OpenStreetMap logo OpenStreetMap

The OpenStreetMap (OSM) project has over 10 million registered members, with around 2 million user profiles having made at least one map contribution. However, a closer look reveals that there has been a slight decline in the number of active contributors over the last three years. Despite the extensive global mapping community, there are instances where individuals or automated bots disregard the consensus norms of the community when editing data. These situations arise due to disagreements regarding the appropriateness of certain tagging or features within the OSM database. To address these issues, a change rollback process, commonly referred to as reverting, is used to combat vandalism and correct ‘mistakes’ by restoring a previous version of the data.

Two years ago, I added additional statistics to the “How did you contribute to OSM?” page for quality assurance purposes. The numbers for each contributor profile were derived from an analysis of the full history OSM planet dump and changeset tags, including the specific editor used. While this pragmatic approach provides valuable insights, it’s important to acknowledge that the obtained numbers are estimations rather than exact figures. Furthermore, I received several inquiries regarding the implementation of the processing involved in identifying the displayed “reverted changes”.

Over the past few weeks, I have developed an advanced processing pipeline. This involved revisiting the comprehensive OSM planet dump and examining the evolution of each entity (node, way, relation) in relation to its previous states. Specifically, an entity with a higher version number was identified as a revert if it had the same latitude/longitude coordinates (for nodes), tags (key-value pairs), and/or members (for relations) as a previous version. In simpler terms, if a mapper changed “X” to “Y” and another mapper subsequently altered it back to “X”, it would be counted as a revert.

The following graph illustrates the amount of reverted map edits, changesets, and the contributors affected per month. This visualisation offers some initial insights into the scope and impact of reverted changes. It’s important to note, that the these numbers don’t include any actions related to reverted data imports.

Alt text

It is also necessary to look more closely at the specific entities that are counted as “reverted”. Are they primarily nodes with a few tags, or are they ways and relations with extensive mapping histories in active areas? These specific aspects, among others, will be explored in an upcoming blog post or possibly published as part of a scientific research study.

What are your thoughts? I think many of you might be curious to discover whether your own map entities have been reverted. Please feel free to leave a comment.

Location: Ciutat de les Arts i de les Ciències, Quatre Carreres, Valencia, Comarca de València, Valencia, Valencian Community, Spain


Comment from TheSwavu on 2 June 2023 at 11:47

Does having something you mapped get deleted count as “reverted”? Or is it just things that have their state returned to their former state?

Comment from Xvtn on 2 June 2023 at 14:57

I’d be curious to see specifically which of my changes were reverted. Is there a way to check that?

Also, just want to say thanks for creating your analysis tools, especially HDYC. It’s really fun to see my stats!

Comment from RicoElectrico on 2 June 2023 at 15:44

However, a closer look reveals that there has been a slight decline in the number of active contributors over the last three years

It’s not as dire if you disregard -

Comment from terminal_burgundy on 2 June 2023 at 19:52

Agree with Xvtn - it says I’ve reverted 7 changesets, but I don’t think I’ve explicitly reverted a changeset more than once

Comment from CaptainCarto on 3 June 2023 at 07:20

Nice job! I think it would be interesting to add the total number of active users on the chart, to compare it to reverted contributors.

Comment from MrMoehritz on 6 June 2023 at 13:39

Hey Pascal,

interesting analysis, could you add the relative share of reverted changesets?

This gives me the idea of creating a “reverted” map just like this “deleted map” I made ( that only features reverted objects. What do you think? Should be fairly simple to adapt the OSHDB code to look for an object state sequence of X->Y->X and then return Y.


Comment from pitscheplatsch on 30 July 2023 at 10:36

Took some time, but HDYC does now contain some links to see your reverted map entities.

@TheSwavu: Yes, normally and so far only entities which “have their state returned to their former state” are counted as reverted.

@Xvtn: Yes, check your HDYC page and the link behind your “Reverted changes” amount.

@terminal_burgundy: Your can now find your “Performed Rollbacks” here:

@MrMoehritz: So far my implemented pipeline counts almost 2.1 million reverted changesets. But I guess a few or more special cases are not yet included.

Comment from SomeoneElse on 31 July 2023 at 21:53

@pitscheplatsch Excellent, thanks! Although of course some “reverts” (for example in my case ) is just tracking data as it changes.

Comment from pitscheplatsch on 2 August 2023 at 08:04

@SomeoneElse You’re welcome! And thanks for your example. IMHO my current pipeline does not properly count the change in this case, so I have to adjust this.

Comment from VictorIE on 25 August 2023 at 00:57

When I see “Rollbacks performed: Reverted 1,754 changesets w/ 3,098 changes of 471 contributors”, it doesn’t resonate with work that I have done. :)

A few times I have noticed vandalism or inadvertent errors, e.g. someone dragged a way onto a relation and things got messy. I usually mention it on the local mailing list and ask for it to be reverted. I don’t know the actual reversion process. So surely those reversions count against someone else?

Once, I got into an inadvertent edit war with someone, where we had each been reversing each others edits without noticing. More than 10 edits on a single node. It related to Tesco having several brand=* tags. When I realise, we could both laugh at it.

Only recently did I see a case of a reversion I did in the wild. Some people have been adding noexit=yes to locations where there are actually exits. If I had mapped the original road, someone else added the noexit=yes and I deleted the noexit=yes, this counted as a reversion, even though I considered it a normal piece of mapping. I was deleting bugs, not seeking to revert someone’s work.

Comment from pitscheplatsch on 31 August 2023 at 19:42

Thank you for your comment, VictorIE.

According to your mapping profile, you are a very active contributor with almost 2.4 million edits and over 110,000 changesets created.

Even if you have never used a “real” revert tool, you can still revert edits using the iD editor. The situation you mentioned is just such an example. Because I don’t think it has been mentioned directly here: For me, a revert is not fundamentally a bad thing. In certain situations it might be unavoidable?

Comment from VictorIE on 31 August 2023 at 21:13

Yeah, one time I was adding the buildings, etc. to a power station and I added everything, but accidentally deleted the power station itself. :)

A revert isn’t wrong, but it may suggest something incorrect was done before it.

Comment from Kogacarlo on 3 October 2023 at 15:49

Hi Pitscheplatsch, on a certain page I read: “Performed Rollbacks of Kogacarlo”. That sentence makes it not clear to me if these are things I rolled back, or if it are things mapped by me and rolled back by others. You may want to change it to “Elements rolled back by Kogacarlo” or something to make that more clear. Have a nice day :-) Carlo

Comment from pitscheplatsch on 5 October 2023 at 18:43

Hi Carlo, yes, thanks.

Comment from Hungerburg on 10 November 2023 at 21:04

In July I did revert a bunch of really sloppy edits by a power mapper. He returned under a different nick and reverted my revert, saying I have to be more careful, there is good stuff also. In his revert, he again removed houses that obviously were not on his outdated aerial, so much for lacking care.

Meanwhile a DWG member reverted this and lots of other changesets of this picky power mapper. I guess, for the hdyc algorithm these are two separate reverts? I learned to live with 1.3% of all the changes in my whole career being reverted ;) I just hope this number will not get used against me.

Does that happen more often, enough to skew statistics?

Log in to leave a comment