The NYT published live data from Edison Research on their website. Each entry withing this data is county or precinct based reporting of vote totals from the counted batches as they were completed. There are very significant anomalies in these reported votes which can indicate nothing other than wide-scale vote fraud.
The below step plot is the total votes in Michigan as they came in. The most unusual values are indicated with blue Triangles underneath them. All 5 of these happened after 3 am, and all 5 represent large pro-Biden votes with minimal to no, pro-Trump component.

The steps were picked out as having a 0.5% or less chance of being part of the rest of the vote reports collected in Michigan. There were 5 entries and all 5 went to Biden and all 5 happened after 3am.
The below graph shows how extreme these points really were.
In recent work, I’ve used straight ratio’s of Trump votes vs total votes to determine significance. This gives symmetrical values between 0 and 1 with 0.5 being the situation where Trump has half of the votes. In this article, I’ve used the natural log of Biden/Trump ratio. This requires that all negative votes be eliminated but it mathematically clean as well. The 50/50 vote occurs at log(B/T) = 1.

The green bars show the distribution of the large vote entries prior to 3am Michigan local time. These large entries are used to avoid rounding issues in the NYT dataset caused by their reporting methods. The time was chosen arbitrarily except that this was when many of the absentee ballots were counted, and these ballots were subject to different rules than regular votes. No validation of voter information was made whatsoever. Note the general right direction shift in the red bars which represent ballot count after 3 am, because this represents some of the expected shift toward Democrats in absentee voting. Most of these votes still stay within the green bars and 6 of these large reports even went to trump.
The print overplotted itself on the graph, this is what it says in order from left to right.
Percent likelihood 0.25208 Row 4545 Time 2020-11-04 20:55:38 Vote Difference B-T 27100
Percent likelihood 0.20651 Row 4550 Time 2020-11-05 00:29:28 Vote Difference B-T 16800
Percent likelihood 0.19212 Row 4500 Time 2020-11-04 12:14:51 Vote Difference B-T 23900
Percent likelihood 0.00173 Row 4449 Time 2020-11-04 08:50:10 Vote Difference B-T 49800
Percent likelihood 1e-05 Row 4495 Time 2020-11-04 11:31:53 Vote Difference B-T 135300
The far right bar representing 135000 votes entered has a likelihood of being part of the standard vote of .00001 percent. One in ten million. This is the single largest vote entry reported in the state of Michigan and it is the single most extreme.
The story is even stronger than that though. It is critical to note that this analysis would have shown extreme Trump votes as well, yet the five most extreme one-sided votes in Michigan were all in favor of Biden, all during the time when absentee ballots were being counted.
The ballots in these five extreme entries were more than sufficient to change the outcome in Michigan. They are statistically impossible without a bias toward one candidate. This is evidence bordering on actual proof of significant fraud in the presidential election.
To reach the level of a statistical proof, validation of this information requires a deep look at actual reporting events from each county. The data Edison Research collected would be sufficient to verify this data and from the general appearance, would likely pinpoint the areas of interest. While the election cycle is likely lost, vote integrity is the foundation of the United States of America. Investigation of the counties and people involved needs to be pursued to the maximum extent of law.