I agree that there's a discrepancy here that we don't understand.
Let's attempt to estimate this from Tesla's Q3 reported figures side, assuming that all Model 3 cells and all PowerWall cells were sourced from Panasonic's 2170 cells at GF1:
- Q3 update letter: 477 MWh storage deployed in Q3, up from 415 MWh in Q2.
- production & deliveries report: 78,837 Model 3's produced. If all are Model SR+ then this is ~4,300 MWh, if all are Model 3 LR pack based then this is ~6,300 MWh. Assuming a 60%/40% mix this is ~5,100 MWh.
- Mystery GF3 battery packs leak from CleanTechnica, if there were 7,000 SR+ of the, would be another 385 MWh of cells.
Total Q3 cell consumption was thus, with realistic product mix assumptions and the GF3 speculation included, 5,962 MWh - annualized to 23.8 GWh.
That's too low IMHO, if we back-calculate end-of-Q3 Panasonic output of 30 GWh by 20% then the Q2 exit rate would have been 25 GWh annualized - and an average production ramp and output for Q3 of somewhere around 27-28 GWh and way too many cells and 3-4 GWh of cells nowhere to go.
So either Panasonic ramped from 25 GWh to 30 GWh at the end of Q3 in an almost binary fashion (hard to imagine this in a complex factory), or there's something going on we don't understand.
In any case, it's hard to make these numbers work at all without assuming around 5-7% of scrap rate - but I do agree with you that it's almost absurd to report production output that gets scrapped a day later when it fails in testing...