When searching with a location bias, then results might get better when the limit parameter is higher.
This happens because the weights used to select the candidates is slightly different than the weights used in rescoring. The candidate search is more biased towards location, which usually makes for better results, unless there are so many matching items near the focus point that all other results are dorpped off the result list.
I considered increasing the limit for candidates. Thing is it makes the queries noticably longer running while only making the problem less prominent but not really solving it at the root, i.e. there may always be more candidates around the focus point than what the limit allows.
In the end, it might be best to have separate queries for candidates that are around the focus point and candidates that cross the importance threshold.
When searching with a location bias, then results might get better when the limit parameter is higher.
This happens because the weights used to select the candidates is slightly different than the weights used in rescoring. The candidate search is more biased towards location, which usually makes for better results, unless there are so many matching items near the focus point that all other results are dorpped off the result list.
I considered increasing the limit for candidates. Thing is it makes the queries noticably longer running while only making the problem less prominent but not really solving it at the root, i.e. there may always be more candidates around the focus point than what the limit allows.
In the end, it might be best to have separate queries for candidates that are around the focus point and candidates that cross the importance threshold.