[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-105283":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":12,"contributorsCount":12,"subscribersCount":12,"size":12,"stars1d":14,"stars7d":14,"stars30d":14,"stars90d":12,"forks30d":12,"starsTrendScore":15,"compositeScore":16,"rankGlobal":9,"rankLanguage":9,"license":17,"archived":18,"fork":18,"defaultBranch":19,"hasWiki":20,"hasPages":18,"topics":21,"createdAt":9,"pushedAt":9,"updatedAt":27,"readmeContent":28,"aiSummary":9,"trendingCount":12,"starSnapshotCount":12,"syncStatus":29,"lastSyncTime":30,"discoverSource":31},105283,"dsh-harness-evolution","HuangShengZeBlueSky\u002Fdsh-harness-evolution","HuangShengZeBlueSky","Evolved DSH plugin harnesses for five agent benchmarks, with versioned artifacts, evolution curves and Life-Harness comparisons",null,"HTML",67,0,52,11,33,62.6,"MIT License",false,"main",true,[22,23,24,25,26],"agent-benchmark","dsh","harness","leaderboard","plugin-evolution","2026-11-01 00:13:27","# DSH Harness Evolution\n\n**English** | [简体中文](README.zh-CN.md)\n\n**Methods, evolution curves, reproducible configurations, and measured plugin artifacts.**\n\nFreeze the actor and evolve its runtime harness. Inspect what each loop changes, how candidates are selected, what happened, and which reproduction steps are actually supported.\n\n[**Interactive summary and all curves**](https:\u002F\u002Fhuangshengzebluesky.github.io\u002Fdsh-harness-evolution\u002F?lang=en) · [**Methods**](docs\u002FMETHODS.md) · [**Reproduce**](docs\u002FREPRODUCING.md) · [**Leaderboard**](LEADERBOARD.md)\n\n## Five benchmark artifacts\n\n| Benchmark | Best released final-selected artifact | Retrospective high point | Life-Harness v3 reference |\n|---|---:|---:|---:|\n| ALFWorld | [95.41% · v18](harnesses\u002Falfworld\u002F20261026-method1-sol61-v18) | — | 88.07% |\n| DBBench | [70.56% · v35](harnesses\u002Fdbbench\u002F20261026-l4-h0-sol61-v35) | — | 65.00% |\n| WebShop | [44.50% · v28](harnesses\u002Fwebshop\u002F20261026-l5-h0-sol61-v28) | [47.00% · v12](harnesses\u002Fwebshop\u002F20261027-original-best-sol61-v12) | 45.50% |\n| OSBench | [38.89% · v17](harnesses\u002Fosbench\u002F20261026-l4-h0-sol61-v17) | [39.35% · v03](harnesses\u002Fosbench\u002F20261030-l4-h0-sol61-v03) | 39.60% |\n| Telecom | [84.17% · v06](harnesses\u002Ftelecom\u002F20261026-start-luna-repaired-v06) | — | 90.00% |\n\nThese artifacts come from different methods and protocols, not one unified algorithm. Each final artifact follows its campaign selection rule; cross-campaign ranking and retrospective peaks use observed tests. Paper budgets, repeats, and Telecom simulator\u002Faccess differ. The full leaderboard retains ALIGN, AgentSquare, Meta-Harness, AHE, and historical Life-Harness versions.\n\n## What the audit established\n\n- **More loop machinery did not reliably improve results.** L4\u002FL5 remain strong; L8 delivered experience without consistent gains, and L9 did not dominate the earlier checkpoints.\n- **Protocol changes are additional variables.** DB\u002FWeb L4 use 300 training tasks and no separate validation; L5 uses 240+60. Their score difference is not a clean component ablation.\n- **Selection leaves headroom.** Web retrospective 47.00% exceeds that campaign final 44.33%; OS early 39.35% exceeds final 38.89%. Original selections are preserved.\n- **Overall gain and mechanism support are separate.** Loading, triggering, and experience delivery are execution facts; intended repairs and global outcome changes require additional evidence.\n\n- **Mean success and reliability can diverge.** OS mechanism feedback reached 32.87% versus summary 29.40%, yet all-three success was 25.00% versus 26.39%. This is a single-search observation.\n\n![Final tests by method](assets\u002Fmethod-final-tests.png)\n\n## Reproduce from a shared core\n\n`evolution\u002F` supplies shared three-benchmark L1–L9 code, 36 profiles, task IDs and a small offline example. Each method documents inputs → edits → evaluation → adoption → final selection, with explicit reproduction coverage.\n\n```bash\npython -m pip install -e evolution\ndsh-evolve profiles\npython evolution\u002Fscripts\u002Fsmoke.py .\u002Fworkspaces\u002Fsmoke\n```\n\nThis scripted-model example exercises real DSH hooks and one adoption cycle; it does not remeasure benchmark scores. Formal runs require models, data and environments. ALFWorld\u002FTelecom and feedback-extension integration boundaries are documented.\n\n## Read and update the data\n\n[Round records and detailed statistics](results\u002Fresearch\u002FREADME.md) · [Metric definitions](docs\u002FMETRICS.md) · [All artifacts](LEADERBOARD.md) · [Publishing](docs\u002FPUBLISHING.md)\n\nTraining, validation and test measurements remain separate. Inspect repeats, valid coverage, adoption\u002Fregression, models and splits. Time fields state their scope; missing costs stay null; three test trials are not three independent searches. New measured records rebuild the site, which shows a collection date rather than pretending to be a live monitor.\n\n[MIT](LICENSE) · [Third-party notices](docs\u002FTHIRD_PARTY.md) · [DSH](https:\u002F\u002Fgithub.com\u002Fdeepseek-ai\u002Fdeepseek-harness)\n",2,"2026-10-31 04:26:18","CREATED_QUERY"]