A quick note about the trending indicator: in T409601, DPE onboarded a data pipeline for WME that classifies articles as trending according to (I think) this methodology. You can see a sample query here. Naturally, your definition of trending doesn't have to agree with that, but this could be a starting point.

In general, I think this is good to consider, though I'm not seeing a very strong priority signal just yet. Can we estimate how incorrect our current completeness estimate is, e.g. with a one-time comparison to wmf.mediawiki_history?

This is great, thanks for the detailed proposal @JMonton-WMF !

@Hghani points out (on a first look) that this can be related to T413027. That ticket also hints at some signals we can incorporate into our classification from here.

Having spoken to @Ladsgroup this morning (thanks!), here's my notes on this task.

Update (from Slack): cuc_agent and cu_changes are no longer bring read in MediaWiki, and writing will tentatively stop on the week of Feb 9.

Sorry I'm late to this, but I basically second Andrew's comment. I think there are two things at play here:

@fkaelin and I just chatted a little more about this, quoting here:

Thanks for checking! What annotation do you have in mind? Something like, "Here's when we enabled temp accounts, and they are included under Anonymous edits"?

Looking at the attached sheet, I see the following must-have actions listed as "complicated":

From a quick discussion at the DE team meeting:

This looks like a very reasonable change, and Q3 sounds feasible; I'm not entirely sure how complex this is, but it looks more like "weeks" of effort than "quarters" or "days".

Sorry I'm a bit late to this - I just caught up a bit on the incident this morning.

I've used the link to set up a donation. I get to the "thank you" page normally on my browser.

@calbon mentioned today that we want to make sure to capture the referrals from x.com/twitter.com . We do look for Twitter in the Referer header, but not X.

Unfortunately it looks like we won't be able to prioritize this for the next couple of months in Data Engineering, but it's still something we want to get to. We'll revisit this in our next planning period towards the end of the year.

@JAllemandou and @OSefu-WMF just had a chat about this one and how we can address it, now and in the future.

Hi, just a quick update after my chat with Sukhbir. We should do this, not only for the value of the dataset itself, but also because it will be an excellent opportunity to make this kind of release a more documented and repeatable process.

I took a look at the upstream issue tracker and found this issue, which looks related. But that's from 2021 and the associated patch, which allows users to view draft dashboards, apparently went into the the 1.2.0 release according to the GitHub tags on the right. We seem to be running Superset v4.2.0 at the moment.

Thanks Valentín, I'm thinking probenet can be a useful signal, but our current focus is to experiment with specific known signals (e.g. presence of DOM properties). We'd like a bit more flexibility to deploy custom logic, and also control over what % of traffic we collect those signals from. More fundamentally, we believe we'll need the X-Request-Id correlation to even be able to evaluate with confidence whether client-side signals, include probenet, are helpful or not.

That is a good question, I was just quoting @BTullis 's wisdom in the original ticket (T383175#10440220).

Nice, this makes a lot of sense! I think Jupyter users are generally used to going to either #data-platform-sre, #talk-to-data-engineering or (less frequently) #working-with-data for support.

Excellent, thanks for coming up with this! It's very cool to see the performance of the real pipeline. My thoughts the initial question:

Thanks KC, and agreed: if we need to make this into a dashboard for ongoing tracking, we should look into importing those logs into the Data Lake.

Thanks Halley, a couple of comments and follow-up questions please:

You're welcome, happy to help!

We had to ask around a little, and unfortunately we don't have a precise answer as we don't really instrument device orientation directly. However, we do instrument device viewport width (aggregated to buckets) for mobile web page views and keep it for 90 days, and I think that can help give you an approximate answer.

I met with @mfossati to discuss this. The main takeaways are:

Excellent discussion, thank you both.

Hey Debra, thank you and welcome as well :) Sorry for the delay here, I wanted to get Virginia's input again to make sure we go in a sensible direction with this.

Yep, I was able to run kinit, set my password and run it again. Thank you!

I was testing these credentials in the past couple of days. I can use Superset (so I'm in analytics-privatedata-users ) and log in to bastion and stat machines (so SSH is also fine).