That's the problem with using data by specimen date. It does give you a more accurate picture over the medium-to-long term, but because there's a timelag you pretty much have to ignore at least the latest four days' data (mainly because we're so useless at turning test results around).
If you look at the difference between one day's figures for the previous 4/5 days and compare them with those published 24 hours later, you would see that the previous day's numbers increase by about 10-fold, the day before that by around 3-fold, the day before that practically doubles and the two days before that increase by around 10-25% each. If you then consider that yesterday's numbers will undergo each of these transitions over the 5-day period, you can see that the most recently published figures are unrecognisable within fairly short order.
That said, it may be the case that there is the beginning of a downward trend in these cities. Whether that actually is the case, whether that trend then sustains and whether those individual cities are representative of the rest of the country are complete unknowns of course.