TL;DR: Contrasting some counties in CO and GA where we actually don’t have the “data leaks” — that I have been examining in recent posts — to strongly identify the mail in vote algos in operation, we find evidence that the algos in CO are “synthesizing” the “late R voter” trend seen there in large counties and that in fact, if anything R voters may tend to surge early in the voting at least when the bulk of voting is in person. So I now strongly suspect that at least in CO the “late R voting” trend story (that even I bought early on) is instead a gaslighting operation. (Update: We now have a failed coverup in Denver County 2020 — go see!)
If we look at the trends of R vs D voting through time in (all mail in voting) Colorado we find variations of an “R votes wait to ‘election day’ to vote” story. (In mail-in voting that just means holding mailing/ drop off until election day or near to it.)
As an example, the straight blue line below is the linear trend of Trump votes per 1000 votes in Larimer County in the CVR file CVRnumber order:
Strongly up and to the right with a 0.81 R2 value. Trump “lost” in Larimer but did win in Douglas which is the deepest red larger county in CO and everyone knows it. For instance, Douglas county actually seceded from the Tri-county health department it was in with Adams and Arapahoe counties so it could ditch the mask mandate (I charted how this natural A/B experiment in “health measures” showed masks/measures have no or possibly even detrimental effects here).
The CVR chart for Douglas looks virtually the same as Larimer’s but running at generally higher votes/1000 levels ending in a moderate Trump win (trend linear with a similar R2 to Larimer at 0.79 vs Larimer’s 0.81!):
R2 values this high in these kind of data sets start to be meaningful and are consistent with an algo manipulating things.
Now let’s go to some deep red counties in Georgia that have very little mail in voting and look at their trends holding aside their mail in votes (which happen early in their counting). Here’s Hall County, GA:
Obviously the R2-value is a lot lower but as voting progresses from early in person voting (about 17-70K) to election day voting the linear trend line is actually down!
And Cherokee, GA is similar:
Again, a weak R2-value but if anything a trend DOWN in Trump percentage. If anything, these weak R2-values suggest the data is more “natural” while making the evidence of a consistent down trend less compelling.
With Trump wins this strong in the early/election day phases of voting we seem to have decent evidence that either the machines have no algos for early/election day or they are either not running or pretty idle. (And remember that none of these counties have the “data leak” signatures in their records so we can’t make strong guesses about when algo manipulations happen.)
And you can clearly see that the mail in voting phase for each of these counties (dotted brown line at a value of 50) is a quarter or less of total votes (that interestingly seem to all/mostly be counted at the start of the CVR sequence) and so had a relatively small influence in reducing the Trump overall win percentage at the end — but reduce it they did.
Finally, you can see the same general algo pattern in the mail phases where things can occasionally or entirely snap close to 50/50 (with some oscillation) but in the case of these strong Trump counties running at 75+ percent of the vote for some stretches (of in person voting at least) it looks like the algos were only able to shave some votes and almost never pull Biden into anything much of an even instantaneous lead.
In conclusion, the mail in phases of these GA counties sure do look manipulated (even though we don’t have the CVR file signature/ “data leak” for that) since they are wildly different from the in person phases which are the vast bulk of the votes. But the huge Trump lean of these counties didn’t allow them the time or algo operating margin to fully paint a late R voter line in the mail in section — although if I had to guess it looks like Cherokee kind of did it gratuitously at the end of mail-in counting as the algo was overwhelmed with raw Trump votes!