LGTM. I would leave out the mention of signals used to infer importance in the target. Right now we just use pageviews in the source.

Another issue that is independent of proper randomization, is that for most use cases, the data produced by the system cannot be used for statistical testing. Let me give an example;

@Nuria
I'm confused about how your statement "a bucket will have control and treatment for 1 experiment". I though that a bucket represents a group of users that get assigned to either the treatment or the control.

@Nuria, @BBlack
I need to clarify that in the example that I gave above, the experiments were not run concurrently, but in sequence.

@Slaporte Do you have any recommendations for the NOTICE file, that apache 2 suggests including in addition to the LICENSE?

@BBlack, @Nuria
In order to run a randomized controlled experiment, you need to ensure that users are randomly assigned to treatment conditions at the start of every experiment and that they remain in their treatment group for the entire duration of the experiment.

I'm fine with using Gerrit. However, Ori should probably complete the request for a new Gerrit repo since you need to choose a code review model and location inside of MediaWiki to do that and I'm not sure what is best.

@ori Who should I add as the copyright holder? Is it me or WMF?

@DarTar Nithum wrote an abstract and it is up on https://www.mediawiki.org/wiki/Wikimedia_Research/Showcase#Upcoming_showcases

Nithum and I are planning on staying below 40 minutes, so that should work out.

See: https://github.com/ewulczyn/wiki-detox/blob/master/src/analysis/Prevalence%20and%20Efficacy%20of%20Moderation.ipynb

The raw diffs are up. See:
https://datasets.wikimedia.org/public-datasets/enwiki/article_talk_diffs_tsv
https://datasets.wikimedia.org/public-datasets/enwiki/user_talk_diffs_tsv/

@santhosh Great! Thank you.

In the last quarter, we focused on building machine learning models to detect personal attacks on user talk pages. Now we will extend that work to the article talk namespace.

Hey @santhosh , I just wanted to double check with you that this is not already happening. Also, if not, what process do you suggest for adding this logging?

Katie is the owner of the private fr github repo.

Hey @kaldari, do you know who I should talk to in order to suggest changes to the admin blocking interface?

For now I am just removing unused files and refactoring the API. I agree that it would make sense to separate the two though.

Our current thinking is to make Detox into its own service that exposes a scoring API. This way ORES can just submit revision ids or diffs to the API and get back scores instead of running the models itself. How does that sound?

The are very few RPA redactions (~400). There are 36k instances of the npa template.

pinged James about how to access the data

@Pcoombe Yes, you will need access to hadoop.

@DarTar He has approval and should sign it soon