How to prevent duplicate keys in archive database?

Question

I am struggling with this problem.
Background: I'm working on a project where I have to make an archive database. The archive database should get
all data of the operational database. It should even save every update ever made, so it literally contains the entire history of the operational database (this is a must and the whole project revolves around this idea). So this is solved by using Change Data
Capture. After that the data should go through a staging area and eventually in the data warehouse database. I came out with a solution and worked it out in the prototype and it seemed to be working all fine. I stupidly forgot to include the foreign keys,
so the archive database didn't have the original structure but it should ofcourse (no wonder it went okay without too much hassle).
Problem: Because we want to store everything in archive, there will be duplicate primary keys (just for instance,
many same contact_id's because telephone number changes a couple of times). I thought to solve this by adding a new primary key which says is auto-increment and purely exist to make a record unique. But when it comes to foreign keys, it's impossible. You want
contact_id to be able to be duplicate and in that case it cannot be a primary key. But foreign key is only able to reference to a primary key or another unique key but not other normal columns.
Any advice on this? It's an absolute must to store all changes.

Answer 1

All of you, thanks for replying, I'm happy you're trying to help me out with this problem. 
Visakh and Louis, thanks that seems like the solution for this case indeed. Yes, the dimensional design appeals more to me as well.
I read the articles and watched some tutorials. But I can't work it around the solution that I had.
More background info: I use CDC to track all the changes done in the operational database and SSIS (following one of Matt Mason's tutorials and with a lot of alterations to make it fit for my project). I have this control flow (don't mind that
error haha):
(Oh apparently I cannot add images yet, so here's the link for the screenshot:) http://nl.tinypic.com/r/w0p1u0/8
Basically I create staging tables in my archive database next to my normal archive tables. Then start CDC control task to get the processing range and then it copies everything from the operational database (joined with a few CDC columns) to the staging
tables. After that the processing range ends so it will only get the rows it hasn't processed before. And then I do some updates on the staging tables and then finally insert everything into the archive tables. The staging tables then can be truncated. After
this the data will go to the staging area for transformations and then finally at last to the DWH. The reason for having a staging area between the archive and DWH is that the archive will not only be used as source for the DWH but also on it's own. The DWH
will not contain 100% the same stuff as the archive (like maybe some transformations, extra columns with calculated fields, plus some columns don't need to be in the DWH at all). When all the ETL stuff is done in SSIS, I have to use SSAS to define all the
facts, dimensions, cubes. 
Example: So I try to work with the SCD type 2. If I understood it correctly (and maybe I didn't): for example, the contact table in archive should have the surrogate key ID (the auto-increment one). The business key is the contact_id
and can be used uniquely with the time range columns. 
Following Visakh's post, the ID becomes the key that the foreign key will reference to. For example: 
Contact table:
ID: 1 contact_id: 100
Name: Glenn start_time: 2014-01-01
End_time: 2014-08-20
ID: 2 Contact_id: 100
Name: Danzig Start_time: 2014-08-20
end_time: NULL
Sorry, I couldn't style it as table somehow. So the employee changed his name. It makes sense that the time period tells when the first name was valid. 
Organisation table: 
ID: 1
org_id: 20 
Contact_id: 1
Start_time: 2014-01-01
End_time:NULL
(it references to ID instead of contact_id as suggested)
The employee belongs to an organisation. It references 1 which is still old data. But this is the last version of the organisation record. 
So then I need a table to link the 2: 
organisation_contact table
contact_id:100
org_id: 20
and then I need another one to join with the surrogate key?
ID: 1
org_id: 20
ID: 2
org_id: 20
(Guess it would make more sense to have org_id in the contact table but for now it's an example)
Problems: I don't quite understand how this works. From the example I saw you have to have another table (the fact table) to link it to the surrogate key. Would this mean I have to have facts and dimension tables in my archive database?
My intention was actually to have all records of the operational databases (all the updates too) in my archive. And after that create the facts and dimensions in the DWH with SSAS. The example looks like I should do it earlier. 
I don't know how to combine this with the cdc solution. I want to get all the data by using CDC. Like how every update gets registered in the accompanying CDC table. Then the archive will get the CDC data. But then how to combine this in use with SCD. I
have the surrogate key in archive (ID) and then I make the start and end time columns. I need to point all references to the ID and then make the other table to keep track of the contact_id (original PK) and another key. At last make another table to track
all the current data in the fact. 
Another question: Would you recommend the SCD task in SSIS. I read it was not that great if you have many rows to work with. What would you think is the best method to implement it. 
Thanks so much again.
EDIT: What about slowly changing dimensions type 4? It looks like you don't have to change the references of the foreign key then. Why do you prefer 2 over 4?