Artifact for: When Do Tabular Foundation Models Pay Off for Cardinality Estimation?

Reproducibility artifact (v1 snapshot) for an ICDE 2027 Experiment, Analysis, and Benchmark submission. Contains the frozen evaluation protocol and manifests, analysis and audit code, per-query inputs, labels and predictions needed to recompute every table and figure, the post-hoc diagnostics (sampling-reliability regimes, review checks, call fractions, oracle/PRICE diagnostics where complete), runtime analysis summaries, and the manuscript sources. Third-party code, datasets and model weights are not redistributed; fetch scripts and SHA-256 hashes are provided. Not redistributed: PRICE and STATS-CEB upstream files (decision D5); reproduce/fetch_d5_upstream.py fetches them from the public upstream repositories at pinned commits and verifies SHA-256. Our own PRICE run outputs that the frozen records bind (statistics pickles, feature outputs, run records) are included. Held: the remaining PRICE-derived results (decision D5) and the curated decision records except the published E2 amendment-2 erratum (decision D9); raw runtime service records are excluded (decision D4). See HELD.md, EXTERNAL_ASSETS.tsv and EXCLUSIONS.md in the archive. Licensing: code under MIT; documentation and our derived data under CC BY 4.0 where upstream terms allow (see LICENSE-DATA.md); TabPFN 3.5 outputs under the TabPFN-3.5 Non-Commercial License notice (NOTICE-TABPFN.txt). Verify downloads with: sha256sum -c --ignore-missing SHA256SUMS