DepLab Dataset v1.0.0: Empirical Python Dependency Compatibility Experiments

24,922 real-world experiments testing whether pairs of Python packages can be installed and used together. Each experiment records the full outcome across three stages: dependency resolution, import, and a runtime smoke test. This dataset was built to answer a simple question: how often does an environment that resolves successfully still fail in practice? In the development split, 3,049 environments (14.2%) passed resolution but failed at import or smoke-test time. Resolvers cannot see these failures because they only read published metadata and never install or run anything. The dataset powers DepLab (https://github.com/Abhisek12378/DepLab), an ML-guided tool that predicts working dependency combinations. Coverage: 50 popular Python packages, Python 3.8 through 3.14, Linux x86_64, multiple versions per package across several dependency families. Each record includes the exact commands run, exit codes, captured stdout/stderr, normalized error text, wheel metadata with compatibility decisions, a PEP 751 lockfile hash, and the exact runtime environment (Python, uv, kernel, libc). Development split (21,490 experiments): 12,031 pass, 6,410 resolution failures, 2,448 import failures, 601 smoke-test failures. Sealed validation split (3,432 experiments): 639 pass, 2,714 resolution failures, 60 import failures, 19 smoke-test failures. The validation split was kept sealed during model development. Files: full experiment results for both splits (JSONL) and 142 model-ready engineered features for the development split (CSV), covering version, constraint, wheel, and changelog-mined signals.