When do you breach a certain infection trend, this is a statistical problem and in my opinion you can solve it with the Poisson cumulative distribution. The Poisson distribution is for event probabilities, it is like PDF(x,lambda) where lambda is the average event repetition rate, and where x is a chosen rate. The CDF is the primitive of the PDF, the CDF is used for computing criteria intervals.
Our current rate of infections is 8000 per day, so the problem is, when could you claim a breach in that trend. 8000 on average means that every 10.8 seconds (on average) someone is infected during a day, this is the lambda parameter in the Poisson distribution.
Next you want the CDF of the Poison distribution to be above 90% or below 10% (this choice is arbitrary). For x=6 you will see that CDF_poisson(6,10.8) = 8.7% and for x=15 the CDF_poisson(15,10.8) = 91.77%.
For 1 infection every 6 seconds you find an upper bound of 14400 infections per day and for 1 infection every 15 seconds we get a lower bound of 5760 infections per day. The confidence interval [5760 .. 14400] is pretty wide open.
All discussions where people say, see, it is 500 more (or less) than yesterday are in my opinion therefore meaningless, you need more of a change to get out of the current regime.