Already thought about that, will have an answer soon (@WurstEver, you taken a look yet?) If anything, the data on their website could be injected into this solution to give a more rounded result to compensate for the regions with lack of extract info...if you know what I mean...

 

It will be very interesting to look at this in detail. Additional (good) data is rarely a bad thing. And this looks pretty good. Bottom line is that everything that's been done in this thread has been based on extracts, so I wouldn't expect there to be any serious inconsistencies

 

So, I'm just starting to get my head around this, but the first thing that sticks out is that only some of the newly published information is related to the analyses presented in this thread. Breaking the new information down into sections, here's how I see it:

1. Reference summary panels: these map references and sub-references to production date ranges and production volumes. There is no direct mapping between serial numbers and production dates here. There are, however, some photos of absolutely bloody beautiful watches!

2. Table A: This information maps references and sub-references to serial number ranges and production volumes. There is no direct mapping between serial numbers and production dates here.

3. Table B: Here we get the first new information that is of direct relevance to the analyses presented in this thread. This table maps serial numbers to reference numbers and years of production for cal. 321 speedies. This information should be consistent with the information we have collated from extracts.

4. Table C: Same information as Table B, but for cal. 861/1861 speedies. As above, this information should be consistent with the information and analyses presented in this thread.

So, it might only really be Table B and Table C that have a direct bearing on the matter of how serial numbers and production dates are related. Testing for consistency should be a simple matter - just like the comparisons with Hartmann's table which kicked this all off. However, sadly, I'm not going to be able to do any number crunching on this for about a week

Finally, it's worth noting that the new tables still only give production date estimates to within a calendar year (or two... or three!). The methods that we've been working on here are still the only game in town when it comes to estimating production dates down to the month

 

Hear ye, hear ye, these folk got time machines I tell ye!

Thanks @WurstEver. Also spent a bit of time checking them out. What has crossed my mind : All 3 Tables (although Table A and B have the same info) could be used to eg. guess the watch reference the movement was put in. Table B and C might be able to be used roughly for estimation, depending on how one splits the overlapping SN's and Dates (i.e. 1962~1964 = 19.583.xxx ~ 20.525.xxx and 1964 ~ 1966 = 20.521xxx ~ 22.825.xxx -> clearly some overlap in both SN and Date, and there is a lot of this) but this would add a big guess factor...better done by actual contradictory extracts and a statistical best guess thereupon...

 

I have a little time available right now and I want to use it to say the following:
guys like you, your efforts and passion are the reason I love being here.

 

Using Tables B and C and some educated guess for gaps I was able to put together an estimate of s/n's and years as follows:

1957 15499xxx - 15747xxx
1958 15748xxx- 15996xxx
1959 15997xxx- 17301xxx
1960 17302xxx- 17761xxx
1961 17764xxx- 17765xxx
1962 18419xxx- 19583xxx
1963 19834xxx- 20520xxx
1964 20521xxx- 20525xxx
1965 20527xxx- 21673xxx
1966 22825xxx- 22825xxx
1967 24066xxx- 24537xxx
1968 25310xxx- 26555xxx
1969 26554xxx- 28420xxx
1970 28781xxx- 30306xxx
1971 31009xxx- 32192xxx
1972 31625xxx- 32240xxx
1973 32241xxx- 32856xxx
1974 32857xxx- 32859xxx
1975 31312xxx- 32124xxx
1976 37870xxx- 39185xxx
1977 38562xxx- 39253xxx
1978 39254xxx- 39945xxx
1979 39946xxx- 44818xxx There are probably no 40xxxxxx - 448xxxxx
1980 44469xxx- 44819xxx There is a lot of overlap between 145.022-78 and 145.0022 references
1981 44121xxx- 44820xxx
1982 44708xxx- 44822xxx
1983 45295xxx- 45168xxx
1984 45882xxx- 45514xxx
1985 46470xxx- 45860xxx
1986 47057xxx- 46206xxx
1987 47644xxx- 46552xxx
1988 48232xxx- 48244xxx
1989 48245xxx- 48257xxx
1990 48258xxx- 48268xxx
1991 48269xxx 48280xxx
1992 48281xxx- 48294xxx
1993 48295xxx- 48309xxx
1994 48310xxx- 48323xxx
1995 48324xxx- 48338xxx
1996 48339xxx- 48352xxx
1997 48353xxx- 48369xxx
1998 48370xxx- 48389xxx
1999 48390xxx 77009xxx

Note the 5 million gap in s/n's between the 145.022-76 and 145.022-78 references and overlap with the 145.0022's.

Also note from Tables B and C that some references with a year were actually first produced in the following year (e.g. the 105.012-63/64/65/66 series noted as first produced in 64/65/66/67). The author confirmed this is correct on the other thread: https://omegaforums.net/threads/spe...-from-the-moonwatch-only-website.58897/page-2

 

I've updated my calculations from the Moonwatch Tables B and C and used our data for interpolating big gaps (81-88 and 88-96):

1957 15499xxx - 15664xxx
1958 15665xxx - 16321xxx
1959 15831xxx - 17301xxx
1960 17301xxx - 17764xxx
1961 17763xxx - 17765xxx
1962 18091xxx - 19896xxx
1963 19126xxx - 20520xxx
1964 20211xxx - 21288xxx
1965 20524xxx - 22824xxx
1966 21676xxx - 23757xxx
1967 23681xxx - 24990xxx
1968 24689xxx - 27136xxx
1969 25621xxx - 29489xxx
1970 29489xxx - 30559xxx
1971 30559xxx - 32197xxx
1972 31471xxx - 32204xxx
1973 31933xxx - 32474xxx
1974 32395xxx - 32859xxx
1975 31312xxx - 35248xxx
1976 35248xxx - 39185xxx
1977 38389xxx - 38907xxx
1978 38908xxx - 39426xxx
1979 39427xxx - 44818xxx There are probably no 40xxxxxx - 448xxxxx
1980 44819xxx - 44819xxx There is a lot of overlap between 145.022-78 and 145.0022 references
1981 44121xxx - 45180xxx
1982 44821xxx - 45737xxx
1983 45737xxx - 46347xxx
1984 46347xxx - 46959xxx
1985 46959xxx - 47510xxx
1986 47510xxx - 48038xxx
1987 48038xxx - 48135xxx
1988 48135xxx - 48244xxx
1989 48245xxx - 48257xxx
1990 48258xxx - 48268xxx
1991 48269xxx - 48280xxx
1992 48281xxx - 48294xxx
1993 48295xxx - 48309xxx
1994 48310xxx - 48323xxx
1995 48324xxx - 48338xxx
1996 48339xxx - 48352xxx
1997 48353xxx - 48369xxx
1998 48370xxx - 48389xxx
1999 48390xxx - 77009xxx

Here's how Moonwatch, Hartmann and WurstEver (linear) compare:



Not bad except in the mid to late '70's.

 

https://omegaforums.net/threads/ilovemyspeedmaster-com-speedmaster-production-dates.59392/

We should continue analyses here...

Thanks to all those involved in this effort...looking forward to the continuation thereof with you.

 

I was wondering why the estimate for my 105.003-65 was still off by 11 months after I had uploaded my extract to www.ilovemyspeedmaster.com.
Now looking into this original thread again, I found these two extracts, that explained, what happened here:
25.449.122 25.01.1967 -> Germany
25.449.761 29.01.1967 -> Stock
25.449.777 04.01.1968 -> Germany

I am interpreting "Stock" as it was put on the shelf (BTW: 29.01.1967 was a Sunday?)
My assumption: These were all manufactured in January '67 or earlier and the dates given on the extract are actually the delivery dates from the factory. If papers for the delivery are missing, the next available data point back in the chain is used and e.g. the date of internal storage is used.

View attachment 381042

View attachment 381043

 

Edit: just read back through this very interesting thread and realised a lot of clever technique has been shown so please take the comment below as a first impression, not measured critisism in any way. The interpolation methods are likely the best we have since the data itself is dirty and may have been corrupted by human error at the archive. /edit

I realise these are best guess estimates and credit is due for going to the trouble but I would point out there are several verified extracts for 24m serials in 1966, particularly in Ed Whites. Serials for those jump from 22m to 24m in the space of a couple of months. Ed Whites seem to mess up the main pattern as so few were sold in later years. 145.012s too seem to jump about with no defined break between 67s and 68s. Overlap between variants and use of serials out of chronological order is the problem here, I don’t think a table by year can ever be perfect, a table by model might be more useful in many cases.

 

Hi @Georgieboy58

Thanks for resurrecting this thread. And thanks for submitting your extract. It was received, has been marked for inclusion, but has not yet made its way to the WWW

Here is the nearest data (to yours, which is 25449777 on Jan 1968) currently used :

25009701 1968/3/27
25443000 1967/11/8
25443800 1967/12/1
25445000 1967/12/1
25445219 1967/11/29
25445323 1967/11/22
25445328 1967/11/26
25445925 1967/11/29
25446800 1967/11/22
25446895 1967/12/15
25449122 1967/1/25
25449716 1967/1/29
26070000 1968/4/1

See how it jumps around?

By visual inspection the nearest extracts to @Georgieboy58 are Jan 1967, and it is clear that that is what (linear) interpolation should give. If I run this SN through some algorithms, here is what I get :

Linear Interpolation : 1967 / 1.0
Statistical method in use : 1967 / 2.0

Hardly a vast improvement, but a step in the right direction. The current implementation could be refined to include a broader neighbourhood (for better results) but the lack of later data prevents that...Anyhow, wrong is wrong

So why is this wrong? Currently 2 scenarios jump to mind...

1) This is the first extract in this SN range that is 1 year later than others

or

2) https://omegaforums.net/threads/someone-does-know-michael-collins-145-012’s-serial-number-extract-issue.74281/page-2

Hope this helps.

@WurstEver , @JMH76 care to comment?

 

Assuming 2), does anyone have the email adddress of the archive department to confirm whether my extract does not also have a typo?

 

I've also sometimes wondered about how consistent the meaning of the "Production Date" given on extracts is. If you're so inclined it would be really interesting to hear what you can find out from the museum about your movement number. Seems to me that it's going to teach us something either way: another observation to help us understand the rate of human errors in the process, or fairly good evidence that the meaning of production date may not have always been consistent. Very cool!

 

Extracts are known to contain errors sometimes. My own contained an error in the Watch Ref No., if you can believe that. It was corrected when I pointed out the error. If they can make that mistake why not production date? In their defense they are only human and there is a lot of manual look-up involved. It would be more difficult to challenge them on the date, however. It might just be a case of using a movement that was on the shelf for a long time; that's been discussed before.

 

Good to hear from you again, @WurstEver.

 

Cool idea and thanks for the tool! It will come in useful and is saved on my search engines favorites bar!