Go to:
https://static01.nyt.com/elections-...0-11-03/race-page/pennsylvania/president.json
That is the official NYT feed from the "Edison Research" data. It is at least official enough that it's what the media have used to call races.
You should be able to click somewhere to view as JSON as opposed to raw data or just headers. (If anyone has Visual Studio, Notepad++, etc, you can actually download the data as its own .json file and analyze there.
Once you're viewing it as json, you're seeing it displayed as human-readable structured data.
Click on the drop-down for "data", then "races", then "0", then "timeseries".
Each timeseries is a numbered data update from the pollsters. (This is the PA presidential race.) Each contains (among other things) an entry for total votes, each candidate will have a % of votes, and it will have a timestamp.
Click through to the update #s 187 and 188:
Find the raw numbers and do the math yourself. Multiply the total # of votes for that update by each candidate's % of votes. That will give you each candidate's vote total for each of these updates.
Report back here what you find, please.