Perhaps someone with better insight can educate me on this matter?
Alternatively, it could be a category of transactions that eventually almost always see high fraud (e.g. kids paying for essays using parent's money, parents reporting the transactions afterwards).
There are certain categories of transaction that see a lot of disputes.
Payment processors, banks, and card schemes (like Visa/MasterCard) don't like disputes - they are costly to deal with.
I can understand if they're saying "we think you're about to get a lot of disputes, and you may not be able to pay for all of them, so we're going to hold some of your money for a few days just to see".
But that isn't what they're saying.
Essay mills fall under the category of "deceptive" products and services. It's clear we made a mistake here in looking at this particular business and we've re-enabled the account (https://news.ycombinator.com/item?id=34802504). In particular, we didn’t correctly distinguish between augmenting the ability to write content vs. buying essays to commit academic fraud. We'll be revisiting our internal policies to better dilineate between these categories.
We also work with our partners to relax or remove unnecessary restrictions where we can. Being overly restrictive towards legitimate businesses flies in the face of growing the GDP of the internet—it's a precondition for us to operate. A small component of that is being reactive: seeing where we make mistakes and noticing where our processes need to be adapted. (As we saw today with essay mills!)
And if you know the constraints of classifiers where recall and precision are a tradeoff, and consider that basically fraud cannot happen for economic reasons, you can imagine that yanking up the recall (loss avoidance) at the expense of precision (stomping over innocent false positives) is tempting and lucrative, at least on short term.