Among the things that failed in the design for healthcare.gov: the overall plan was to have each component built, to a specification, and then any two teams that were integrating components would test integration, and the whole thing would work.
Three pieces of the puzzle missing:
1) It appears there was no honest-to-God real end-to-end testing in the plan: no team to own it, no chain of responsibility if it didn't work. So cross-functional issues like total system performance were ownerless, and it is the nature of government contracting work to push blame away, not assert ownership where none is demanded by the contract.
2) Even in integration testing, responsibility was unclear when it failed. If the specification itself proved faulty, neither side of the protocol had ownership over fixing the problem; faulty specs led to pissing matches of blame-pushing instead of any party saying "Okay, the spec is now this, just fix it."
3) Since nobody had holistic ownership, no core infrastructure decisions were made. The whole project, under the hood, was a mismash of different databases, languages, messaging services, and protocols, owned by (and chosen by) each contractor with no thought paid to long-term maintenance of the app as a singular service. Each team had their own ad-hoc dashboards with no top-level view to allow anyone to understand why the end-to-end site experience was broken. There was nobody to own a concept equivalent to "Ops" or "Site reliability" who had authority to make judgement calls here; long-term maintenance was going to look like full understanding of dozens of similar systems doing similar tasks.