Chulinz/Text2SQL-Decisions
Text2SQL-Decisions v0.3 28,081 English-only examples across two separate configurations. Both are template-generated, execution-validated drafts, not human-reviewed benchmarks. No model has been fine-tuned as part of this release. Configuration Rows Task License default 25,000 Four-candidate SQL plan selection on public sensor and e-commerce data CC BY 4.0 meter 3,081 Bounded meter planning decisions, including conversational context CC0-1.0 The configurations… See the full description on the dataset page: https://huggingface.co/datasets/Chulinz/Text2SQL-Decisions.
067
1# Replaying the synthetic fixture2 3`meters.jsonl` and `readings.jsonl` are exports of the actual PostgreSQL fixture used to validate the dataset. Timestamps use UTC. The evaluation clock is fixed at `2026-10-09T05:00:00.000Z` with `Asia/Bangkok` timezone. These original synthetic rows and generation files are CC0-1.0, as described in `../meter/LICENSE`.4 5`data.sql` defines the tables and cumulative-reading interval view. `data-fixture.ts`, `data-config.ts`, and `data-schema.ts` document the deterministic generator. The fixture and generator source fingerprints are recorded in `../meter_validation_report.json`. There are no production connections or credentials in this release.6 7To replay, use a new empty PostgreSQL database. The commands below assume the dataset repository is your working directory and `psql` is installed. `PGDATABASE` selects your disposable test database and normal PostgreSQL environment variables supply your own connection settings. The final two GRANT statements in `data.sql` refer to the application's `analyst` role; omit those grants if that role does not exist in your test database.8 9```sh10python3 -c 'from pathlib import Path; print("\n".join(line for line in Path("meter_fixture/data.sql").read_text().splitlines() if not line.startswith("GRANT ")))' | psql --set=ON_ERROR_STOP=111psql --set=ON_ERROR_STOP=1 --file=meter_fixture/replay.sql12```13 14The replay inserts all rows in a single transaction, so an interrupted or failed replay rolls back. It expects empty fixture tables and fails on duplicate primary keys. Rerun after an interrupted transaction; after a successful replay, reuse that database rather than inserting again.15 16Example gold query records contain PostgreSQL SQL plus positional parameters. Bind the `parameters` values to the SQL rather than interpolating them. Result hashes include the runtime evidence structure and cannot be recreated by merely concatenating SQL result files; evaluate the individual rows and aggregates when using another client.17