Scale exactly as torflow does?
legacy/trac#33775 (moved) shows that if sbws calculates low consensus bandwidth because of missing descriptors and the rest of bwauths are also using sbws, it'd enter in a spiral in which it'd keep measuring low. I think this is because while torflow multiplies the calculated ratio by the descriptor observed bandwidth [0], while sbws multiplies the ratio by the minimum of all descriptor bandwidth values and the consensus, which was added in legacy/trac#28598 (moved). So maybe the new consensus bandwidth should not depend on the previous one, or not as the minimum. For a relation on how bandwidth values depend on each other, see [1]
[0] https://onbasca.readthedocs.io/en/latest/torflow_aggr.html [1] https://onbasca.readthedocs.io/en/latest/bandwidth_tor.html#bandwidth-values-origin