It is worth being precise about the Chinese data problem, because the loose version of the complaint is wrong and leads to bad decisions. The published series are not generally inaccurate. The problem is that the most decision-relevant quantities are either unpublished or published in a form that cannot be decomposed.
What exists and is reasonably good: monthly customs crude imports by volume and origin, monthly crude production, monthly refinery throughput in aggregate, monthly product output and monthly product trade. That is a substantial dataset and it supports real analysis.
What does not exist in usable form: any regular inventory series, the split between commercial and strategic stocks, independent refinery run rates, and product stocks at the hub level. Those are precisely the series that would let you distinguish a demand story from a stockpiling story, which is the question that matters most.
Publication lag and revision behaviour add a further layer. Monthly data arrives with a lag of weeks, some series are revised, and definitional changes have occurred without prominent notice. Any long history assembled from these series should be treated as a series whose definition may have moved, and the burden is on the analyst to check rather than on the publisher to flag.