Test results

SQL Query Optimizer for D1, SQLite and Turso: test results

Tested 2026-10-03, skill version 1.0.2 at the time of loading this page. We run every skill on a strong and a weak model before it is listed, and publish both verdicts, including where the weak one fails.

Verdicts

Date
2026-10-03
Strong · claude-sonnet-5-5 (Claude Code alias "sonnet")
Named the planted problem with the right code in all 11 cases that had one and answered 'No findings.' for the clean schema. Its fixes were working SQL: an index on (scope, created_at) for the gap, a partial index on status = 'pending' for the empty poll, a range on the raw timestamp, keyset pagination, chunks of 50 for D1's 100-parameter limit, a backfill in the same migration, the partial index's WHERE copied into ON CONFLICT. At first 2 of 12 were scored as failures because our pattern accepted only one of the two fixes the skill names; Sonnet chose the other (a latest-prices table kept by the writer; day bounds computed in SQL), both were confirmed with EXPLAIN QUERY PLAN, and the same answers rescored at 12 of 12.
Weak · claude-haiku-4-5-20251001 (Claude Code alias "haiku")
Found the planted problem with the right code and a working fix in all 11 cases that had one, including the Bulgarian request. On the clean schema it invented a finding under a code of its own and proposed DELETE ... LIMIT, which standard SQLite rejects as a syntax error. With Haiku, run any suggested statement before relying on it.

With and without the skill

Tested 2026-10-03.

Results with and without the skill, for Sonnet and Haiku
SonnetHaiku
withwithoutwithwithout
Cases passed12/1210/1211/123/12

One run per model and case. Without the skill Sonnet missed an OR that switches the index off and a function wrapped around an indexed column. Haiku missed nine, among them the D1 limit of 100 bound parameters, a LIKE prefix that scans, and a new flag column that turns the whole history into pending work. On the clean schema Haiku failed with and without the skill: with it, it invented a finding.

Same cases and the same checks with and without the skill. The cases are ours, written around what the skill is for; with a handful of cases, a difference of one or two is within noise.

Note

Twelve fictional schemas and queries: eleven with one planted problem each (aggregate scan, index gap, NOT IN, OR on a flag, empty queue poll, function on a column, prefix LIKE, OFFSET with COUNT, D1 parameter limit, flag column without backfill, ON CONFLICT against a partial index) and one clean; one request in Bulgarian. Every planted plan and every accepted fix was confirmed with EXPLAIN QUERY PLAN on SQLite 3.53 without ANALYZE. Machine checks look for the finding code and a working fix; one run per model and case. The rescoring of stored answers after two patterns were widened is in results-2026-10-03-recheck.md.

What was not measured

  • Models other than the two named above were not run.
  • Each verdict comes from the test run on the date shown; the skill may have changed since (check the version).
  • Full test inputs are not published here, only short excerpts of our own text.
  • Results on your own texts, languages and domains can differ.

Back to SQL Query Optimizer for D1, SQLite and Turso · Card (JSON)