If there were requirements published (big if) they were very poor or done at the last minute.
A. No one understood the difficulty of designing with requirements to that many (each customized hundreds of times) legacy programs... therefore they half-assed it. Has it EVER been done before? Without upgrading the legacies first??
B. I'm going to bet that the requirements changed WEEKLY as someone else insisted on functionality from front-end to back that no one thought of in the original design... as a way to Cover their exposed backsides...
Step one would've been to normalize all the underlying systems (a 5 yr project, most likely at gov's usual pace). So that they were consistent in basic features and functionality...
THEN, one could build the front end. And one would ALSO begin to deal with the problem of exporting relevant data to the insurance companies who are using their own proprietary-designed systems... sigh. ***
I'm guessing here: work has JUST begun on the IRS systems to account for the verification of income, amount of subsidy, and consequent penalty - if any? How about the employer reporting systems? [Lets just go on and crash the IRS system in the process of building that - the US gov doesn't care how much money it receives in revenues daily... sigh...]
Honestly, a project of this magnitude - even if it were well-designed from the get-go, and problems simulated and solved as part of the design process, could be expected to raise continual glitches and outages. What happens when one of the legacy systems is upgraded in the future???
Even when upgrading EVERYTHING involved, all at once -- things popup and require awkward work-arounds and "make-dos". The pipedream of integrating data from multiple sources into some dashboard type AI brain... is still a pipedream, Watson notwithstanding.
*** PS: and then there are the encryption and security issues and ID validation protocols at every step along the way...