They saw a - possible - release of a haswell refresh (whatever than means) in Q2 '04 and called it that '04 will not have broadwell!
But wait a minute: Q2 is Q2! it's not the whole of the year! If Intel will start having 14nm products by the beginning of 04 it comes to reason they will not sit on an industry that provides cheaper products for the performance and leave it unused.
In fact it makes perfect sense that Q3+ of '04 will have 14nm chips, it's the point the distance from Haswell reaches a regular length (about 1 and a half year, I even graphed it based on past history of the past 3-4 iterations, it makes perfect sense to be Q3).
In fact (#2), I'm not even convinced 'refresh' is not nomenclature for broadwell, but even without that, it's perfectly possible to have both broadwell and a refresh in the same year.