The statistics refer to diagnosis of 2019-nCov based on conclusive test results. They do not include diagnosis solely made on symptoms - if somebody in China had flu-like symptoms and died from pneumonia, that's not considered a diagnosis of 2019-nCov infection and thus it's not an 2019-nCov death unless proper tests were performed.

Even if we assume that all the test results are reported, we know that large quantities of suspected cases in Wuhan are not tested - hospitals report there's an insufficient quantity of RNA test kits available, there are some reports that outpatients aren't tested, and bed capacity is limited, so anybody who does not have space in the hospital system is not tested. And if they die, they're excluded from statistics of 2019-nCov patients because they were not tested.

So if you have a system that's capable to test X people per day, even if you report all the test results perfectly honestly, you're not going to see more than X new infections per day... no matter how many new infections you actually have. If it's nontrivial for Japan to test 3600 people quarantined on a ship, how would it be possible to test everyone who's sneezing from 11 million people quarantined in a city?