Procurement Concentration and Potentially Recoverable Competitive Savings in New York City Contracting — Audit and Provenance Ledger and Reproduction Dataset (v1.0)

This dataset is the audit and provenance ledger plus the reproduction and audit package for the paper "Procurement Concentration and Potentially Recoverable Competitive Savings in New York City Contracting" (v1.0; DOI 10.5281/zenodo.21447217). Contents: a standard-library reproduction script (reproduce_procurement.py) that verifies the SHA-256 of two frozen vendor CSVs against a manifest, recomputes Sole Source and Negotiated Acquisition concentration (vendor counts, top-N shares, Gini) on the clean vendor rows, checks all five Table 3 savings bases, and with a live mode re-pulls the MTA FY2024 cross-check from data.ny.gov; the two hashed vendor CSVs and the manifest; and the complete audit trail (audit-reports/) with every Model Council member report and synthesis across the three rounds. A self-test mode perturbs each checked value and confirms the comparator detects it, and an adversarial trap in the source data catches any reproduction that fails to exclude the summary and footnote rows. The paper converged under Audit Framework v3.0, Standing Rule 7. F-PC-R1 (on draft v0.3) returned non-pass on one verified HIGH finding (a false quantitative assurance in the basis-disclosure note); its findings were absorbed. F-PC-R2 (on v0.4) and F-PC-R3 (on v0.5) were qualifying passes by independent, non-producer-lineage members with zero critical and zero high findings, advancing the consecutive qualifying-pass count to two and closing the rotation with a member new to the second pass. The producing lineage (Claude, Anthropic) was quarantined from every convergence-bearing verdict. The ledger records the six-element findings and dispositions for all three rounds, the verified figures, the tier discipline (byte-deterministic reproduction, disclosed method, and a refused Tier-3 legitimacy boundary), and a producer-accountability note documenting the errors introduced during revision that the audit caught. Two caveats are recorded rather than concealed: the two consecutive passes rest on materially-identical substance (the audited versions differ only in wording and tooling, no figure moved), and convergence is met inside this producer-defined governance framework while the externally-verifiable core (every figure reproducing from the public Comptroller and data.ny.gov sources, and the comparator being tamper-evident under the adversarial trap) holds independently of it. Convergence certifies structural honesty and reproducibility; it does not certify that the savings estimate is recoverable or policy-ready. Core (methodology and this program): structural honesty verification cross-instance audit reproducibility audit trail provenance ledger convergence Model Council tamper-evidence SHA-256 Gini coefficient vendor concentration Domain (what the data is about): public procurement NYC Comptroller non-competitive contracting sole source negotiated acquisition open data