Beyond award snapshots: measuring public procurement obligations over the contract lifecycle

# Beyond award snapshots: replication materials ## Description This record contains the data and code needed to reproduce the tables and figures for “Beyond award snapshots: measuring public procurement obligations over the contract lifecycle.” The article reconstructs U.S. federal contract actions for 2010–2025 and distinguishes five ways of representing procurement over time: the contract-origin event, the initial obligated state, signed period flow, the cumulative obligated state, and the ordered contract path. The empirical results compare initial-event and lifecycle measures and describe contract trajectories, official modification purposes, observed dollar and end-date consequences, pricing form, initial duration, and competition. ## Replication routes The archive offers two one-command routes: 1. `bash run_from_clean_data.sh` reproduces the deposited article results from the included frozen inputs and verifies the generated tables and figures against `results/reference/`. 2. `bash run_from_web.sh` downloads current public-source records, cleans them, runs the same analysis, and writes a comparison with the frozen reference. The clean-data route writes to `results/reproduced_clean/`; the public-source route writes to `results/reproduced_web/`. Neither route overwrites the deposited reference results. Each route creates or reuses a local `.venv` and installs the exact package versions listed in `requirements.txt`. The cleaned inputs consist of the minimized 18-column action file `data/cleaned/contract_actions.parquet`, the quarterly deflator file `data/cleaned/gdpdef.csv`, and the ingestion summary `data/cleaned/source_counts.csv`. `DATA_DICTIONARY.md` defines every released variable, `OUTPUT_CROSSWALK.md` maps article results to files, and `QUALITY_CHECKS.md` documents the checks enforced by the programs. ## Public data sources and frozen vintage The contract data come from the USAspending transaction-download service for award type codes A, B, C, and D and action dates from 2010-01-01 through 2025-12-31. The frozen source extract was retrieved between 2026-03-08 and 2026-03-13. It consists of 253 download archives covering 192 calendar months; large months were divided into shorter intervals to satisfy API row limits. After 1,436 exact duplicate rows were removed, the analytical file contains 77,479,236 unique actions associated with 62,251,484 contracts. The real-dollar sensitivity uses the Gross Domestic Product: Implicit Price Deflator (`GDPDEF`), produced by the U.S. Bureau of Economic Analysis and distributed through FRED. The frozen CSV was retrieved 2026-07-14 at 13:21:58 UTC. The analysis uses all 64 quarters from 2010Q1 through 2025Q4 and expresses each action in 2025 dollars using its own action-quarter index and the 2025 annual average. USAspending and GDPDEF are live sources. Agencies can revise earlier contract records, and macroeconomic series can be revised. The included cleaned data are therefore the article-exact route. A later web run reproduces the collection and construction method on the records then available, but its values can legitimately differ from the frozen reference. ## Reproduced material The output covers Figures 1–3 and Tables I–III in the article, together with Supplementary Tables A1–A9. These exhibits report the data and sample bridge, initial-event versus action-flow measures, contract trajectory classes, common 12/24/36-month outcomes, modification-purpose and end-date results, supplier-identity sensitivity, pricing and duration comparisons, constant-dollar sensitivity, competition descriptives, and agency-clustered small-sample inference. ## Software and computing requirements The programs require Python 3.11 or newer and a Bash shell. Internet access is needed for initial dependency installation and for the public-source route. Both commands default to four processing workers and a 48 GB DuckDB memory limit. The cleaned-data route requires substantial temporary disk space; the public-source route additionally retains raw downloads under `work/web/` and can take many hours. The validation host uses Ubuntu 24.04.3 LTS (x86-64), Python 3.12.3, an Intel Xeon E-2276M processor (6 physical cores and 12 logical CPUs), and 62 GiB of RAM visible to the operating system. ## Data rights and license Original code and documentation are available under the MIT License. That license does not apply to or relicense the third-party source data. The minimized action file excludes supplier names, DUNS numbers, UEIs, and identity-bearing intermediate keys; its supplier linkage is a deterministic dataset-local integer surrogate. Detailed source-by-source notices are in `DATA_RIGHTS.md`. ## Source citations U.S. Department of the Treasury. (2026). *USAspending.gov contract transaction data* [Data set]. USAspending.gov. Contract action dates 2010-01-01 through 2025-12-31; retrieved 2026-03-08 through 2026-03-13. <https://www.usaspending.gov/>. U.S. Bureau of Economic Analysis. (2026). *Gross Domestic Product: Implicit Price Deflator* [GDPDEF] [Data set]. Retrieved 2026-07-14 from FRED, Federal Reserve Bank of St. Louis. <https://fred.stlouisfed.org/series/GDPDEF>.