I think you're confusing occasional accuracy with reliable medium range forecasting.
If you think back to the old Countryfile forecasts they would only ever look out towards the end of the week, just 4-5 days ahead, and frequently the predctions for end of the week were very vague. Now the forecasters routinely look out a week to 10 days with similar levels of accuracy.
The verification stats show the improvement in model accuracy so my view is backed by the evidence.
Little more than 40 years ago there were occasions when the evening forecast was entirely wrong just 12-24 hours later.
Originally Posted by: Gandalf The White