It sounds like the fuel valves which control the engine's throttle took longer to turn up or down the rocket than expected.
The control system that is constantly processing rocket position and turning that into throttle outputs couldn't handle how long it took for decisions to turn into actions. That tends to cause wobbling in most controllers (overshoot, undershoot, etc).
The "lag" is a delay between your steering signal and the state of the system changing in response. In this case, because of static friction, valves controlling fuel flow reacted with delay (I'm not sure if they're talking about Merlin's valves or attitude control engines though) - so the rocket did not react immediately after being told to do so.
There's a whole math field around that topic, known as "control theory". There are ways to determine the limits of inputs and outputs that will allow your control system to keep everything in the desired state. In this case however, the input lag wasn't accounted for and it pushed the system out of the space of stable states, beyond the ranges where the control unit could keep up with the changes, so it ended up oversteering. Fortunately, in this case it's mostly a software problem, i.e. they could tweak the controller to deal with the lag.
The best way to imagine it is to recall your first experiences with a bike or ice skates - trying to balance yourself on them is exactly the kind of thing the rocket was doing, probably with similar results.
My control theory is a bit rusty nowadays, but I hope this clears up some things.
[0] - if you ever hear about "PID controller", this is exactly where the name comes from - proportional, integral, derivative.
So looks like if they can solve for modeling this latency (likely a software problem), there's a decent shot they'll nail it the next time around?
Since valves cannot open instantaneously, the control system has to know how long the valve takes to open/close, and start that process a little bit early so the valve is open when it needs to be open. If something goes wrong with the valve, it doesn't open as quickly, and so the computer model thinks it should be open and isn't, which could easily go off into an unstable state.