July 11, 2026
SUM Like Nobody’s Watching
Show HN: Sqlsure – deterministic semantic checks for AI-generated SQL
This tool says your AI can write code that looks right but secretly gets the numbers wrong
TLDR: sqlsure is a new tool that tries to catch sneaky wrong answers in AI-written database queries before they run, including double-counted totals and privacy leaks. Commenters were split between “finally, a safety net” and “maybe this is solving bad setup with fancy branding.”
A new Show HN post came in with a bold promise: your AI assistant may be confidently writing database questions that return totally wrong answers, and this tool wants to stop the chaos before anyone hits run. The project, sqlsure, checks whether a query is meaningfully correct, not just whether it runs. The pitch is spicy: it can catch silent mistakes like double-counted revenue or private patient details leaking out, all in a blink, offline, with no data leaving your machine.
But the real popcorn moment was the comment section. The creator showed up calm and confident, basically saying, this is about stopping one of the most annoying “looks fine, is wrong” bugs in data work. Then the skeptics arrived. One commenter absolutely dragged the vibe of the launch writeup, saying they "really struggle comprehending AI written READMEs," which is the kind of insult that hits extra hard in a product aimed at the AI age. Another pushed back on the core premise, arguing the example bug didn’t prove this problem is common enough to need a whole library, and bluntly suggested the issue might just be bad database design dressed up as innovation.
So the mood? Equal parts intrigue and side-eye. Some readers see a much-needed lie detector for AI-made analytics. Others are stuck on a more basic question: is this a lifesaver, or just a very polished fix for a mess teams shouldn’t have made in the first place?
Key Points
- •Sqlsure is described as a deterministic SQL semantic checker that runs before query execution and targets silent logic errors such as fan-out double-counting, invalid aggregations, unsafe joins, and sensitive column exposure.
- •The article says sqlsure was run on 2,568 expert-written queries from the Spider and BIRD benchmarks, producing 45 flags with zero false alarms.
- •It claims sqlsure identified a BIRD benchmark gold query that was wrong by 8× and found a schema defect that was filed upstream.
- •Sqlsure derives its rulebook from existing schema metadata such as dbt tests, PK/FK declarations, or direct database introspection rather than from LLM-based review.
- •The tool is presented in three deployment modes: CI gate, MCP server for AI agents, and Python library integration for text-to-SQL systems or evaluation workflows.