4 ms·

The original Healthcare.gov was a clusterfuck because contractors did a shit job. They did a relaunch, but I'm wondering how much of it was actually rewritten.

The original Healthcare.gov was a clusterfuck because contractors did a shit job. They did a relaunch, but I'm wondering how much of it was actually rewritten. I'm putting my money on brokers/agents having weak passwords and someone did some guessing like firstname.lastname@something.gov/<password>

heelix 8y ago
I was one of the poor souls sent in as part of the tech surge to fix it. Ah, what a cluster fuck that was. CGI spent most of the time making UML diagrams, with the hopes that a UML > Java > XML generator would do the job... only to discover they missed out on the data modeling. Always fun to see CNN show a twitter feed the moment you take down something for patching.
- simplify 8y ago
  Oh wow. I would love to hear more stories if you have any.
  - heelix 8y ago
    Day 0, when it went live, ACA successfully processed six people. :P Will give a few more tomorrow, when I have something more than a phone.
- josh-wrale 8y ago
  Can you expound on the data modeling they missed out on? Is the UML code generator a bad way to go in your opinion or did they just not do due diligence with data modeling? (Asking because I'm starting a new data-heavy project and I'm considering generating code from UML.)
  - heelix 8y ago
    At the core, they sort of missed the problem. With the advent of ACA, you could no longer use medical history to set rates. You had age, sex, smoker -- and the killer location. What they were actually after is a bit of a rules engine, as one person may have 30k+ offerings, another location none. Think actuarial tables at the zip code level. The data model itself was a bit ... modeled around people and missed the rate part. I think they had about 2 years - and spent much of that time modeling. The cracks were not evident until way too late. One of my favorites was the way they serialized the POJOs. Data object turned to XML. Send to a process that added more stuff. Send to another process, lose all the non-base stuff. Lots of data corruption. The model being wrong required them to try and tack on all sorts of extra stuff... but the framework really did not support it. They tried to match a handful of A players with a bunch of C grade developers. Then they pulled all the A players into never ending meetings. I saw little to no code review of what was actually going on. Folks literally copied switch blocks, because the code worked, and left in the old case statements. Exceptions eaten. Text book example after example of what you might expect in the daily wtf type code.
    - pbreit 8y ago
      Does legislation take into account at all how difficult or complicated it is to implement?
      - cobalt 8y ago
        hah
briandear 8y ago
Contractors did a shit job because the people in charge had no idea what they were doing.
stephenhuey 8y ago
I was surprised to hear there’s been a breach because Nava works on it now and I’ve talked to them and they seem to know what they’re doing. https://www.navapbc.com/ https://www.navapbc.com/

heelix 8y ago

I was one of the poor souls sent in as part of the tech surge to fix it. Ah, what a cluster fuck that was. CGI spent most of the time making UML diagrams, with the hopes that a UML > Java > XML generator would do the job... only to discover they missed out on the data modeling. Always fun to see CNN show a twitter feed the moment you take down something for patching.

simplify 8y ago
Oh wow. I would love to hear more stories if you have any.
- heelix 8y ago
  Day 0, when it went live, ACA successfully processed six people. :P Will give a few more tomorrow, when I have something more than a phone.
josh-wrale 8y ago
Can you expound on the data modeling they missed out on? Is the UML code generator a bad way to go in your opinion or did they just not do due diligence with data modeling? (Asking because I'm starting a new data-heavy project and I'm considering generating code from UML.)
- heelix 8y ago
  At the core, they sort of missed the problem. With the advent of ACA, you could no longer use medical history to set rates. You had age, sex, smoker -- and the killer location. What they were actually after is a bit of a rules engine, as one person may have 30k+ offerings, another location none. Think actuarial tables at the zip code level. The data model itself was a bit ... modeled around people and missed the rate part. I think they had about 2 years - and spent much of that time modeling. The cracks were not evident until way too late. One of my favorites was the way they serialized the POJOs. Data object turned to XML. Send to a process that added more stuff. Send to another process, lose all the non-base stuff. Lots of data corruption. The model being wrong required them to try and tack on all sorts of extra stuff... but the framework really did not support it. They tried to match a handful of A players with a bunch of C grade developers. Then they pulled all the A players into never ending meetings. I saw little to no code review of what was actually going on. Folks literally copied switch blocks, because the code worked, and left in the old case statements. Exceptions eaten. Text book example after example of what you might expect in the daily wtf type code.
  - pbreit 8y ago
    Does legislation take into account at all how difficult or complicated it is to implement?
    - cobalt 8y ago
      hah

briandear 8y ago

Contractors did a shit job because the people in charge had no idea what they were doing.

stephenhuey 8y ago

I was surprised to hear there’s been a breach because Nava works on it now and I’ve talked to them and they seem to know what they’re doing. https://www.navapbc.com/ https://www.navapbc.com/