My open-source text-to-SQL layer is #1 on BIRD Mini-Dev in all three dialects

The maintainer of schemagate, an Apache-2.0 text-to-SQL library, reported that it now ranks first on the BIRD Mini-Dev leaderboard in all three dialects, scoring 81.0 on SQLite, 72.6 on PostgreSQL and 72.4 on MySQL, ahead of the next entries at 75.6, 69.0 and 49.2. The approach selects only the tables a question needs, passes those plus BIRD's column notes to Claude Opus 5.5 to write a single query, and retries with the error message if the query fails, with no candidate voting or agent loop. The maintainer noted that schemagate's core purpose is access control, stripping tables the caller may not read before the model sees the schema, and that a first result on the hidden main BIRD board was 72.11 with a rerun pending.

BIRD Mini-Dev is 500 text-to-SQL questions over 11 databases, in SQLite, PostgreSQL and MySQL versions, scored by whether the query returns the right rows. BIRD keeps the leaderboard and checks the submissions. schemagate, the Apache-2.0 library I maintain, is now first in all three columns: | | schemagate | Next entry | |---|---|---| | SQLite | 81.0 | 75.6 | | PostgreSQL | 72.6 | 69.0 | | MySQL | 72.4 | 49.2 | The setup is plain: schemagate picks the tables a question needs, the prompt carries only those plus BIRD's column notes, and Claude Opus 5.5 writes one query. No candidate voting and no agent loop. If the query errors, the model sees the error and tries again. schemagate's main job is access control. It removes the tables the caller may not read before the model sees the schema. A benchmark doesn't test that part, but it shows the filtering doesn't cost accuracy. What it isn't: the main BIRD board, which uses a hidden test set. Our first test result there was 72.11, and a rerun is pending. pip install schemagate