Align EvoDev-Bench dataset card and Croissant metadata

README.md

@@ -12,7 +12,7 @@ tags:
 - multi-turn
 - executable-evaluation
 - arxiv:2605.24110
-pretty_name: EvoCode-Bench
+pretty_name: EvoDev-Bench
 configs:
 - config_name: default
   data_files:
@@ -20,15 +20,16 @@ configs:
     path: metadata/task_viewer.csv
 ---
 
-# EvoCode-Bench
+# EvoDev-Bench
 
-EvoCode-Bench is a benchmark dataset for evaluating coding agents in persistent multi-turn software engineering interactions. It uses the Harbor official multi-step task format, and this release provides a task-level viewer manifest plus downloadable executable archives. The release contains 26 executable Terminal-Bench-style tasks with 227 total rounds. Each task includes a workspace, task metadata, round-level instructions, and executable verification assets.
+EvoDev-Bench is a benchmark dataset for evaluating coding agents in persistent multi-turn software engineering interactions. This public release is forked from EvoCode-Bench and retains its 26 executable tasks and 227 rounds. It uses the Harbor official multi-step task format and provides a task-level viewer manifest plus downloadable executable archives. Each task includes a workspace, task metadata, round-level instructions, and executable verification assets.
 
 ## Links
 
 - Paper: https://arxiv.org/abs/2605.24110
-- Dataset: https://huggingface.co/datasets/UnipatAI/EvoCodeBench
-- Code & evaluation (GitHub): https://github.com/UniPat-AI/EvoCodeBench
+- Dataset: https://huggingface.co/datasets/eachsheep/EvoDev-Bench
+- Code & evaluation (GitHub): https://github.com/EachSheep/EvoDev-Bench
+- Upstream dataset: https://huggingface.co/datasets/UnipatAI/EvoCodeBench
 - Interactive results: https://unipat-ai.github.io/EvoCodeBench/
 - Evaluation framework: Harbor official multi-step (https://harborframework.com/docs/tasks/multi-step); single-round fast-forward fork: https://github.com/UniPat-AI/harbor-official-fast-forward
 
@@ -49,7 +50,7 @@ before each agent phase). The v1 leaderboard and trajectories (June 13–16) are
 **superseded**. The same re-release also corrected one contaminated task (d12_w1) and 11
 task/test defects. Per-(task, model, round) detail and the full changelog are in the
 GitHub repo's README ("Known issues & responsible disclosure") and
-[`CHANGELOG.md`](https://github.com/UniPat-AI/EvoCodeBench/blob/main/CHANGELOG.md). If you
+[`CHANGELOG.md`](https://github.com/EachSheep/EvoDev-Bench/blob/main/CHANGELOG.md). If you
 evaluate with stock upstream Harbor in shared mode, apply the same sanitization or use
 separate-verifier mode.
 
@@ -82,9 +83,9 @@ The dataset is not intended for model training, human programmer assessment, saf
 ## Citation
 
 ```bibtex
-@misc{shen2026evocodebench,
-  title = {EvoCode-Bench: Evaluating Coding Agents in Multi-Turn Iterative Interactions},
-  author = {Haiyang Shen and Xuanzhong Chen and Wendong Xu and Yun Ma and Liang Chen and Kuan Li},
+@misc{shen2026evodevbench,
+  title = {EvoDev-Bench: Evaluating Coding Agents in Multi-Turn Iterative Interactions},
+  author = {Haiyang Shen and Xuanzhong Chen and Wendong Xu and Liang Chen and Kuan Li and Yun Ma},
   year = {2026},
   eprint = {2605.24110},
   archivePrefix = {arXiv},
@@ -97,4 +98,4 @@ The dataset is not intended for model training, human programmer assessment, saf
 
 The benchmark uses synthetic and curated software tasks. It is designed not to include personal or sensitive information. Users should still run their own secret/PII checks before redistribution.
 
-Dataset URL for submission: `https://huggingface.co/datasets/UnipatAI/EvoCodeBench`
+Dataset URL for submission: `https://huggingface.co/datasets/eachsheep/EvoDev-Bench`

croissant.json

@@ -60,11 +60,11 @@
     "sha256": "sc:sha256"
   },
   "@type": "sc:Dataset",
-  "@id": "https://huggingface.co/datasets/UnipatAI/EvoCodeBench#dataset",
-  "name": "EvoCode-Bench",
-  "alternateName": "EvoCode-Bench: Evaluating Coding Agents in Multi-Turn Iterative Interactions",
-  "description": "EvoCode-Bench is a benchmark dataset for evaluating coding agents in persistent multi-turn software development. The release contains 26 executable Terminal-Bench-style tasks and 227 total rounds, with task-only and sanitized trajectory variants.",
-  "url": "https://huggingface.co/datasets/UnipatAI/EvoCodeBench",
+  "@id": "https://huggingface.co/datasets/eachsheep/EvoDev-Bench#dataset",
+  "name": "EvoDev-Bench",
+  "alternateName": "EvoDev-Bench: Evaluating Coding Agents in Multi-Turn Iterative Interactions",
+  "description": "EvoDev-Bench is a benchmark dataset for evaluating coding agents in persistent multi-turn software development. This release is forked from EvoCode-Bench and contains 26 executable Terminal-Bench-style tasks and 227 total rounds, with task-only and sanitized trajectory variants.",
+  "url": "https://huggingface.co/datasets/eachsheep/EvoDev-Bench",
   "license": "https://creativecommons.org/licenses/by-nc/4.0/",
   "conformsTo": [
     "http://mlcommons.org/croissant/1.1",
@@ -107,7 +107,7 @@
     "executable tests",
     "iterative development"
   ],
-  "citeAs": "Shen, Haiyang; Chen, Xuanzhong; Xu, Wendong; Ma, Yun; Chen, Liang; Li, Kuan. EvoCode-Bench: Evaluating Coding Agents in Multi-Turn Iterative Interactions. arXiv:2605.24110, 2026.",
+  "citeAs": "Shen, Haiyang; Chen, Xuanzhong; Xu, Wendong; Chen, Liang; Li, Kuan; Ma, Yun. EvoDev-Bench: Evaluating Coding Agents in Multi-Turn Iterative Interactions. arXiv:2605.24110, 2026.",
   "sameAs": [
     "https://arxiv.org/abs/2605.24110",
     "https://github.com/UniPat-AI/harbor_multiturn"
@@ -118,7 +118,7 @@
       "@id": "task_manifest",
       "name": "Task manifest",
       "description": "CSV manifest listing task identifiers, release variants, number of rounds, metadata fields, and relative task paths.",
-      "contentUrl": "https://huggingface.co/datasets/UnipatAI/EvoCodeBench/resolve/main/metadata/task_manifest.csv",
+      "contentUrl": "https://huggingface.co/datasets/eachsheep/EvoDev-Bench/resolve/main/metadata/task_manifest.csv",
       "encodingFormat": "text/csv",
       "contentSize": "12079",
       "sha256": "ef010b063f5a892611a9585c663195e382e8da165acf98798c214c2ca4fc75ff"
@@ -128,7 +128,7 @@
       "@id": "task_only_archive",
       "name": "Task-only release archive",
       "description": "Zstandard-compressed tar archive containing task directories without eval_runs; includes environment, round directories, instruction.md, and task.toml.",
-      "contentUrl": "https://huggingface.co/datasets/UnipatAI/EvoCodeBench/resolve/main/archives/evocodebench_wotraj.tar.zst",
+      "contentUrl": "https://huggingface.co/datasets/eachsheep/EvoDev-Bench/resolve/main/archives/evocodebench_wotraj.tar.zst",
       "encodingFormat": "application/zstd",
       "contentSize": "70967720",
       "sha256": "02e735e38ad3a77dfc1496808be97715166621fc67fb9b75ba70aa78a23c0142"
@@ -138,7 +138,7 @@
       "@id": "trajectory_archive",
       "name": "Sanitized trajectory release archive",
       "description": "Zstandard-compressed tar archive containing task directories with eval_runs. Docker image files and external local symlinks are replaced by markdown placeholders, and raw telemetry has been removed.",
-      "contentUrl": "https://huggingface.co/datasets/UnipatAI/EvoCodeBench/resolve/main/archives/evocodebench_wtraj.tar.zst",
+      "contentUrl": "https://huggingface.co/datasets/eachsheep/EvoDev-Bench/resolve/main/archives/evocodebench_wtraj.tar.zst",
       "encodingFormat": "application/zstd",
       "contentSize": "426694211",
       "sha256": "299f69943dd4669f88a3e1e70bd58480cd5ff644c7e917301b7c9c85756006af"
@@ -148,7 +148,7 @@
     {
       "@type": "cr:RecordSet",
       "@id": "tasks",
-      "name": "EvoCode-Bench task variants",
+      "name": "EvoDev-Bench task variants",
       "description": "One record per task and release variant.",
       "key": [
         {