Have you considered looking at the usage (audit) and determine whether the reports are actually used?

Several years ago, I had a client in a similar situation (a little bit smaller at around 2000 reports).  By looking at the audit data, we were able to get them down to around 300 reports.

We put some rules in place in order to determine if it was something worth keeping or archiving.  For example: if the report ran once in the past year...we would determine if it was an annual report.  If it was not, we would archive it.  A lot of the reports turned out that someone ran it to see what it was and did not use it again.  

I am pretty sure we looked at a 3 or 4yr period and not just 12 months.  One simple rule was if it was not used in 36 to 48 months...it was automatically archived (no questions asked).  From what i recall, that one simple rule removed 50 to 60percent of the reports.

Going forward, the client continued to look at the report usage to help manage the overall count.  Today they are hovering around 400 to 450 reports.

After a little over a year, the archived reports were deposed of (aka deleted).

Hope this helps.

one idea is to download the specs using the REST API and then use an AI to identify similarity.

In detail: 
using the REST API you can manage to get all information about all objects which are in the team content folder (with java sdk I guess even what is in my folders)
Then you know all reports, reportViews with ID and search path. It is also possible to find out all related models.
With this information you can download all report specs also with the REST API. Only for uploads you might need JAVA SDK.

The use an AI for the spec analysis.

Another approach is to copy the table with the report specs from the content store repository into a relational db environment and connect an AI with this table to do the analysis. Then you only have the internal ids from the content store but it is possible to join this with other content store tables to find out the object id you know from the portal.

Thanks for all your responses.

Yes, my plan is to first compare the audit db with the actual objects in the content store to identify unused reports.

Once I have the list of used reports, I would do some kind of structure fingerprinting, like number of queries, query items used etc and use classification to cluster similar ones.
If you have done something like this before, what would be good candidates for the fingerprint items?
I don't think I could use a LLM or AI on all the full specs - could I?

thanks again

one idea is to download the specs using the REST API and then use an AI to identify similarity.

In detail: 
using the REST API you can manage to get all information about all objects which are in the team content folder (with java sdk I guess even what is in my folders)
Then you know all reports, reportViews with ID and search path. It is also possible to find out all related models.
With this information you can download all report specs also with the REST API. Only for uploads you might need JAVA SDK.

The use an AI for the spec analysis.

Another approach is to copy the table with the report specs from the content store repository into a relational db environment and connect an AI with this table to do the analysis. Then you only have the internal ids from the content store but it is possible to join this with other content store tables to find out the object id you know from the portal.

Thanks for all your responses.

Yes, my plan is to first compare the audit db with the actual objects in the content store to identify unused reports.

Once I have the list of used reports, I would do some kind of structure fingerprinting, like number of queries, query items used etc and use classification to cluster similar ones.
If you have done something like this before, what would be good candidates for the fingerprint items?
I don't think I could use a LLM or AI on all the full specs - could I?

thanks again

one idea is to download the specs using the REST API and then use an AI to identify similarity.

In detail: 
using the REST API you can manage to get all information about all objects which are in the team content folder (with java sdk I guess even what is in my folders)
Then you know all reports, reportViews with ID and search path. It is also possible to find out all related models.
With this information you can download all report specs also with the REST API. Only for uploads you might need JAVA SDK.

The use an AI for the spec analysis.

Another approach is to copy the table with the report specs from the content store repository into a relational db environment and connect an AI with this table to do the analysis. Then you only have the internal ids from the content store but it is possible to join this with other content store tables to find out the object id you know from the portal.