What were some big model busts?

Do any of you recall any specific instances where the numerical models screwed up on basic weather fields such as pressure, frontal positioning, gradients, moisture, and so forth? This does not count precipitation, precipitation type, and so forth.

I'll investigate and post some examples here.

April 20, 2005 comes to mind. I believe that was the day. sw NE looked amazing the night before the event. The day of was a bust with much weaker wind fields.
Some random examples off the top of my head:

September 16, 2006... For the system setting up the southeastern SD tornadoes, run upon run of the NAM had significantly stronger sfc thru 500mb lows than the GFS, even up through the 06Z model that morning (i.e. 985mb sfc low vs 995 from GFS). The GFS nailed it, and the NAM "came around" by 12Z; either way, the shear was sufficient for signficiant tornadoes.

April 29, 2005... the NAM and the GFS vastly overestimated the intensity of a powerful system progged to move into AR/MS/TN... which ended up deamplififying and shoving a strung-out cold front through the area instead. Haven't looked closely at this one; at the time I was too busy driving to the pineforests of southeast AR.

July 18, 2004... NAM and RUC both underestimated boundary layer dewpoints in the northern Plains by around 10F. MLCAPE was thus much stronger than progged, which helped fuel an unexpected rash of violent northwest-flow tornadic supercells over eastern ND.

May 22, 2004... the 12Z RUC mixed the dryline all the way to far southeastern Nebraska/eastern KS by 21Z. A good example of the at-times mix-happy RUC.

I couldn't find examples of the latter ones without going into GRIB data, but here's an interesting composite of the 4/20/05 patterns. Red is the 24 h NAM forecast, black is the 00 h initialization for that evening. You can really see how the expected short wave trough in Colorado vanished.