Gracefully handle source regressions

[pnoll1]

Machine accepts bad data from source making it the last successful run and what gets packaged in the data downloads.

Example:Franklin County WA  28,619 addresses to 37

Marking the run as failed and returning the previously cached result would be much better for data consumers.

[iandees]

Machine doesn't know that it's bad data, but it could check to see if the row count changed significantly and flag it as an error. I think the system that @ingalls is working on to replace machine should provide for this.

[pnoll1]

Is there a roadmap or public code to track progress on this?

[ingalls]

Yes, you can see the new processing system here: https://github.com/openaddresses/batch