Editorial policy
How we use sources, distinguish examples from measurements and handle corrections.
Useful explanations, traceable evidence
We focus on concrete questions from engineers operating LLM systems. Articles should explain a decision, show its assumptions and give readers a way to verify the result. We link vendor documentation, model cards and project repositories rather than treating copied summaries as evidence.
Three kinds of evidence
- Reference guide: a documentation-based explanation and our analysis; not a claim of lab validation.
- Worked example: an explicitly hypothetical calculation or workflow, with stated inputs.
- Measured result: requires hardware and software versions, commands, workload, raw results and limitations. None are currently published here.
Corrections and review dates
Send corrections to [email protected]. Substantive changes receive an updated date and a note describing what changed. Original publication dates remain visible. We do not refresh dates solely to make content look new.
AI assistance and originality
AI tools help draft and maintain this publication. They are not evidence of accuracy or hands-on experience. Our goal is to add useful explanations, calculations and decision frameworks to cited sources, without presenting invented logs, benchmarks or credentials.
Advertising and content boundaries
Technical advice must remain separate from advertising. The site does not solicit ad clicks, publish paid rankings disguised as testing or accept unreviewed public posts. Corrections, attribution concerns and inappropriate content can be reported through the contact page.
Updates
We prioritize corrections, changed upstream behavior and reader questions over a fixed article quota. The updates page and RSS show actual changes; a publishing schedule is not a substitute for evidence.