Bloom Filter

A Bloom filter is a probabilistic data structure that lets you check "is this item in a set?" cheaply, with no false negatives and a tunable rate of false positives. In Bitcoin, Bloom filters were introduced by BIP-37 (2012) as a way for SPV clients to request only transactions that might match their addresses.

The original workflow:

1. SPV client builds a Bloom filter encoding its addresses (with some intentional fuzziness via false positives).
2. Client sends the filter to a full node.
3. Full node sends back only blocks/transactions that match the filter (plus some unrelated false-positive matches).
4. Client checks each result locally to see which are real matches.

The fatal privacy flaw: by inspecting the filter, a malicious or curious full node can probabilistically reverse-engineer which addresses the SPV client cares about. The false-positive rate is supposed to provide cover, but in practice it's too easy to extract patterns when a client repeatedly queries with overlapping filters.

By the late 2010s this was widely understood, and many Bitcoin Core nodes disabled BIP-37 Bloom filter relay entirely (it became opt-in, then disabled-by-default in Core v0.19).

The successor is BIP-158 compact block filters, where the full node generates the filter and the client downloads and queries it locally - so the node learns nothing about which addresses the client cares about. Same primitive idea, opposite information flow, much better privacy. Modern SPV-style mobile wallets use BIP-157/158, not BIP-37.

Bloom filters as a general data structure are still useful in many other contexts (databases, caches, distributed systems). In Bitcoin specifically, they're a historical artifact you'll encounter mostly in older docs.