1. Correctness and popularity are different

2. Not everyone looks for correctness, even when looking for a database. See: MySQL.

> as its not a piece of your architecture that you can tolerate having constant outages with

Reliability and correctness are not binary scales of 'works' vs. 'does not work'. Again, see MySQL.

> there is no evidence of MongoDB being systemically unreliable

There is plenty of evidence of MongoDB losing writes because of its (honestly) stupid design: writing to memory maps instead of files, never fsyncing, sharding and distributing without confirming disk flush.

---

MySQL: For a long time, one needed to be aware of certain caveats and corners and limitations of MySQL or be surprised with data inconsistency and loss. That is not to say that MySQL could not be operated reliably; it could, just that the barrier was higher than it appeared.

From truncating data (with a warning) to not accepting and silently (i.e., without warnings) ignoring CHECK constraints, I've had clients who lost business data and realized it only too late to go back to the original data source.

I used to think MySQL was an bad database due to its concessions of correctness and poor documentation, until MongoDB came along with their lofty marketing and high-and-mighty "Nobody needs SQL anymore, we are Web-scale" blare-horn and failed to deliver on even keeping the data it claimed to have saved.