---
language: en
title: "Candidate set"
description: "A candidate set brings together passages selected before reranking. Learn how to inspect this stage to find gaps in retrieval."
canonical: "https://tropk.ai/en/resources/glossary/candidate-set"
last_modified: "2026-10-12T16:13:08.288Z"
---

# Candidate set

Reviewed on 2026-10-06.

## What it means

A candidate set is a list of documents or passages selected by an initial [retrieval](/en/resources/glossary/information-retrieval) stage for a query. This list defines the material a subsequent stage, such as a reranker, can evaluate. Its composition depends on the available collection, the search method, the filters, and the number of results requested at that processing stage.

## Why it matters for marketing

For those evaluating a search experience, distinguishing candidates from final results helps pinpoint the problem more precisely. If a needed passage never made it into the set, discussing its final position will not resolve its absence. A content team may need to fix [indexing](/en/resources/glossary/indexing), filters, or document coverage before requesting adjustments to the model that reranks results.

## Example in practice

In a hypothetical scenario, an insurer tests a question about electric bike coverage. The correct clause is in the collection, but it does not appear among the twenty candidate passages. A filter excluded documents with the newly created category. The team identifies the exclusion in the intermediate list and requests a filter fix instead of changing the generated answer text.

## How to apply or verify

Inspect the retriever’s output before any [reranking](/en/resources/glossary/reranking) or generation. Record the query, filters, passage IDs, and result limit. Mark which documents would be needed to answer and check whether they were selected. Compare different limits using a controlled sample, observing both the inclusion of useful evidence and the cost of processing more candidates per query.

## Important distinctions

A candidate is not necessarily relevant, nor will it necessarily be cited in the answer. The initial set may bring together matches based on words, vectors, or combined methods. In a stage that only reranks a fixed list, a missing document cannot be promoted. Other workflows may perform new searches; in that case, identify which round each set belongs to.

## Sources and references

- [Retrieve & Re-Rank — Sentence Transformers documentation](<https://www.sbert.net/examples/sentence_transformer/applications/retrieve_rerank/README.html>) — Sentence Transformers

## Continue a leitura

- [Information retrieval](</en/resources/glossary/information-retrieval>)
- [Passage retrieval](</en/resources/glossary/passage-retrieval>)
- [Reranking](</en/resources/glossary/reranking>)

## About the author

[Vinicius Pacheco](</en/team/vinicius-pacheco>)

Responsible for Tropk’s technical content, focusing on crawling, indexing, structured data, and agent access to content.

---

[All pages in Markdown](<https://tropk.ai/sitemap.md>)