It is possible for a reported lower-bound to be proven incorrect, by a subsequently reported plan which has smaller cost. We need a policy for how to handle these situations.
Suggest the following:
- Remove all lower-bounds from the same batch as the erroneous instance (we no longer trust any of them)
- Revert the claim to the previous best-known value. Note that this requires we keep a history of lower-bound claims for each instance.