I found (page 33) the method of finding fidelity from fit by "interleaved and reference decay" according to the sequence fidelity formula: $$F_{ref}=Ap_{ref}^{m}+B,$$ where $p_{ref}^{m}$ is sequence decay.
But if we look at the following datasets:
we can find that the reference sequence fidelity curve is close to $1$ and the other curves corresponding to the gates - around $0.998-0.999$ (which corresponds to decay $p_{gate}$).
How could it be that the difference between interleaved and reference decays is around $0.999$?