1. It has O(1) lookups (it has to probe at most 2 constant-size buckets) and O(1) amortised insertions (the probability of needing more than k swaps to find an empty slot decreases exponentially with k).
2. It's geared for low false positives, 3% or lower.
3. It does have false positive on lookups: signatures are lossy.
4. It has comparable space requirements to bloom filters for the same error rate and set (not universe) cardinality.
2. It's geared for low false positives, 3% or lower.
3. It does have false positive on lookups: signatures are lossy.
4. It has comparable space requirements to bloom filters for the same error rate and set (not universe) cardinality.