That's a good question! This paper also mentions "Dele. BF [43]" - the "The Deletable Bloom Filter: A New Member of the Bloom Family" https://www.researchgate.n

That's a good question! This paper also mentions "Dele. BF [43]" - the "The Deletable Bloom Filter: A New Member of the Bloom Family" https://www.researchgate.net/publication/260668624_The_Deletable_Bloom_Filter_A_New_Member_of_the_Bloom_Family https://www.researchgate.net/publication/260668624_The_Delet... . As far as I see, lookup is identical to the regular Bloom filter. Yet the vacuum filter paper lists the throughput of the Deletable Bloom Filter to be almost twice as the Bloom filter (1.95x). How can that be? Maybe I misunderstand something here. Yes, a peer review is supposed to catch completely incorrect numbers. But in many cases, a reviewer doesn't have access to the source code, and no way to re-run the tests. There's the "Replicated Computational Results initiative": https://jea.acm.org/rcr_initiative.cfm https://jea.acm.org/rcr_initiative.cfm . Papers that pass this test will get a stamp certifying that its results are reproducible. The xor filter paper (where I'm a co-author) took part in this initiative.