I suspect it's situation specific: there are probably situations where a model like this would work very well, and others where it would be a terrible idea. I'm also a little wary of government created 'pseudo-markets', as these often become bureaucratic nightmares that cost more to administer than if the government had just done the thing itself. 'Pseudo-markets' are kinda like machine-learning algorithms: they'll optimise towards your success metric, but sometimes in unexpected and undesirable ways (i.e. game the system).

Also, for certain services, you'd need to ensure equity of access (i.e. cover all use cases). Tax returns are a good example. There's some pretty esoteric stuff in tax returns that are only relevant to a handful of people (e.g. reporting franked dividends distributed from closely-held unit trusts via an interposed corporate entity, or whatever). It might never be profitable for a private market to cover this case, meaning the government would have to further 'pseudo-regulate' its 'pseudo-market', offer further subsidies, or cover this use-case itself.

Interestingly, on your second 'failure mode', you could flip it around and view it as a desirable outcome. Sticking with income tax reporting, most reporters have pretty simple tax affairs. They just report their annual income (which the government already knows), maybe claim a deduction or two, and that's it. But because the government must cover every possible use case, people are forced to wade through a 40 page form instead of a 1 page form. There might be more gain to society from doing this:

1) Government offers the fixed subsidy, but makes clear it is only guaranteed for, say, 2 years. 2) After 2 years, offers a lower (or no) fixed subsidy for the 'cherry-pickers' (who will still be profit positive) 3) Reallocate the savings as higher fixed subsidies for the remainder of the market

By iterating this process a few times, the market would naturally segment according to complexity, allowing the government to accurately 'price discriminate' on the basis of complexity. I dunno, I'm just spit-balling here, no clue if this is actually a good or bad idea. What are your thoughts?

Of course, all of this is only possible if the government publishes a 'tax return' API that's easy to use (i.e. does not create high implementation costs for private providers). Even though publishing an API sounds (and, frankly, is) simple, you'd be astounded at how often the government screws this kind of this up (often by contracting out to IBM, Fujitsu and their ilk). Or not publish one at all, even though the potential benefits are blindingly obvious...