvious? Or pointless? Maybe someone has tried something like this before? I didn't check. But if so, I'd be keen to see what the results were. If members find this interesting and would like to share the details of their extracts I'd be happy to incorporate new observations into the data set and see if this goes anywhere.

 

Cool stuff.

I wish there was more data in the lower end of those serial numbers.

I don't know how they do it, but if you add up all the numbers in a speedmaster movement number and divide it by the day of the lunar cycle, it will come out to 666. No idea how that works.



Have you considered overlaying the references to see how it lines up with mwo and sp101?

 

Problem is serials where not put into production in order. Hence a very late 39m -76 with the accepted-74 dial.

My understanding was they had the movements pre made and in storage and just used any container that coincided with forecasted production. Hence they are out of order inside each sub reference.

 

I may be mistaken, but I think that theory has fallen out of favor.

Here's some info, https://omegaforums.net/threads/omega’s-extract-of-the-archives-–-how-it-works.42587/

 

Very cool. I'd like to see more stuff here like this and Andy K's work on the Speedmaster 125.

 

Hey, thanks, that's a good idea! But something I've learnt through this exercise is that often the extracts don't show sub-reference numbers. So I could do an overlay at the reference level (eg., 145012 vs 145022), but not really at the sub-reference level (eg. 145022-69 vs 145022-71). While this analysis is addressing a slightly different question, I just cross-checked with Speedmaster101. Everything I found was consistent with the ranges given there apart from just two watches, which are interesting for reasons written in this article: http://speedmaster101.com/blog/145-003-whats-the-story/

These examples (the Jan 29th 1967 one is from the very article linked above) are just a whisker outside the top end of the serial range given on Speedmaster101 for the 105.003. But it's so close as to barely be worth mentioning. Bottom line is that everything I have seen in the course of this analysis supports those ranges.

 

My #2.49mil was produced in June 1967, much later than your #2.5 range . Uhmm

 

Hey cool! Many thanks! Nice application of curve fitting! I never thought about applying it this way

Two things : 1) you could increase the size of your data-set by reaching out to the forum members to PM you their EFTA...must be hundreds if not thousands of EFTAs just in OF. Hell, there are probably individual members with as many watches

2) I'm interested in the non-linear spike near the end of your data (i.e. the more modern watches), not because modern is cooler than vintage, but for completeness sake (modern Speedies equally deserve love). Could this be caused due to at some point (perhaps the 70,s/80's) Omega modernizing their production line, and thus producing more watches then they sell? Perhaps more data here will yield in a model for decent prediction...

 

I think you're all good. 2495xxxx is well within the range given on Speedmaster101 for a 105.003 (it should be a -65 according to the table). And the model I presented in the OP gives the following estimate:

Production month = (7.6916*24.950000)-63.8171 = 128.09
Month number 128 = August, 1967

I've rushed through this a bit in my excitement, but I *think* the model is off by just two months in the case of your watch. Thanks! This is the first test of the model with a watch outside of the original sample

PS - gorgeous watch! I'm green with envy!

 

Then go ahead, how does this one fit in ? It's a 105.012-66, 254438××

 

No worries! I think you're right on both counts. It would be great to get some more data. To that end, if anyone would like to either post their extract details or PM me, I'll add them to the data set.

The 1861 spike at the end is weird, isn't it? It would be interesting to get more data to see if it's indicative of a change in slope, intercept or both. That might give us a clue as to what's driving the change. Of course for my first stab at the problem I took the easy approach to dealing with non-linearity ... that is, I ignored it

 

Hey, thanks! Let's see, this time I'll show the working ...

Predicted month number = (7.6916 * 25.443800) - 63.8171 = 131.89
Predicted Year = 1957+ROUNDDOWN(131.89/12,0) = 1967
Predicted Month = ROUND(MOD(131.89,12),0) = 12

So it looks like the model is pretty much spot on for your watch! Of course, I'm rushing through this a bit now, so if anyone can see a problem with the working, please do let me know. You won't hurt my feelings ... much

 

Cool!

How does a 145.012-67 with a 26.074.xxx fit with a production date of May 20 1968 fit?

 

All righty, let's see ...

Predicted month number = (7.6916 * 26.074000) - 63.8171 = 136.73
Predicted Year = 1957+ROUNDDOWN(136.73/12,0) = 1968
Predicted Month = ROUND(MOD(136.73,12),0) = 5

Looks pretty good!

 

Sorcery!

 

Outstanding research. I'll pm you the serial/production dates I have as on as I can dig them out.

 

Here is one I found (sorry to say this aint mine)...

 

Would you believe mine was produced six days later! It is a 65, Simon, who was at STS then, showed me the case back when I took it for servicing in March. It's odd that an earlier production date on a 145.003 has higher reference number. This is a fascinating topic, I wonder if any other OF members can get closer numbers/dates for their Speedmasters?

 

Off topic but congratulations on your 200th post!

 

True that! Thanks @SouthernScot. I am new to forums, but honestly its quite easy here...I am sure you will be there in no time...