DMAIC-Bench: From Self-Reflection to Self-Restraint for Agents in Multi-Stage Analytical Tasks
Paper • 2610.32305 • Published
task_id stringlengths 19 19 | case_id stringclasses 10
values | industry stringclasses 3
values | seed int64 0 7 | is_fault bool 2
classes | instruction stringlengths 12.8k 13.2k | data_blocks_json stringlengths 637 5.17k |
|---|---|---|---|---|---|---|
adb_c27-T3-s0-clean | adb_c27 | process_industry | 0 | false | # Task adb_c27-T3-s0-clean
## 1. Business background
We are the process engineering group at a fine-chemicals plant. Yield on one intermediate has been low and swings from batch to batch; the internally accepted range is 80 to 100 percent. Production has recorded every batch's yield in charge order, and the archive h... | {"define": {"goal": "bring batch yield inside its accepted range and hold it there", "lsl": 80.0, "usl": 100.0}, "msa": {"measurements": {"operator1": [[91.0024, 91.1981, 90.7567], [88.1826, 88.8958, 88.4649], [89.2513, 89.1766, 89.3551], [86.0227, 85.5796, 86.3415], [89.5819, 89.3479, 89.4342], [90.1051, 90.0193, 89.9... |
adb_c27-T3-s1-clean | adb_c27 | process_industry | 1 | false | # Task adb_c27-T3-s1-clean
## 1. Business background
We are the process engineering group at a fine-chemicals plant. Yield on one intermediate has been low and swings from batch to batch; the internally accepted range is 80 to 100 percent. Production has recorded every batch's yield in charge order, and the archive h... | {"define": {"goal": "bring batch yield inside its accepted range and hold it there", "lsl": 80.0, "usl": 100.0}, "msa": {"measurements": {"operator1": [[89.9813, 90.368, 91.1385], [87.943, 87.9054, 88.0488], [88.7021, 87.7911, 87.6965], [90.7042, 91.1402, 91.0279], [92.8288, 93.185, 92.8427], [91.9278, 91.9816, 92.2117... |
adb_c27-T3-s2-clean | adb_c27 | process_industry | 2 | false | # Task adb_c27-T3-s2-clean
## 1. Business background
We are the process engineering group at a fine-chemicals plant. Yield on one intermediate has been low and swings from batch to batch; the internally accepted range is 80 to 100 percent. Production has recorded every batch's yield in charge order, and the archive h... | {"define": {"goal": "bring batch yield inside its accepted range and hold it there", "lsl": 80.0, "usl": 100.0}, "msa": {"measurements": {"operator1": [[89.52, 89.2849, 89.4704], [90.8251, 90.3887, 90.8674], [91.3032, 90.9195, 91.4685], [90.7227, 90.1485, 90.5083], [92.1097, 92.2492, 92.4403], [90.7749, 90.3569, 90.463... |
adb_c27-T3-s3-clean | adb_c27 | process_industry | 3 | false | # Task adb_c27-T3-s3-clean
## 1. Business background
We are the process engineering group at a fine-chemicals plant. Yield on one intermediate has been low and swings from batch to batch; the internally accepted range is 80 to 100 percent. Production has recorded every batch's yield in charge order, and the archive h... | {"define": {"goal": "bring batch yield inside its accepted range and hold it there", "lsl": 80.0, "usl": 100.0}, "msa": {"measurements": {"operator1": [[89.3964, 89.3599, 88.9562], [90.491, 90.3328, 89.8476], [90.0828, 90.1918, 89.898], [87.9235, 88.0822, 87.8918], [91.4379, 91.6117, 91.0022], [92.6886, 92.2912, 92.634... |
adb_c27-T3-s4-clean | adb_c27 | process_industry | 4 | false | # Task adb_c27-T3-s4-clean
## 1. Business background
We are the process engineering group at a fine-chemicals plant. Yield on one intermediate has been low and swings from batch to batch; the internally accepted range is 80 to 100 percent. Production has recorded every batch's yield in charge order, and the archive h... | {"define": {"goal": "bring batch yield inside its accepted range and hold it there", "lsl": 80.0, "usl": 100.0}, "msa": {"measurements": {"operator1": [[85.9327, 85.7052, 84.2894], [93.7602, 89.7024, 91.5639], [91.2823, 92.4553, 89.3874], [90.0308, 89.6041, 87.6001], [83.9054, 85.7962, 88.8155], [91.9093, 94.6481, 94.7... |
adb_c27-T3-s5-clean | adb_c27 | process_industry | 5 | false | # Task adb_c27-T3-s5-clean
## 1. Business background
We are the process engineering group at a fine-chemicals plant. Yield on one intermediate has been low and swings from batch to batch; the internally accepted range is 80 to 100 percent. Production has recorded every batch's yield in charge order, and the archive h... | {"define": {"goal": "bring batch yield inside its accepted range and hold it there", "lsl": 80.0, "usl": 100.0}, "msa": {"measurements": {"operator1": [[87.9291, 87.8957, 88.008], [85.6997, 86.0302, 85.6002], [91.3676, 91.5483, 91.9109], [90.7877, 90.8471, 90.7164], [86.005, 85.8592, 85.8926], [87.7278, 88.2022, 87.431... |
adb_c29-T3-s0-clean | adb_c29 | discrete_manufacturing | 0 | false | # Task adb_c29-T3-s0-clean
## 1. Business background
We are the quality group at a components plant. One critical characteristic has an internally accepted range of 40 to 100, and the line has left a record of sixty consecutive pieces in the order they came off. The customer has asked us to state where this character... | {"define": {"goal": "state where the characteristic stands against its accepted range", "lsl": 40.0, "usl": 100.0}, "process_run": {"data": [66.266, 58.163, 59.975, 53.496, 60.483, 62.874, 65.453, 61.539, 75.187, 63.684, 66.003, 70.171, 70.304, 57.429, 60.645, 54.521, 54.84, 67.938, 58.447, 59.381, 58.173, 61.298, 64.1... |
adb_c29-T3-s1-clean | adb_c29 | discrete_manufacturing | 1 | false | # Task adb_c29-T3-s1-clean
## 1. Business background
We are the quality group at a components plant. One critical characteristic has an internally accepted range of 40 to 100, and the line has left a record of sixty consecutive pieces in the order they came off. The customer has asked us to state where this character... | {"define": {"goal": "state where the characteristic stands against its accepted range", "lsl": 40.0, "usl": 100.0}, "process_run": {"data": [54.952, 78.187, 62.369, 76.752, 84.967, 69.035, 71.812, 69.541, 64.681, 72.903, 55.262, 63.18, 66.448, 73.214, 51.398, 74.547, 75.242, 72.791, 61.316, 84.919, 62.237, 74.036, 74.0... |
adb_c29-T3-s2-clean | adb_c29 | discrete_manufacturing | 2 | false | # Task adb_c29-T3-s2-clean
## 1. Business background
We are the quality group at a components plant. One critical characteristic has an internally accepted range of 40 to 100, and the line has left a record of sixty consecutive pieces in the order they came off. The customer has asked us to state where this character... | {"define": {"goal": "state where the characteristic stands against its accepted range", "lsl": 40.0, "usl": 100.0}, "process_run": {"data": [64.009, 60.881, 58.327, 59.419, 68.776, 58.003, 57.319, 65.932, 58.622, 58.686, 59.456, 64.455, 61.792, 59.419, 65.646, 58.972, 67.72, 60.12, 58.888, 58.271, 62.497, 60.622, 66.37... |
adb_c29-T3-s3-clean | adb_c29 | discrete_manufacturing | 3 | false | # Task adb_c29-T3-s3-clean
## 1. Business background
We are the quality group at a components plant. One critical characteristic has an internally accepted range of 40 to 100, and the line has left a record of sixty consecutive pieces in the order they came off. The customer has asked us to state where this character... | {"define": {"goal": "state where the characteristic stands against its accepted range", "lsl": 40.0, "usl": 100.0}, "process_run": {"data": [68.615, 60.736, 72.303, 69.2, 61.408, 74.745, 53.975, 71.322, 70.492, 64.61, 96.4, 70.599, 82.64, 43.6, 91.594, 84.114, 57.342, 76.739, 55.345, 75.069, 87.835, 68.445, 48.06, 66.8... |
adb_c29-T3-s4-clean | adb_c29 | discrete_manufacturing | 4 | false | # Task adb_c29-T3-s4-clean
## 1. Business background
We are the quality group at a components plant. One critical characteristic has an internally accepted range of 40 to 100, and the line has left a record of sixty consecutive pieces in the order they came off. The customer has asked us to state where this character... | {"define": {"goal": "state where the characteristic stands against its accepted range", "lsl": 40.0, "usl": 100.0}, "process_run": {"data": [65.702, 62.387, 60.174, 78.151, 78.817, 66.224, 81.525, 78.538, 73.22, 75.824, 71.315, 57.902, 72.863, 69.323, 70.732, 75.411, 71.107, 75.798, 73.523, 83.37, 65.979, 89.2, 66.501,... |
adb_c30-T3-s0-clean | adb_c30 | discrete_manufacturing | 0 | false | # Task adb_c30-T3-s0-clean
## 1. Business background
We are the process team on a honing line that finishes cylinder bores. Assembly has returned units for bore finish, and the accepted band for roughness is 0.40 to 1.60 um Ra with 1.00 as the target. The line records a roughness reading for every bore in the order i... | {"define": {"goal": "bore finish inside the accepted band and holding steady", "lsl": 0.4, "unit": "um Ra", "usl": 1.6}, "msa": {"measurements": {"operator1": [[1.255, 1.2124, 1.1209], [0.2024, 0.212, 0.3026], [1.1814, 1.2104, 1.155], [1.1119, 1.1624, 1.2237], [1.1228, 1.1046, 1.1333], [0.8936, 0.848, 0.7852], [1.1742,... |
adb_c30-T3-s1-clean | adb_c30 | discrete_manufacturing | 1 | false | # Task adb_c30-T3-s1-clean
## 1. Business background
We are the process team on a honing line that finishes cylinder bores. Assembly has returned units for bore finish, and the accepted band for roughness is 0.40 to 1.60 um Ra with 1.00 as the target. The line records a roughness reading for every bore in the order i... | {"define": {"goal": "bore finish inside the accepted band and holding steady", "lsl": 0.4, "unit": "um Ra", "usl": 1.6}, "msa": {"measurements": {"operator1": [[0.9242, 0.8629, 0.9333], [1.3477, 1.3823, 1.3887], [0.1998, 0.1758, 0.2318], [1.0161, 1.0214, 1.0308], [1.0136, 1.0323, 1.1009], [0.9889, 0.9907, 1.0895], [0.6... |
End of preview. Expand in Data Studio
DMAIC-Bench
Paper: arXiv:2610.32305 · Code and evaluation framework
DMAIC-Bench evaluates whether agents use intermediate evidence to continue, divert, or stop in multi-stage analytical workflows. The paper describes 97 seeded tasks across 15 industrial cases and a 16-task life-sciences extension.
Scope and usage
The benchmark is an interactive evaluation: generated data is released in stages by a steward, and a separate verifier evaluates execution. Browsing or loading dataset rows alone does not reproduce the full agent evaluation. Refer to the GitHub repository for task generation, Harbor integration, scoring, dependencies, and instructions.
This public export contains 78 records: 65 clean tasks across 10 public cases and 13 fault variants, generated using the public-only build of the GitHub release. Each JSONL row includes the task ID, case, industry, seed, fault flag, agent instruction, and synthetic data blocks as a JSON string. All blocks are visible in this export; access restrictions and the steward are not enforced by this dataset viewer. Reference solutions, verifier truth and runtime code are not included here. See the GitHub repository to construct and run the full environment. The life-sciences extension is described in the paper but is not included in this export. The original industrial campaign per-trial archive is not included in the public GitHub release. Reported experiments covered all 97 industrial tasks before publication; do not describe those results as a held-out generalization evaluation.
License
The generated benchmark corpus is distributed under CC BY 4.0. Software in the GitHub repository is separately licensed under AGPL-3.0-only. The paper is separately licensed under CC BY 4.0.
Citation
@misc{peng2026dmaicbench,
title={DMAIC-Bench: From Self-Reflection to Self-Restraint for Agents in Multi-Stage Analytical Tasks},
author={Peng, Shan and Cheng, Shuaiqi and Peng, Miao and Zi, Chenyi and Li, Jia},
year={2026},
eprint={2610.32305},
archivePrefix={arXiv},
primaryClass={cs.AI},
url={https://arxiv.org/abs/2610.32305}
}
- Downloads last month
- 42