Thank you for sharing your insights here. Yes, I’m aware that SMT needs a bit of time to “stabilise” so the readings are taken after driving quite a bit.Hmm - I think you just demonstrated how difficult it is to estimate your battery capacity with significant accuracy. It looks like your battery is an LG M48 like my car (and for @Bouba). You mention that your FPWN is 74.5 kWh (same as mine). That makes 1% battery, 0.74 kWh or thereabouts.
First off, purely as an example, depending on the accuracy of a specific type of data measurement (which are often subject to rounding), measurements that seem different could in reality be very similar. I refer to your 3% and 5% above - it would not take much of a measurement variation and then data rounding to make those practically the same.
Next, when did you take the SMT readings? Straight away or a few minutes after Bluetooth connection? Sometimes it takes a while for SMT to get a current value from the CAN bus and sometimes it will only do that if the car has been driven. It depends on the parameter, but, you could have errors there too.
I appreciate all that sounds a bit bla bla, but the example is rational.
I use NFP data over time in order to assess battery capacity, It appears good for +/- 0.5 to 1 kWh error on total capacity. I could comfort that data with the SMT CAC values (when they were working in SMT before last December!). The other ways of estimating capacity seem to give about a 1 kWh error as discussed in previous posts. On that basis I would be inclined to believe your NFP calculation of 6.7% (but that's my opinion only and happy to be called out). Logically that value would also depend on lifetime average SoC, average storage temperature and n° of charge cycles as per @AAKEE posts. In the end best to charge to 100% and drive to display 0% when it's moderately warm, avoiding phantom drain wherever possible, then you'll know for sure for the current battery capacity.
What I don't get is the result from the health test. The "HV Battery Health Test Confusion" talked around that without arriving at a conclusion. I wonder if it's actually measuring other parameters Tesla considers critical to a healthy pack - things like pack imbalance where a cell is significantly down on the others in the same module? There's a commercial post discussing this here
Once aspect which we all need to consider is that, on Tesla there is a buffer energy in the battery - 3,2kWh below zero for M3 LR with LG battery. So, 1% when NEE is more or less 0,713 and decreasing according with degradation.