UB_HF_DATASET_VALIDATION.md

UndoBench Hugging Face Dataset Validation (UB-HF1)

HF_OWNER = TanmaySah HF_DATASET = undobench HF_REPO = TanmaySah/undobench

TOTAL_TASKS = 36 DEV = 14 VALIDATION = 10 TEST = 12 DOMAINS = 8

MISSING_TASK_IDS = 0 EXTRA_TASK_IDS = 0 DUPLICATE_TASK_IDS = 0

FOUR_AUTHORS_LISTED = YES CITATION_HAS_4_AUTHORS = YES CITATION_AUTHOR_ORDER_MATCHES_ARXIV = YES

DATASET_VIEWER_WORKS = YES LOAD_DATASET_WORKS = YES

PAPER_LINK_PRESENT = YES HF_PAPER_LINK_PRESENT = YES GITHUB_LINK_PRESENT = YES

LICENSE = CC-BY-4.0 SECURITY_SCAN_PASS = YES

CANONICAL_TASKS_CHANGED = NO SCIENTIFIC_RESULTS_CHANGED = NO

---

Detailed Audit & Verification Log

1. Repository & Publication Details

• Hugging Face Repository: TanmaySah/undobench
• Public URL: https://huggingface.co/datasets/TanmaySah/undobench
• Visibility: Public (private: false)
• License: CC BY 4.0 (license: cc-by-4.0)
• Paper Association: arxiv:2610.05622 (papers: ["2610.05622"])
• Code Link: https://github.com/tradertanmay/undobench

2. Task & Split Breakdown

• Split DEV: 14 workflows (data/dev.parquet)
• Split VALIDATION: 10 workflows (data/validation.parquet)
• Split TEST: 12 workflows (data/test.parquet)
• Total Workflows: 36 canonical workflows across 8 enterprise operational domains:
  1. Cloud Infrastructure (4 tasks)
  2. Customer Relationship Management (CRM) (4 tasks)
  3. Database Transactions (5 tasks)
  4. Git Version Control (4 tasks)
  5. Messaging Systems (5 tasks)
  6. Payments & Billing (5 tasks)
  7. Storage & Blob Management (5 tasks)
  8. Ticketing & Incident Management (4 tasks)

3. Canonical Authorship

• Hosting Namespace: TanmaySah (Personal Hugging Face account used for dataset discovery and hosting)
• Canonical Authors:
  1. Dolly Sah
  2. Tanmay Sah
  3. Harshul Jain
  4. Tanya Sah
• Order Verification: Exactly matches the public arXiv record arXiv:2610.05622.

4. Client Verification (datasets.load_dataset)

from datasets import load_dataset

ds = load_dataset("TanmaySah/undobench")
# Result:
# DatasetDict({
#     dev: Dataset({features: [...], num_rows: 14}),
#     validation: Dataset({features: [...], num_rows: 10}),
#     test: Dataset({features: [...], num_rows: 12})
# })

5. Security & Secret Scan

• Hugging Face tokens: 0 detected
• GitHub tokens: 0 detected
• OpenAI/Anthropic/API keys: 0 detected
• Private keys: 0 detected
• Local filesystem paths (/Users/): 0 detected
• Status: PASSED