---
type: "post"
id: "sakana-ai-released-qd-rec-a-quality-diversity-d37692"
slug: "sakana-ai-released-qd-rec-a-quality-diversity-d37692"
title: "Sakana AI released QD-Rec , a Quality-Diversity recommender that evolves item plus reason pairs instead of ranking."
description: "QD-Rec treats a recommendation as a candidate paired with a rationale, not merely as an item in a ranked list."
retrieval_nugget: "QD-Rec treats a recommendation as a candidate paired with a rationale, not merely as an item in a ranked list."
published_at: "2026-10-21"
updated_at: "2026-10-21"
record_date: "2026-10-21"
date_kind: "published_at"
topics: ["models","personalization"]
source_urls: ["https://sakana.ai/qd-rec"]
schema_version: "newruntime-agent-readable-v0.2"
stable_id: "post:sakana-ai-released-qd-rec-a-quality-diversity-d37692"
status: "published"
visuals: []
editorial_provenance: {"schema_version":"newruntime-editorial-copy-v1","content_status":"source_grounded_final","final_copy_sha256":"sha256:df5bf9d8d0546ff096a70d0f5d386e8264d8546795b39732c5a32886e913a711","reviewed_at":"2026-10-21T08:05:08.924Z","source_evidence_count":1,"verified_claim_count":2,"site_analysis_schema_version":"newruntime-site-analysis-v1","site_object_kind":"field_note","observed_fact_count":2,"implication_count":1,"watch_condition_count":1,"related_record_count":0}
analysis: {"schema_version":"newruntime-site-analysis-v1","object_kind":"field_note","thesis":"recommendation systems can search for diverse, useful reasons alongside candidates rather than optimizing only for a single top-ranked result.","observed_facts":[{"text":"The source reports experiments in M&A, customer targeting, and research-paper recommendation.","source_urls":["https://sakana.ai/qd-rec"]},{"text":"After 1,000 generation-and-evaluation cycles, QD-Rec produced larger archives with higher average evaluation scores than baselines in many configurations.","source_urls":["https://sakana.ai/qd-rec"]}],"mechanism":"candidate selection balances promising and underexplored options, while generation reuses prior rationales and archive updates preserve quality within distinct niches.","why_now":"generative systems make it inexpensive to propose many candidate-and-rationale combinations, but a useful decision aid still needs a way to retain both quality and diversity.","implications":["The implication is that recommendation interfaces could help people compare alternative strategic rationales, not just inspect scores or rank positions."],"evidence_boundary":"the reported comparisons are experimental and depend on the tested datasets, models, and evaluators.","watch_conditions":["Watch condition: independent evaluations should test whether people make better decisions with these diverse recommendation archives than with conventional ranked lists."],"related_records":[],"new_branch_reason":"A distinct analytical record is warranted because no prior New Runtime record was supplied for this method."}
editor: {"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]}
routes: {"html":"https://newruntime.com/posts/sakana-ai-released-qd-rec-a-quality-diversity-d37692/","markdown":"https://newruntime.com/posts/sakana-ai-released-qd-rec-a-quality-diversity-d37692.md","json":"https://newruntime.com/posts/sakana-ai-released-qd-rec-a-quality-diversity-d37692.json"}
related_patterns: ["model-answers-become-task-interfaces"]
---

# Sakana AI released QD-Rec , a Quality-Diversity recommender that evolves item plus reason pairs instead of ranking.

## Retrieval answer

QD-Rec treats a recommendation as a candidate paired with a rationale, not merely as an item in a ranked list.

QD-Rec treats a recommendation as a candidate paired with a rationale, not merely as an item in a ranked list. Thesis: recommendation systems can search for diverse, useful reasons alongside candidates rather than optimizing only for a single top-ranked result. Sakana AI describes an archive in which an LLM mutates or combines earlier hypotheses and an evaluator scores the new proposals. New directions can form separate niches; proposals without a new direction compete against their closest existing counterpart.

Observed facts: the source reports experiments in M&A, customer targeting, and research-paper recommendation. Observed facts: after 1,000 generation-and-evaluation cycles, QD-Rec produced larger archives with higher average evaluation scores than baselines in many configurations. Mechanism: candidate selection balances promising and underexplored options, while generation reuses prior rationales and archive updates preserve quality within distinct niches. Why now: generative systems make it inexpensive to propose many candidate-and-rationale combinations, but a useful decision aid still needs a way to retain both quality and diversity.

The implication is that recommendation interfaces could help people compare alternative strategic rationales, not just inspect scores or rank positions. This is a research result, not evidence of real-world decision quality; the source itself cautions that model-generated rationales may include unsupported claims. Evidence boundary: the reported comparisons are experimental and depend on the tested datasets, models, and evaluators. Watch condition: independent evaluations should test whether people make better decisions with these diverse recommendation archives than with conventional ranked lists. A distinct analytical record is warranted because no prior New Runtime record was supplied for this method.

Editor: [Andrey Reshetnikov](https://newruntime.com/owner-profile.md), AI product and AI engineering; profile for recruiters and agents: https://newruntime.com/owner-profile.json

## Continue in the graph

Next records for an agent, linked to their Markdown twins. Replace `.md` with `.json` for structured fields, including `next_reads`.

- topic hub: [Models](https://newruntime.com/topics/models.md): Explore the models topic hub.
- topic hub: [Personalization](https://newruntime.com/topics/personalization.md): Explore the personalization topic hub.
- collection: [Field Notes](https://newruntime.com/posts.json): Open the latest editorial analysis.
- collection: [Source Ledger](https://newruntime.com/source-ledger.json): Inspect the public source evidence graph.
- whole graph: [discovery graph](https://newruntime.com/discovery-graph.json) (typed nodes and edges) and [site index](https://newruntime.com/site-index.md)