Rated Ranking Evaluator

Rated Ranking Evaluator (RRE) is an open-source, offline search-quality evaluation library for Apache Solr and Elasticsearch, built and maintained by Sease. Where Quepid is an interactive web dashboard, RRE is a JVM/Maven library designed to run as part of a build — making it the natural fit for CI/CD regression testing of search relevance. Apache 2.0 licensed.


Why It Exists

RRE’s core philosophy is incremental, iterative, and immutable search development: rather than mutating a config in place, you version it, and RRE tracks each version’s metrics and the delta between versions. This turns “did my change help or hurt relevance?” into an automatable, measurable question you can run on every build — search’s answer to unit/regression testing.

Domain Model

RRE organizes an evaluation as a hierarchy, with metrics computed at the leaf (query) level and aggregated upward by arithmetic mean:

  • Corpus — the dataset indexed for evaluation.
  • Topics — top-level thematic groupings of information needs.
  • Query Groups — sets of equivalent/related queries under a topic.
  • Queries — the individual query executed against the engine.
  • Ratings / Judgments — per-(query, document) relevance grades (the ground truth), supplied as JSON.
  • Metrics — computed per query, then rolled up to Query Group, Topic, and whole-evaluation levels.

Metrics

RRE ships a broad set of IR measures out of the box:

Each metric is a first-class, query-level value; the aggregated (“Mean…”) variants come from averaging up the topic hierarchy.

Architecture & Modules

RRE is a set of Maven modules on the JVM:

ModuleRole
rre-coreThe evaluation engine (domain model + metric computation)
rre-search-platformPluggable abstraction over the search engine (Solr / Elasticsearch)
rre-maven-pluginRuns evaluations inside a Maven build
rre-maven-reporting-pluginProduces human-readable reports (spreadsheet / PDF) for non-technical stakeholders
rre-persistence-pluginPersists evaluation output (JSON)
rre-serverWeb control panel; refreshes results in real time after each build
rre-maven-archetypeProject template to bootstrap an RRE setup

Primarily Java, with supporting Python/JS. Because the engine is behind rre-search-platform, the same evaluation can target different backends.

Outputs

  • JSON — machine-readable evaluation results (for pipelines / diffing).
  • Spreadsheet / PDF reports — via the Maven reporting plugin, aimed at non-technical readers.
  • RRE Server — a live web dashboard that updates after each build cycle, showing metrics and version-over-version deltas.

RRE Enterprise

RRE Enterprise is Sease’s commercial layer built on the open-source library — the same metrics/engine wrapped in a full UI and a simplified, less code-centric user experience.

RRE vs. Quepid

Both come from the search-relevance consulting world but differ in shape:

Rated Ranking EvaluatorQuepid
FormJVM/Maven libraryWeb application
Primary modeAutomated, in-build (CI/CD)Interactive, exploratory
AudienceEngineers wiring relevance tests into pipelinesAnalysts + non-technical raters, team judging
EnginesSolr, ElasticsearchEngine-agnostic — any HTTP endpoint (Solr, ES, OpenSearch, Vespa, Qdrant, custom APIs)
MakerSeaseOpenSource Connections

See Relevance Evaluation Tools Compared for the fuller landscape including Search Relevance Workbench and Elasticsearch Relevance Studio.

Companies

  • Sease — creator and maintainer (also offers RRE Enterprise)

Comparison