Editorial Policy

China AI Bench is a small desk running an automated pipeline. An automated analysis framework monitors Chinese AI releases and assembles the source material; AI tools assist with research, drafting, and translation; a human editor verifies every claim, number, and link before anything publishes. That final step is not automated, and it never will be.

Every brief links a verifiable public source: official blogs, model cards, API docs, or named outlet reporting. We don't link to AI-farm re-blogs, and private conversations are never cited. Vendor-published numbers are labeled vendor-reported and cross-checked against independent evals where those exist.

Every piece runs under an organizational byline: News Desk for daily briefs, Benchmarks Desk for leaderboards and task-set runs, named editors for analysis pieces. These bylines stand for the desk process behind the work: we don't fabricate individual authors, and AI tools are never listed as authors.

Every item we publish adds something the source doesn't: a hands-on test result, a verified data point, or an independent take. Pure aggregation — restating a press release or re-printing a headline — is not published. Brief items carry a one-line take on why global developers should care.

Mistakes happen. Email us and we verify within 48 hours. Confirmed errors are corrected as a dated update at the top of the original piece — what changed and when. We never silently edit, and errata stay part of the record.

When a model version changes, pages built on the old results are re-checked: spot checks within 48 hours of a major release, full re-tests on a schedule, re-dated when results change.

AI tools are part of the pipeline: research, drafting, summarization, translation — and we say so plainly. AI is a tool in the pipeline, not an author, and not a source of fact. Images generated with AI are labeled with the tool used. Nothing publishes without human review.