• New logic-v2 matcher: We're including a new matching system, logic-v2. This new system reflects feedback from users of logic-v1 and introduces a more precise, explainable, and culturally-aware way to match names of people and companies. logic-v2 reduces false positives, is fully deterministic, improves cross-language and cross-script matching, runs fast, and provides detailed explanations of its decision-making path. Please be aware that we'll still be adjusting logic-v2 based on user feedback, so don't expect every scoring decision to be set in stone yet. We look forward to your feedback, be it as an issue report on our GitHub, or on our discussion forum or via the support team.
• Field deprecations and renames:
  • The matcher field in /match responses will be removed in a future version of yente. Equivalent information is available via the /algorithms endpoint.
  • The /match endpoint response has gained a new explanations field that is a strict superset of features. In addition to the partial scores produced by each component of the matching system, it includes textual descriptions of the matching decisions from each of these subsystems. These descriptions are designed for display in analyst workbenches or can be passed to generative AI tools to help interpret screening alerts. The features field will be removed in a future version of yente.
  • The cutoff parameter on the /match endpoint has been deprecated and will be removed in a future version of yente. If you care about low-scoring results being returned, please set the threshold parameter instead.
  • The /algorithms endpoint has gained a new docs field that is a strict superset of features. In addition to a documentation of the algorithm features (yes, those featured in the explanations field in /match), it contains a list of configuration knobs to tune the algorithm to your individual needs. Currently, only logic-v2 offers these knobs.
  • These algorithm configuration options are passed in a new config field on the /match request.
• Locking mechanism to prevent concurrent reindexes: One of the most common issues with yente, both in our hosted deployment an for external users, was two re-index jobs stepping on each other's toes or overloading the Elastic backend, doing the same thing twice. Before we start a reindex, we now write a little lock to the Elastic backend so that other instances of yente who might be planning to do a reindex know that one is already in progress. Regardless, if you're running multiple instances of yente, you should still configure a separate re-indexing cronjob. See documentation on how to deploy yente for more information.
• Audit log of index operations: A log of re-indexing and index cleanup operations is written to a special index in Elastic. This allows users to get an accurate record of what data became available when in their yente without digging through log messages. To read the audit log, simply run yente audit-log (or use another CSV viewer of your choice).
• Google Cloud Logging compatible request logging: If you're running yente on Google Cloud Run, handy little badges with information about the HTTP request as well as a little button to filter logs from only this trace will show up in the Logs Explorer. Careful: the format of the log message logged for each request changed, so if you're parsing the JSON logged by yente, please update your infrastructure accordingly.
• FollowTheMoney 4.0: This release pulls in an updated version of the data model underpinning yente. If you're ingesting custom data sources that you're generating yourself, please check out the release notes. The only change that may be notable for users of yente is probably the rename of CryptoWallet:managingExchange. Users relying on recent published OpenSanctions data can ignore this change.
• Documentation moved to yente.followthemoney.tech: yente is part of the wider open source ecosystem around the FollowTheMoney data model, and the move of its documentation reflects this. The documentation is now part of the source repository and can be edited by anyone. Documentation is never perfect, and we welcome your PRs! The OpenSanctions website remains a good place to read about the data underpinning yente and our hosted API.
• Celebrity-friendly scoring in /search: While /match is at the heart of what yente does best, the /search endpoint is what users usually hit first when they type in a search query on opensanctions.org. By deploying cutting-edge research from our search engine labs, we ensure the Putin ranks first when you type in "putin". Much wow!
• The /updatez endpoint is now disabled by default. Set the UPDATE_TOKEN authentication token to a secret value to enable it.
• The usual round of dependency upgrades, among them an upgrade of cryptography from 45.0.5 to 45.0.6