Some people won't agree with this, but if you've gone over one, then you know.
HP 8640B.
The supposedly ultra-stable low-phase noise design suffers plenty from short-sided design choices when it's working. Let's start with the supposed phase noise claim. That phase noise is only available if:
1. You're locked to a 5 MHz ultra-stable source (and remember, using basic logic circuits to "divide down" a 10 MHz source induces what? Phase noise.)
2. You're ok with 100 Hz tuning steps. Yeah. Because you do not get the low-phase noise unless you can engage the PLL Locker. And the locker won't engage unless all the digits are shown on the 5 digit display. Oh, sure, you can lock and then de-tune the PLL circuit with a knob, which induces, you guessed it, more phase noise, as you're suddenly at the mercy of a variable resistor.
But ok, fast forward and look at this beast honestly.
Mechanics:
It has a ridiculous linked-(un)limited-slip differential scheme going on, using plastic gears, between the FM deviation and the frequency range knob. Yes, multiple layers of complex gears. It's like taking apart a mechanical alarm clock. The gears crack over time and then begin to slip or just fall into pieces.
Then there's the slider switches on the front panel. These break off because the enormous lever arm action coupled with the stiff click-click of the slider causes the brittle plastic to snap off. You can fix this by making your own metal replacements! Yea! And then you can glue them with epoxy to whatever is left inside the HP 8640B. Fun times!
Don't forget the camshaft that the range knob operates. Oh, yes, not only does the range knob operate a nasty arrangement of plastic gears in the front, but it also operates an enormous enigma of camshaft switches deeper inside which switch around how the UHF cavity oscillator is divided down.
But there's more to hate here. The PLL boards mate to the main board using these horrible and unreliable finger connectors. If yours hasn't failed yet, then it will!
And then there's the usual. RIFA capacitors at the rear, sealed pots that need to be cleaned inside somehow, the occasional failed capacitor. We all expect that stuff.
Oh and don't forget the broken gold fingers on the wafer-like switches. Yes, instead of normal wafer switches, HP etched switch paths on gold PWBs, sort of like how a modern handheld DMM handles the big switch. Except, to attach the wipers of the switch, they hot-melted the little gold fingers into (you guessed it) another plastic piece, which of course slowly falls apart. Don't forget to literally shake the gold pieces out when you first open one of these up -- they are basically impossible to replace without cannibalizing another instrument.
Ultimately, a small amount of relays or even solid-state switching, plus some standard switches, would have saved this instrument from such an early demise. Don't believe me? Instruments made before and after it used other not-completely-fantastic-plastic methods to control the circuits.
After many hours of gluing, grinding, probing, replacing, and swearing, you end up with a signal generator with low frequency resolution and phase noise specifications that it will never meet unless you have a cesium clock at 5 MHz to feed it with. To be fair I considered such (just for fun).