A released fork of GizmoSQL v1.38.0 (a server embedding DuckDB) whose embedded DuckDB v1.5.5 carries a small series of engine changes:

• GizmoSQL: https://github.com/bddppqs/intent-gizmosql, tag v1.38.0-clickbench.1 (the release this entry installs: portable Linux amd64 and arm64 builds).
• DuckDB: https://github.com/bddppqs/intent-duckdb, tag v1.5.5-clickbench.1 (changes documented in that repository's CLICKBENCH-FORK.md: batched RE2 character-class runs, CountZeros builtins with selection-fed HyperLogLog, an evictable decoded-dictionary cache with admission back-off, and a cross-request memo of literal regexp_replace domains over dictionary vectors).

The scripts in this directory are the upstream gizmosql entry's, unchanged, except:

• install, which downloads the pinned release zip for the machine's architecture (amd64 or arm64) and verifies its SHA-256 instead of running the vendor's network installer;
• query, whose failure check only looks at the client's own diagnostics (exit code, Error: lines) and no longer at result rows: a URL or title containing the word "error" in a result window over tied counts (e.g. Q39's LIMIT 10 OFFSET 1000) turned a valid run into a null timing in the upstream version.

The server runs with its default configuration, the upstream schema (create.sql) and the upstream load path; there is no extra index, pre-aggregation or query-specific configuration.

Disclosure: the DuckDB changes were developed and evaluated against the 43 ClickBench queries on the ClickBench dataset. They are general engine mechanisms that do not change query results, the storage format or the SQL surface. Because of that development history the entry is labelled tuned: yes at the maintainers' request; the ClickBench maintainers decide its final name and labels.

Both repositories keep their upstream licenses (DuckDB: MIT; GizmoSQL: Apache-2.0).