As crude as can be, avalanche pulse generator (waveform into 50 ohms shown as Ref), BNC tee to binding posts and scope; binding posts to twisted pair up to gate. Waveforms are interpolated from a 10GS/s acquisition (equivalent time sampled, 350MHz BW).
Note that the scope and source act in parallel, hence Rsrc = 25 ohms.
Decimals are incidental (why bother rounding down?). The RLC network is simulated with a modest timestep because it's easily written that way (naive Newton integration), but high frequency elements are easily divergent at this scale (e.g., if C is made to be a few pF, or L1 or L2 is made similarly small), and a better integration method would be beneficial (trapezoidal, RK2..). I don't feel like writing those out in a spreadsheet, much faster to load up another few thousand cells and run it again.
The L1-C-L2 parameters were intended to simulate the twisted pair, but it seems a best fit doesn't quite fit with those alone (i.e., the high frequency ringing isn't captured well). I think an LC tank in series with what's shown would get closer (i.e., simulating mode conversion because twisted pair). Anyway, that's just the tight squigglies, which it seems got averaged over pretty well on the best-fit, so that's nice.
The R value seems robust; it acts like a vertical offset to the 'rebound' phase of the waveform, and nearby values of R and C2 don't fit nearly as well as these do.
Tim