For the same reason Grammarly does it too, I'd assume.

To track when the user corrects it. Otherwise you can't adapt if somehow the correction is not what the user wanted.

If there are a bunch of these corrections you know something is wrong there. IMO 30 days is quite modest and if this is properly anonymized..

Edit: dear HN user who decided to silently downvote - you could do better by actually voicing your opinion

> dear HN user who decided to silently downvote - you could do better by actually voicing your opinion

Sure, I'll bite. Let's address the obvious issue first: what you're saying is speculation. I can only provide my own speculation in return, and then you might or might not find it agreeable, or at least claim either way. And there will be nothing I can do about it. I generally don't find this valuable or productive, and I did disagree with yours, hence my silent downvote.

But since you're explicitly asking for other people's speculation, here I go. Advanced "spellchecking" necessitates the usage of AI, as natural languages cannot ever be fully processed using just hard coded logic. This is not an opinion, you learn this when taking formal languages class at university. It arises from formal logic only being able to wrangle formal logic abiding things, which natural languages aren't (else they'd be called formal languages).

What the opinion is, and the speculation is, is that this is what the feature kicks off when it sends over input data to MS's servers for advanced "spellchecking", much like what I speculate Grammarly does too. Either that, or these services have some proprietary language engine that they'd rather keep on their own premises, because why put your moat out there if you don't strictly have to.

Technologically speaking, at this point it might be possible to do this locally, on-device now. This further didn't use to be the case I believe (although I do not have sources on this), and so this would be another reason why you'd send people's inputs to the shadow realm.

It’s hard to read writing packed with defensive clauses.

Better to say what you need to say. Leave the defense for the occasion someone misunderstood what you meant to say.

It's further pretty hard to write like this, but I still prefer it over getting trivially checkmated by ill meaning people, and over being misinterpreted silently and that causing issues downstream. It's at this point an instinctual defense mechanism, that I've grown to organically develop in the low-trust environments that are forums like this.

I 100% agree with the principle, but (regrettably) in practice you can't do this in a lot of places where the community is critical (which isn't a bad thing by itself) but doesn't call out/downvote/moderate bad criticism (which is bad).

I can't count the number of times on HN that I've seen responses to posts that took advantage of the poster not writing defensively to emotionally attack them in ways that absolutely break the HN guidelines, and weren't flagged or downvoted. And on other sites, like Reddit, it's just the norm.

The defensive writing will continue until morals improve.

This is often (though not always) blanket statement.

Logs are always generated, and logs include some amount of data about the user, if only environmental.

It's quite plausible that the spellchecker does not store your actual user data, but information about the request, or error logging includes more UGC than intended.

Note: I don't have any insider knowledge about their spellcheck API, but I've worked on similar systems which have similar language for little more than basic request logging.

Pii is stored _at most_ for 30 days.