It's the philosophy behind them. Let's say Google Search improved relevance by showing less low quality sites. However what if those sites also showed a lot of Google Display Ads and thus generated a lot of ad revenue for Google versus other sites. Given this email it's very possible that such a change would get rolled back or discouraged due to the drop on Google revenue from it.

If there had been such a change or a discussion about whether to make such a change, presumably that'd be what the DoJ would have used as evidence instead of this.

The DOJ can't magically pull out a time machine to record conversations that happened but were never written down.

The critical thing that appears to have happened is that a feature was rolled back because it hurt search queries. What feature it was is unclear from the email, since they never discuss it directly, but it was noted that users in the experiment were asking if it was a bug in public. Furthermore, the feature was newly added months prior.

So if I'm following it correctly, the email boils down to "please revert $POPULAR_CHANGE because it makes us less money on search ads, but don't tell people that's why we're doing it!"

(Someone who is a more diligent sleuth than I could probably figure out, but I don't know how to search Reddit for threads in ~April 2019 asking about bugs in Chrome related to features that launched ~October 2018.)

> The critical thing that appears to have happened is that a feature was rolled back because it hurt search queries.

Where are you getting there being a rollback? The earliest emails in the thread say there won't be a rollback, and the newest email asks for that decision to be reconsidered. Nowhere does it say that a rollback actually happened.

(The reddit thread isn't about a rollback, it's about an ablation experiment on a small % of users.)

As I read it, the ablation experiment is the experiment seeing what would happen if they rolled back, rolled out on a small % of users.

Sure, but that's not a rollback. That's just figuring out whether a rollout / which rollout caused a revenue drop. (It's basically as if they'd not done a 100% rollout in the first place, but kept a 1% holdout group. The point isn't to get more revenue from that 1%, it's to be able to measure long-term impact from the original change.)