I think it's pretty clear that when a fair portion of your data (polls) are misleading or incorrect for whatever reason - your algorithm will have a very hard time producing meaningful output.
The way I view it the Clinton camp knew they were in serious trouble and reacted accordingly.
Do you have a source for the internal poll collapse 1.5 weeks before the election?
Edit: My hypothesis is the likely voter estimates were pure fiction.