Self Evolve 中文 Start
Value LSH / English mirror

Value LSH decides what to review next. It does not decide what is true.

The queue compresses 2,236 GitHub, paper, social/X, and blog materials into triage classes. Its job is narrow: choose what to inspect, what to repair, and what not to cite yet.

Open full visual queue Read generated report Project evidence guide
Snapshot

Enough structure to guide reading, not enough to replace review.

Generated 2026-06-05
2,236materials
26triage facets
174LSH buckets
24embedding clusters
1,085 materials

needs review

Needs a human pass before strong public use.

851 materials

priority review

Inspect first, then verify source trail and claim strength.

300 materials

repair or avoid citing

Repair evidence or avoid citation until the source chain improves.

How to use

Turn a large corpus into a review queue.

Use the queue to select the next deep read, then leave the queue and inspect the underlying paper, repository, model card, report, benchmark, or raw capture. The final judgment should be written in the reviewed page, not inferred from the score alone.

Priority review sample

Items to inspect before citing.

github · priority review

AgentEvolver

raw-github/modelscope_agentevolver.md

tag coverage: stronger trail 21 positive facets 1 negative facets
github · priority review

CUGA Agent

raw-github/cuga-project_cuga-agent.md

tag coverage: stronger trail 20 positive facets 1 negative facets
github · priority review

Agent Memory Benchmark

raw-github/vectorize-io_agent-memory-benchmark.md

tag coverage: stronger trail 20 positive facets 1 negative facets
social · priority review

0166 cloud tencent com AI 100

raw-social/0166-cloud-tencent-com-AI-100.md

tag coverage: stronger trail 21 positive facets 1 negative facets
social · priority review

0171 36kr com 7 GPU 36

raw-social/0171-36kr-com-7-GPU-36.md

tag coverage: stronger trail 21 positive facets 1 negative facets
github · priority review

Awesome Agent Harness (AutoJunjie)

raw-github/autojunjie_awesome-agent-harness.md

tag coverage: stronger trail 20 positive facets 1 negative facets
github · priority review

Yunjue Agent

raw-github/yunjuetech_yunjue-agent.md

tag coverage: stronger trail 20 positive facets 1 negative facets
github · priority review

EvoAgentX

raw-github/evoagentx_evoagentx.md

tag coverage: stronger trail 20 positive facets 1 negative facets
Repair queue sample

Weak evidence is useful only when it stays labeled.

github · needs review

agentgpt

github:reworkd/agentgpt

github · needs review

webarena

github:web-arena-x/webarena

github · needs review

CoML

github:microsoft/coml

github · needs review

OpenELM

github:carperai/openelm

github · needs review

openagents

github:xlang-ai/openagents

github · needs review

FunSearch

github:google-deepmind/funsearch