Synthetic Data Warehouse for Combinatorial Bridge Tables

Reproducible synthetic dataset and evaluation code accompanying the article "Combinatorial Bridge Tables: Efficient Representation of Many-to-Many Relationships in Data Warehouse Systems" (A. Plamenac, S. Tomović, P. Dakić). No real, patient level, or otherwise non-public data are contained in, or were used to produce, this deposit. All data are fully synthetic and generated by the included seeded script (seed 20260708). The corpus reproduces the structure of a large inpatient data warehouse: 740,731 visits across four hospitals over 2013–2017, with three multivalued dimensions represented as bridge tables — diagnoses (n = 700 synthetic ICD-9-style codes), procedures (n = 300), and medications (n = 250). Group sizes follow a two-component (acute / multimorbid) mixture with a heavy right tail; item frequencies follow a Zipf law. All codes are placeholder identifiers with no clinical meaning. The deposit contains:- build_synthetic_case_study.py — seeded generator, combinatorial encoder, lossless-reconstruction validation, and storage accounting- cns.py — combinatorial number system (RankGroup / RankGroupInverse) with self-test- dimension tables, the encoded combinatorial fact table, and an illustrative classic-bridge sample- the realized group-size distribution (Table 1 of the article), result summary, and manuscript figures- etl_pipeline.puml — PlantUML source of the ETL pipeline figure Key results: losslessness verified by exact round-trip reconstruction on all 740,731 diagnosis groups with zero mismatches; bridge-table storage reduced 4.19× under fixed-width accounting and 7.06× under exact variable-length accounting. Running build_synthetic_case_study.py regenerates every reported number and figure exactly.