It seems possible to purchase reserved instances based on average usage instead of initial usage. For example, if I purchased 2 RIs, ran 1 instance on day 1 and ran 3 instances on day 2, it will be essentially the same usage rate as if I ran 2 instances for 2 days. Will RI calculation see it that way, or will I be charged 3 days of RI usage and 1 day of on-demand usage?
There's no spread across hours.
Each hour is considered separately, and underutilized reserved instance hours don't roll forward or backward into hours where you have more running instances than reserved.
Customers using both Reserved and On-Demand instances will have Reserved Instance rates applied first to minimize costs. You pay the low hourly usage fee for every hour in your Reserved Instance term (which means you’re charged the hourly fee regardless of whether any usage has occurred during an hour). When your total quantity of running instances during a given hour exceeds the number of applicable Reserved Instances you own, you will be charged the On-Demand rate. (emphasis added)
https://aws.amazon.com/ec2/pricing/reserved-instances/buyer/
While not explicitly stated, of course, "you will charged the On-Demand rate" refers to running instances each hour beyond the number of reserved instances.