Atomic Common-Day Invoice Clearing: Pseudonymized Invoice Records and Reproducibility Data, 2012–2023

Description

This dataset contains 749,952 retained atomic invoice records issued from 2012 through 2023 (EUR 99.7050155695 billion) and 1,240 observed 2024 records through 8 February 2024 used only as the terminal execution bridge for the 2023 cohort. Direct business identifiers have been removed. Each record is represented by repository-specific pseudonymous invoice, debtor, and creditor codes, integer euro cents, issue date, due date, and issue cohort. No firm names, VAT/CIF identifiers, invoice numbers, source-row identifiers, source filenames, or reversible crosswalks are included. The deposit also contains annual runner inputs, curation audits, bridge non-reuse checks, phase-replay validation summaries, annual results, and daily cumulative payable-mass-reduction curves. The 2021 issue cohort uses the separately curated twelve-month 2021 workbook corpus rather than the longitudinal source's 2021 slice. Pseudonymous identifiers preserve deterministic tie-break ordering required by the published implementation. Because exact dates, amounts, and network relations are retained, the transaction-level records are pseudonymized rather than guaranteed irreversibly anonymous; open publication should therefore be used only where the governing data-use agreement permits it. Source of data IMI, The Ministry of Economics, Rumania Related article Atomic Common-Day Invoice Clearing under Causal Daily Scheduling: Path-Enabled and Bounded-Cycle Policies Keywords temporal invoice graph; multilateral clearing; trade credit; working capital; operations research; causal scheduling; supply networks; payable-mass reduction; reproducibility; decentralized clearing

Steps to reproduce

Download mendeley_annual_runner_inputs_anonymized.zip and extract it to data/annual_inputs/. It should contain one normalized atomic-record CSV for each year, 2012.csv through 2024.csv. Download or clone the companion software repository linked under Related Links. Python 3.11 or newer is required. From the repository root, create a Python environment and install the package and reproducibility dependencies: python -m pip install -e '.[dev,optimization]' Run the automated test suite: pytest -q The 17 tests cover common-day feasibility, path/cycle accounting, causal scheduling, optimization, attribution, replay, topology, input/output handling, and bridge non-reuse. Run the rolling causal daily clearing experiment: python scripts/run_sequential_cohorts.py \ --data-dir data/annual_inputs \ --output build/sequential \ --start-year 2012 \ --end-year 2023 \ --last-bridge-end 2024-02-08 \ --cycle-bound 8 Each policy maintains one chronological residual-state stream. January-February invoices of year (y+1) may enter cohort (y)'s terminal bridge, but each invoice UID is introduced only once. Amounts consumed in a bridge cannot be reused in the next year; only residual balances continue. Cohorts 2012-2022 use the complete following January-February bridge. The 2023 cohort uses the available follow-up through 8 February 2024, reproducing the observed bridge horizon. Compare outputs with mendeley_reproducibility_results_and_audits.zip, especially: results/annual_sequential_bridge_results.csv results/aggregate_sequential_bridge_summary.json results/boundary_nonreuse_audit.csv results/cycle_phase_validations.json results/path_phase_validations.json results/curves/annual_*_sequential_bridge_curves.csv Aggregate results should be approximately: bounded-cycle clearing (L=8): 43.3468% payable-mass reduction; path-enabled clearing: 48.2018% payable-mass reduction; path advantage: 4.8551 percentage points. This corresponds to approximately EUR 43.219 billion versus EUR 48.060 billion of PMR over EUR 99.705 billion of issue-cohort invoice mass. In boundary_nonreuse_audit.csv, reintroduced_at_full_face_value should be False at every annual boundary. Global validation should confirm that each invoice UID is introduced once and no record is consumed above its original amount. Compare generated daily cumulative PMR curves and annual terminal results with the deposited reference outputs. Runtime may vary by hardware, but deterministic clearing and accounting results should reproduce independently of runtime. Invoice and firm identifiers are repository-specific pseudonyms. No company names, VAT/CIF identifiers, invoice numbers, source-file identifiers, or reversible mapping are needed to reproduce the published experiments. Code in https://github.com/peplluis7/onemillioninvoices

Institutions

• Universitat de Girona

  Catalonia, Girona

Categories

Related Links

Licence