# rcb-official-20261010.md

Source: https://internscience.github.io/ResearchClawBench-Home/
Retrieved: 2026-10-10

## ResearchClawBench

Evaluating AI Agents for Automated Research from **Re-Discovery to New-Discovery**

40

Tasks

10

Domains

### Frontier   Pass@1Pass@5  i  Pass@1 reports single-run leaderboard scores after rerunning failed jobs when possible, emphasizing completed single-run capability. Pass@5 uses five attempts per task, takes the best score for frontier/ranking, and treats failed or unscored tasks as 0 to emphasize stability and coverage.

Best score per task across all agents. 50 = matches original paper, 100 = surpasses it.

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/qiushi.png)Qiushi EngineDeepSeek-V4-Pro-Preview

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/innoclaw.jpg)InnoClaw (2.0)GPT-5.5

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/innoclaw.jpg)InnoClawGPT-5.5

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/ligase.png)LigaseGPT-5.6-Sol

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/autoscirub.png)AutoSciRubCodex GPT-5.6-Sol

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/evolab.png)EvoLabGPT-5.5

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openscience.png)Open ScienceClaude-Opus-4.8

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/anthropic.svg)Claude CodeClaude-Opus-4.6

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/evo.svg)EvoScientist (0.1.1)GPT-5.4

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/evo.svg)EvoScientist (0.0.4)GPT-5.4

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openai.svg)Codex CLIGPT-5.4

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openclaw.svg)OpenClawGPT-5.4

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/researchclaw.svg)ResearchClawGPT-5.4

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/asx.svg)ARIS CodexGPT-5.4

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/nanobot.svg)NanobotGPT-5.4

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/anthropic.svg)Claude-Opus-4.8

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/anthropic.svg)Claude-Opus-4.7

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/anthropic.svg)Claude-Opus-4.6

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/glm.webp)GLM-5.2

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/glm.webp)GLM-5.1

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/minimax.png)MiniMax-M3

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/qwen.png)Qwen3.7-Max

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/qwen.png)Qwen3.6-Plus

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/qwen.png)Qwen3.5-397B-A17B

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/kimi.png)Kimi-K2.6

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/kimi.png)Kimi-K2.5

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/gemini.png)Gemini-3.5-Flash

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/gemini.png)Gemini-3.1-Pro

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/deepseek.png)DeepSeek-V4-Pro

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openai.svg)GPT-5.5

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openai.svg)GPT-5.4

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/mimo.png)MiMo-V2.5

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/mimo.png)MiMo-V2-Pro

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/grok.png)Grok-4.1

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/grok.png)Grok-4.3

![](https://internscience.github.io/ResearchClawBench-Home/static/logos/hunyuan.png)Hy3-Preview

Frontier

Human Level (50)

Note: All standalone LLM results below are evaluated with [ResearchHarness](https://huggingface.co/spaces/InternScience/ResearchHarness).

### Leaderboard 2026-10-10 07:33:28 UTC   Pass@1Pass@5  i  Pass@1 reports single-run leaderboard scores after rerunning failed jobs when possible, emphasizing completed single-run capability. Pass@5 uses five attempts per task, takes the best score for frontier/ranking, and treats failed or unscored tasks as 0 to emphasize stability and coverage.

Slide to view more agents

By Domain

Slide to view more domains

| Agent/LLM | Overall | Source | Astronomy | Chemistry | Earth | Energy | Information | Life | Material | Math | Neuroscience | Physics |
| --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- | --- |
| 🥇![](https://internscience.github.io/ResearchClawBench-Home/static/logos/qiushi.png) [Qiushi Engine](https://oxelra.com/)DeepSeek-V4-Pro-Preview | 38.6$7.62h 55m | Community | 42.8$5.62h 09m | 30.6$114h 23m | 44.4$3.21h 14m | 45.7$2.352m 15s | 39.8$145h 24m | 30.3$2.353m 45s | 41.2$103h 52m | 33.2$9.53h 39m | 32.6$103h 57m | 45.7$7.42h 49m |
| 🥈![](https://internscience.github.io/ResearchClawBench-Home/static/logos/innoclaw.jpg) [InnoClaw (2.0)](https://github.com/SpectrAI-Initiative/InnoClaw)GPT-5.5 | 34.9$8.939m 39s | Community | 41.6$3.515m 35s | 32.7$9.642m 42s | 27.6$3.414m 58s | 30.4$241h 47m | 36.0$5.926m 10s | 32.8$7.131m 24s | 32.4$6.830m 14s | 34.2$3.816m 55s | 35.9$221h 39m | 45.2$2.711m 49s |
| 🥉![](https://internscience.github.io/ResearchClawBench-Home/static/logos/ligase.png) [Ligase](https://gradence.com.cn/products/wecode.html#biopharma)GPT-5.6-Sol | 34.8$7.747m 49s | Community | 37.4$7.949m 32s | 22.9$7.949m 14s | 34.3$5.635m 10s | 29.9$7.144m 12s | 40.1$7.345m 54s | 32.5$7.345m 49s | 35.1$9.157m 03s | 39.7$7.245m 15s | 35.3$9.61h 00m | 41.0$7.446m 05s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/autoscirub.png)[AutoSciRub](https://github.com/zjunlp/AutoSciRub)Codex GPT-5.6-Sol | 33.2$7.849m 03s | Community | 42.3$4.528m 12s | 30.1$121h 15m | 30.2$6.037m 14s | 31.7$6.339m 07s | 33.5$7.647m 40s | 26.2$7.043m 29s | 34.5$7.647m 48s | 28.8$8.653m 52s | 30.0$111h 06m | 44.4$8.150m 37s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/evolab.png)[EvoLab](https://evo-lab.org/)GPT-5.5 | 32.7$6.528m 41s | Community | 44.8$4.118m 04s | 27.5$1147m 51s | 29.8$5.725m 21s | 33.9$7.031m 19s | 38.2$6.127m 13s | 28.8$3.917m 09s | 24.9$6.026m 29s | 27.6$5.423m 47s | 20.1$9.441m 38s | 51.3$6.327m 54s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/innoclaw.jpg)[InnoClaw](https://github.com/SpectrAI-Initiative/InnoClaw)GPT-5.5 | 28.8$7.131m 35s | Community | 41.6$3.515m 35s | 18.0$6.327m 53s | 27.6$3.414m 58s | 30.4$241h 47m | 24.1$5.424m 13s | 24.8$5.122m 33s | 32.4$6.830m 14s | 27.9$7.332m 33s | 16.3$6.428m 32s | 45.2$2.711m 49s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openscience.png)[Open Science](https://github.com/ai4s-research/open-science)Claude-Opus-4.8 | 22.8$3.417m 11s | Internal | 29.4$2.412m 13s | 9.8$6.030m 07s | 19.6$1.99m 19s | 18.4$2.110m 20s | 20.6$4.522m 20s | 16.8$1.68m 09s | 25.8$4.824m 02s | 27.2$6.733m 25s | 13.3$3.718m 40s | 40.0$1.89m 10s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/anthropic.svg)Claude CodeClaude-Opus-4.6 | 21.5$5.326m 15s | Internal | 30.2$2.311m 27s | 9.3$141h 09m | 22.7$2.311m 39s | 21.7$5.527m 30s | 25.0$4.019m 49s | 15.6$1.88m 51s | 25.5$141h 10m | 27.5$4.522m 16s | 5.5$2.512m 24s | 32.3$1.78m 29s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/evo.svg)[EvoScientist (0.1.1)](https://github.com/EvoScientist/EvoScientist)GPT-5.4 | 18.8$4.136m 17s | Community | 26.4$2.118m 43s | 9.9$4.741m 58s | 18.8$2.724m 08s | 19.9$5.145m 26s | 18.9$3.127m 27s | 16.7$4.035m 18s | 15.2$7.61h 07m | 15.5$3.934m 39s | 11.4$3.934m 57s | 34.9$3.733m 01s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openai.svg)Codex CLIGPT-5.4 | 18.4$2.017m 53s | Internal | 26.5$1.311m 21s | 7.6$3.430m 28s | 22.2$1.412m 52s | 23.1$1.916m 45s | 17.0$1.815m 35s | 14.4$1.714m 50s | 13.0$2.824m 52s | 20.8$2.118m 16s | 8.6$2.724m 24s | 31.0$1.19m 28s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openclaw.svg)OpenClawGPT-5.4 | 16.6$0.696m 06s | Internal | 28.4$0.494m 21s | 6.0$0.776m 51s | 17.3$0.615m 23s | 22.0$0.766m 47s | 14.0$0.625m 32s | 15.7$0.686m 00s | 12.9$0.635m 38s | 14.3$0.716m 19s | 8.3$0.998m 49s | 27.6$0.605m 18s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/researchclaw.svg)ResearchClawGPT-5.4 | 16.3$0.827m 15s | Internal | 23.1$0.544m 46s | 8.5$1.211m 07s | 17.1$0.534m 42s | 19.0$0.474m 12s | 14.9$0.817m 12s | 14.0$0.797m 03s | 19.3$1.09m 00s | 12.8$1.08m 56s | 4.2$1.412m 15s | 30.1$0.383m 22s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/evo.svg)[EvoScientist (0.0.4)](https://github.com/EvoScientist/EvoScientist)GPT-5.4 | 15.5$1.210m 37s | Internal | 26.0$0.817m 10s | 4.4$1.917m 15s | 17.0$0.706m 13s | 24.0$1.210m 33s | 7.4$1.311m 11s | 16.4$0.766m 47s | 13.4$1.714m 57s | 14.3$0.907m 59s | 5.3$1.816m 20s | 26.3$0.877m 45s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/asx.svg)[ARIS Codex](https://github.com/wanshuiyin/auto-claude-code-research-in-sleep)GPT-5.4 | 13.6$0.746m 36s | Internal | 21.3$0.534m 41s | 7.4$1.311m 43s | 15.1$0.494m 22s | 13.6$0.675m 58s | 6.0$0.655m 46s | 16.9$0.544m 47s | 12.4$0.837m 23s | 11.4$0.847m 28s | 6.9$0.918m 08s | 24.7$0.665m 51s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/nanobot.svg)[Nanobot](https://github.com/HKUDS/nanobot)GPT-5.4 | 12.8$0.494m 21s | Internal | 22.3$0.312m 45s | 6.2$1.210m 49s | 14.3$0.413m 37s | 13.5$0.464m 05s | 11.1$0.686m 01s | 12.9$0.383m 23s | 13.0$0.443m 53s | 12.1$0.262m 21s | 3.3$0.322m 49s | 19.4$0.423m 44s |
| 🥇![](https://internscience.github.io/ResearchClawBench-Home/static/logos/anthropic.svg)Claude-Opus-4.8 | 21.1$4.019m 52s | Internal | 34.2$1.89m 05s | 17.9$1153m 58s | 21.9$1.99m 40s | 18.5$2.512m 42s | 20.3$2.512m 29s | 13.2$1.99m 34s | 24.7$3.919m 41s | 16.4$8.542m 20s | 4.5$4.421m 55s | 38.9$3.215m 52s |
| 🥈![](https://internscience.github.io/ResearchClawBench-Home/static/logos/glm.webp)GLM-5.2 | 20.7$1.242m 20s | Internal | 36.3$0.9833m 25s | 9.1$1.756m 26s | 19.1$0.6923m 40s | 23.3$1.239m 48s | 11.1$1.138m 09s | 17.6$0.6120m 53s | 22.3$1.655m 41s | 26.4$1.91h 03m | 9.7$1.653m 43s | 35.8$1.136m 28s |
| 🥉![](https://internscience.github.io/ResearchClawBench-Home/static/logos/anthropic.svg)Claude-Opus-4.7 | 20.7$5.125m 18s | Internal | 32.9$1.78m 34s | 4.2$141h 08m | 22.5$2.512m 34s | 23.2$2.914m 27s | 13.9$8.844m 01s | 12.8$3.215m 57s | 24.1$9.346m 43s | 18.9$5.426m 56s | 9.7$3.919m 36s | 34.2$2.813m 48s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/anthropic.svg)Claude-Opus-4.6 | 19.9$6.030m 14s | Internal | 35.1$4.823m 52s | 7.2$6.633m 10s | 30.4$2.110m 34s | 26.0$3.517m 22s | 12.8$1157m 13s | 12.7$2.311m 20s | 20.0$171h 26m | 22.4$3.819m 07s | 7.4$3.014m 49s | 35.0$3.819m 10s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/minimax.png)MiniMax-M3 | 19.8$0.4542m 45s | Internal | 30.5$0.2523m 32s | 11.9$0.861h 22m | 18.7$0.3634m 13s | 24.0$0.3634m 01s | 13.2$0.3936m 54s | 12.7$0.3431m 56s | 17.3$0.5754m 17s | 21.7$0.651h 02m | 8.5$0.4643m 22s | 35.1$0.2523m 28s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/qwen.png)Qwen3.7-Max | 18.7$0.4211m 08s | Internal | 23.8$0.379m 47s | 6.9$0.6718m 00s | 14.8$0.287m 36s | 25.3$0.359m 17s | 21.8$0.318m 14s | 10.3$0.205m 18s | 23.9$0.6818m 00s | 22.3$0.307m 56s | 7.0$0.6818m 07s | 38.3$0.236m 10s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/glm.webp)GLM-5.1 | 18.2$1.330m 13s | Internal | 29.8$1.227m 17s | 11.4$2.148m 11s | 20.6$0.7116m 34s | 20.4$0.8920m 47s | 16.2$0.7818m 13s | 12.2$0.9421m 51s | 18.6$2.81h 03m | 18.0$1.842m 53s | 5.8$0.9822m 46s | 28.9$0.8519m 41s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/qwen.png)Qwen3.6-Plus | 18.0$0.5223m 02s | Internal | 30.8$0.5926m 16s | 7.7$1.147m 21s | 16.6$0.2511m 20s | 21.2$0.4017m 36s | 17.7$0.5524m 29s | 12.1$0.3615m 52s | 19.6$0.6227m 33s | 19.1$0.6930m 53s | 4.6$0.4118m 23s | 30.7$0.2410m 38s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/kimi.png)Kimi-K2.6 | 18.0$1.234m 41s | Internal | 27.8$1.647m 00s | 3.1$1.336m 33s | 22.9$1.236m 20s | 18.1$0.8524m 47s | 12.8$1.337m 01s | 13.8$0.5917m 12s | 22.3$1.338m 18s | 19.5$1.956m 41s | 8.3$1.336m 41s | 24.0$0.5917m 11s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/gemini.png)Gemini-3.5-Flash | 17.9$1.217m 50s | Internal | 28.1$1.826m 17s | 5.5$0.8913m 09s | 24.0$1.0014m 45s | 18.3$1.420m 12s | 24.1$1.623m 46s | 13.4$0.6810m 05s | 19.7$1.522m 49s | 22.0$0.8913m 13s | 2.2$1.318m 58s | 26.8$0.7310m 53s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/deepseek.png)DeepSeek-V4-Pro | 17.1$0.2825m 33s | Internal | 25.2$0.1514m 08s | 7.5$0.5147m 05s | 20.1$0.2522m 38s | 21.8$0.2219m 59s | 4.6$0.2422m 27s | 13.3$0.2119m 02s | 24.6$0.4642m 06s | 17.9$0.3734m 04s | 7.7$0.1816m 33s | 28.6$0.2119m 37s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openai.svg)GPT-5.5 | 17.0$1.88m 06s | Internal | 24.0$1.46m 04s | 10.4$1.67m 14s | 18.4$0.984m 20s | 22.6$2.912m 41s | 17.8$2.712m 02s | 11.8$2.29m 40s | 14.7$2.611m 26s | 13.1$1.56m 31s | 6.3$1.66m 58s | 30.9$0.914m 03s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/mimo.png)MiMo-V2.5 | 16.9$0.2313m 39s | Internal | 24.4$0.1710m 00s | 4.4$0.158m 53s | 15.9$0.2313m 49s | 19.5$0.1810m 33s | 12.3$0.6941m 19s | 12.8$0.095m 35s | 19.2$0.1810m 30s | 18.2$0.2313m 40s | 4.7$0.1911m 32s | 34.7$0.169m 33s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/openai.svg)GPT-5.4 | 15.3$2.118m 53s | Internal | 22.1$1.19m 56s | 6.9$2.017m 23s | 18.8$0.847m 29s | 17.1$1.311m 37s | 15.7$1.311m 44s | 14.2$1.614m 04s | 7.7$1.714m 50s | 18.3$1.311m 28s | 4.9$8.71h 17m | 27.0$1.513m 13s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/mimo.png)MiMo-V2-Pro | 15.3$0.6220m 39s | Internal | 21.8$1.549m 05s | 6.3$1.239m 27s | 19.8$0.3311m 03s | 12.8$0.7524m 56s | 9.3$0.4615m 25s | 12.8$0.237m 43s | 21.4$0.4214m 10s | 13.6$0.5217m 19s | 6.2$0.5016m 46s | 26.4$0.248m 03s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/qwen.png)Qwen3.5-397B-A17B | 14.2$0.5821m 22s | Internal | 17.1$0.4115m 20s | 7.8$0.7025m 48s | 18.3$0.4215m 24s | 18.6$0.5219m 26s | 6.0$0.4315m 58s | 11.5$0.3713m 34s | 18.4$0.9434m 53s | 14.3$0.8230m 18s | 4.2$0.6825m 16s | 26.1$0.4817m 39s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/kimi.png)Kimi-K2.5 | 14.0$0.5121m 20s | Internal | 23.4$0.3916m 18s | 7.6$0.5522m 53s | 16.4$0.3413m 59s | 13.3$0.4318m 03s | 9.4$0.4418m 09s | 11.4$0.4318m 07s | 13.1$0.6928m 35s | 16.1$0.7732m 05s | 3.0$0.6024m 58s | 26.5$0.5522m 55s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/grok.png)Grok-4.1 | 13.5$0.046m 30s | Internal | 28.6$0.035m 58s | 2.9$0.035m 50s | 14.9$0.024m 15s | 11.1$0.047m 22s | 11.9$0.035m 22s | 13.2$0.034m 54s | 9.5$0.047m 14s | 12.2$0.036m 00s | 4.8$0.059m 40s | 25.6$0.058m 17s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/gemini.png)Gemini-3.1-Pro | 13.3$1.617m 47s | Internal | 19.3$1.213m 42s | 8.0$1.820m 10s | 13.8$0.495m 29s | 12.0$0.768m 27s | 6.8$1.011m 26s | 9.4$1.819m 36s | 11.3$5.359m 15s | 15.0$1.516m 25s | 1.6$0.323m 32s | 29.7$2.325m 14s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/hunyuan.png)Hy3-Preview | 12.9$0.0314m 18s | Internal | 13.6$0.028m 53s | 15.1$0.0317m 26s | 14.2$0.0316m 13s | 11.1$0.0314m 20s | 14.2$0.0211m 32s | 10.5$0.028m 55s | 16.6$0.0313m 13s | 8.3$0.0314m 33s | 1.4$0.0420m 39s | 26.4$0.0312m 53s |
| ![](https://internscience.github.io/ResearchClawBench-Home/static/logos/grok.png)Grok-4.3 | 12.4$0.4414m 01s | Internal | 24.7$0.309m 43s | 3.5$0.5718m 18s | 15.1$0.3410m 45s | 18.8$0.309m 40s | 2.5$0.278m 43s | 12.5$0.3210m 11s | 12.3$0.6019m 12s | 8.5$0.3711m 55s | 2.6$1.237m 27s | 28.2$0.072m 12s |


The displayed UTC timestamp is generated by new Date() on the client (app.js updateClock), not a result-update date.

Official app.js:
function averageEntry(entries) {
    const scored = entries.filter(e => Number.isFinite(e?.score));
    if (!scored.length) return null;
    if (isPass5) {
      const attempts = scored.flatMap(entry => Array.isArray(entry.runs) ? entry.runs.filter(Number.isFinite) : []);
      const mean = values => values.length ? values.reduce((a, b) => a + b, 0) / values.length : null;
      const attemptMean = mean(attempts);
      const attemptStd = attempts.length && Number.isFinite(attemptMean)
        ? Math.sqrt(mean(attempts.map(v => (v - attemptMean) ** 2)))
        : null;
      const scoreMean = mean(scored.map(e => e.score).filter(Number.isFinite));
      return {
        score: scoreMean,
        mean: attemptMean ?? scoreMean,
        std: attemptStd,
        min: attempts.length ? Math.min(...attempts) : Math.min(...scored.map(e => e.score).filter(Number.isFinite)),
        max: attempts.length ? Math.max(...attempts) : Math.max(...scored.map(e => e.score).filter(Number.isFinite)),
        runs: attempts.slice(0, 20),
        runs_truncated: attempts.length > 20,
        attempts: scored.reduce((sum, entry) => sum + (Number.isFinite(entry.attempts) ? entry.attempts : 0), 0),
        scored_runs: attempts.length,
        tasks: scored.length,
        details_state: 'none',
      };
    }
    const average = key => {
      const values = scored.map(e => e[key]).filter(Number.isFinite);
      return values.length ? values.reduce((a, b) => a + b, 0) / values.length : null;
    };
    return {
      score: average('score'),
      duration_seconds: average('duration_seconds'),
      cost_usd: average('cost_usd'),
      details_state: getEntriesDetailsState(scored),
    };
  }