EstateForge

EstateForge generates the kind of corporate database estate real tools meet: redundant tables, historical naming, incompatible keys, filtered copies, undocumented migrations, and dependencies nobody meant to make permanent. The disorder is planted on purpose, and its answer key is written down.

• 500tables in the largest generated estate
• 19questions with independently recorded answers
• 2×2business domains and SQL dialects: PostgreSQL and SQL Server

A benchmark that knows why the answer is right

A single plan generates tables, views, procedures, ETL flows, rows, and an answer key. The generator decides where a field came from, renders that decision as SQL, and records the truth separately. A discovery tool parses the SQL back in the other direction. Agreement between independently written paths is evidence; a golden generated from either path alone is only a transcript.

Named difficulties, not a generic mess dial

Different pathologies defeat different shortcuts: denormalized copies break value-overlap inference, zero-padded text keys break type filters, and shared vocabulary breaks name similarity. EstateForge names each difficulty so a run can say exactly what it tested. A pristine control estate distinguishes finding real disorder from reporting disorder everywhere.

What it generates

• Energy-trading and US core-banking domains.
• Nested estate sizes of 25, 116, or 500 tables, so an object keeps the same meaning as scale grows.
• PostgreSQL and SQL Server renderings, both gated.
• A JSON fixture with objects, columns, sample values, derivations, questions, and the planted answer key.

Explore it

git clone https://github.com/Solifugus/estateforge.git
cd estateforge
./estateforge list
./estateforge questions
./estateforge test