L2CA reproducibility archive (v1.1.1)

Reproducibility archive (v1.1.1) for the manuscript “Ledger-to-Cell Corrective Allocation for Open-Boundary Urban-Flood Surrogates: Signed Volume Diagnosis and Sequential Spatial Correction in Lower Naengcheon, South Korea,” Reproducibility archive (v1.1.1) supporting the manuscript “L2CA: Signed-volume correction with boundary-exchange updates for flood surrogates”. The deposited release files remain unchanged. Figure and table numbers in the archive and in the Contents and Licensing sections below follow the original archive numbering. Supplementary Software S1, provided separately with the revised manuscript, contains the numbering crosswalk and the synthetic CPU benchmark. Supplementary Software S1 is not included in this Zenodo deposit. L2CA (Ledger-to-Cell Corrective Allocation) is an inference-time correction framework for urban-flood surrogate models in open-boundary settings. It reconstructs signed net open-boundary exchange from predicted depth–velocity fields, diagnoses time-varying signed correction volumes from a source–storage–boundary water balance, allocates them spatially with a lightweight learned allocator, and propagates the corrected state sequentially through a two-pass relaxed projection. Contents: - `code/l2ca_core/`: L2CA projection operator and a synthetic CPU smoke test that requires no study data- `code/topopluvnet_definition/`: weight-free TopoPluvNet architecture definition and a script that verifies its parameter count and output dimensions (7,106,470 parameters)- `aggregate_tables/` and `analysis_outputs/`: deidentified event-level and aggregate tables supporting standalone replay of the 12 headline contrasts, including the full 10,000-replicate bootstrap reconstruction with design hashes; Holm-adjusted allocation contrasts; and analyses of closure targets, ledger components, regional regimes, relaxation factors, and equal requested correction volumes- `figure_reproduction/`: releasable plotting data, rendering scripts, and reference outputs for Figures 2–4, 6–7, S1–S2, and S4. Figures 1, 5, and S3 are data-owner-approved rendered spatial derivatives Integrity and verification: `PACKAGE_MANIFEST.tsv` and `SHA256SUMS.txt` record and verify all 189 payload files. Including these two integrity files, the archive contains 191 file entries in total. `VERIFICATION_RESULTS.txt` documents the verification environment—Python 3.12.3, PyTorch 2.5.1 (CPU-only), NumPy 2.4.6, pandas 3.0.3, and Matplotlib 3.10.9—and the checks executed for this release. Scope: The archive supports inspection and execution of the released L2CA projection operator, inspection of the TopoPluvNet architecture, replay of the reported statistical contrasts, and regeneration of the releasable figures. It does not support rerunning the original SWMM/FVM simulations, model training, or inference because it excludes the original SWMM/FVM inputs and outputs, drainage network, machine-readable boundary geometry, internal node and scenario identifiers, cell-level arrays, DEM-derived inputs, trained weights, and data-split lists. These materials remain restricted under the data owner’s security and redistribution policies. Restricted materials may be requested from the corresponding author ([e-mail]), subject to written approval from the data owner. Licensing: Code is released under the MIT License (`LICENSE`). Author-derived deidentified nonspatial tables and plotting data are released under CC BY 4.0 (`DATA_LICENSE.txt`). Figures 1 and 5 include OpenStreetMap data © OpenStreetMap contributors, licensed under ODbL 1.0. The Natural Earth locator input is in the public domain. Funding: Electronics and Telecommunications Research Institute (ETRI) grant funded by the Korean government [26ZR1300, Development of Technology for the Urban Extreme Rainfall Response Platform].