OK, a few corrections need to be made that justifies a new article. Thanks and h/t to Gabe for his feedback over at My Left Nutmeg.
The latest number is Joe Courtney by 109 votes, which reflects the +100 error for Simmons and the +37 error for Courtney.
The procedure going on right now is called a "canvass" rather than a recount. The canvass has to be completed by Wednesday night. From what I understand, the canvassing includes checking the totals reported by each lever-style mechanical machine against the reported totals. This is where the +100 vote Simmons error and +37 vote Courtney errors came from.
The larger numbers were most likely transcribing errors. The other 1 and 2 vote errors may be overlooked provisional ballots, or minor transcribing errors. Many of the towns have reported these kinds of small errors, and they tend to balance out between the candidates. There is no significant sweep going toward either candidate.
The OS machines are canvassed by re-scanning the ballots through the machines and checking the figures reported. In the event of an actual recount, which the loser may request after the canvass is completed should the difference remain this close, each paper ballot will be manually tabulated. I think a recount can be requested any time the margin is withing 1/2 of 1%, but I'll have to check that. It may be as high as 1%.
All of this is separate from the random audit of 25% of the OS machines that Secretary Bysiewicz's office will perform at some point to verify the accuracy of the machine totals themselves; this was planned before the election and has no connection at all to the canvassing or potential recount. I'll try to find out when the audit is scheduled to take place. Obviously, the canvass and possible recount take precedence.
I'm waiting to hear back from the Secretary of State's office on more details.