Search

These docs describe the development of our search engine and the systems we've used to test it.

The story so far

We began with a data-informed approach to measuring relevance, running candidate queries against each other in AB tests, and taking incremental steps based on some illustrative metrics. After a few months of development, we found that this approach was insufficient to drive the development of new features safely or reliably. Most of this documentation is now outdated as a result.

Instead, we started developing rank, a system which uses known examples of tricky search-term / document pairs to constrain the query. Rank is the system we use to develop queries and mappings today.

Current queries

You can read an explanation of the current queries' structure here.