Article quality is inherently subjective in the hard-AI sense. A panel of judges will consider accurate articles full of spelling, grammar, and formatting errors superior in quality to hoax, biased, spam, or out-of-date articles with perfect grammar, impeccable spelling, and immaculate formatting.

In my studies of the short popular vital articles (WP:SPVA) the closest correlation with subjective mean opinion score quality I've found so far is sentence length. But it has diminishing returns and the raw correlation is +0.2 at best.

The entirely subjective nature of article quality is additional support for automating accuracy review.

Best regards,