validate

Compares Sirius GPU output with DuckDB CPU baseline to identify faulty operators via phased validation.

Updated Mar 18, 2026

Search Tags:#diagnostics #gpu #validate #duckdb #operator #query-debug

Ready to run

npx skills add https://github.com/kindofluke/albatross-data --skill validate-kindofluke

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: validate
Source: https://github.com/kindofluke/albatross-data/tree/main/.vibe/skills/validate
Command: npx skills add https://github.com/kindofluke/albatross-data --skill validate-kindofluke

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Diagnose incorrect query results by comparing Sirius GPU output against a DuckDB CPU baseline, identifying the faulty operator and narrowing down the root cause.

Core Features & Use Cases

Row-count analysis across operators to spot divergences.
Data checksum validation to verify values and detect data drift between GPU and CPU runs.
Phase-guided diagnostics from baseline setup to operator-level deep dives, with user prompts for each phase.
Use cases include debugging queries that return wrong results in GPU-accelerated pipelines and pinpointing the source of the discrepancy for remediation.

Quick Start

Provide the SQL query (or path to your data) and the tool will compare GPU results to a DuckDB CPU baseline to identify the faulty operator and guide remediation.

Frequently Asked Questions about validate

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I diagnose incorrect GPU query results against a CPU baseline?▼

To diagnose incorrect GPU query results, you can compare Sirius GPU output against a DuckDB CPU baseline to identify the faulty operator and narrow down the root cause of the discrepancy. This process uses phase-guided diagnostics.

What is the best way to pinpoint a faulty operator causing data drift in an accelerated pipeline?▼

Pinpointing a faulty operator causing data drift involves performing row-count analysis across operators and validating data checksums to verify values and detect discrepancies between GPU and CPU runs.

How do I debug a query that returns wrong results in a GPU-accelerated pipeline?▼

Debugging a query that returns wrong results in a GPU-accelerated pipeline requires comparing the GPU output to a DuckDB CPU baseline through phase-based validation, starting from baseline setup to operator-level deep dives.

Do I need per-operator instrumentation to validate GPU output with DuckDB?▼

Yes, validating GPU output with DuckDB requires per-operator instrumentation and optional diagnostic logging, including checksum capture and stream synchronization checks, to accurately identify the source of query discrepancies.

Can I run GPU query validation autonomously without manual approval for each phase?▼

GPU query validation supports interactive, autonomous, or semi-autonomous modes, but the default phase-guided diagnostics expect the user to approve each phase before proceeding to the next validation step.