Chengyu Lin

Could you explain the use case here? Most of the actual `Signature` structs already have `Display` derived from `tagged` macro.

I just took a glimpse, ignoring the arity, the difference is already illustrated in page 3 figure 2, the graph 2 vs 3. The key trick here is, just shortcut...

Is this still working? We may still want a short pk version for a smaller stake table. cc @alxiong

To me, the issue feels like an [intel ADX](https://en.wikipedia.org/wiki/Intel_ADX) thing. We may compile 2 images for older CPUs and new ones.

> can i take on this ? Thanks for volunteering. However, we need an internal meeting to make some design decisions. Thus we'll take care by our own.

The other weird observation is, if you do not [warmup a new cuda stream per run](https://github.com/EspressoSystems/jellyfish/compare/main...cl/gpu-profiling#diff-653f41b63c2df09c518f6c718d2126c78945a67fca0f8bdd333776a410d29024R1334), the performance is also not good even with 1 batch. cc @alxiong

Also with one single batch, the type conversion takes one-third of the time. We could eliminate that #516

After this issue, what's the ideal way to get a stake table as a non-participant of consensus? Our light client prover service needs it.

But other parties can perform a liveness attack by submitting signatures with random keys.

Current VID parameter setting is UNSAFE! https://github.com/EspressoSystems/HotShot/blob/39127341371bac1d747ca905e5fd822890d28b2b/crates/types/src/vid.rs#L56 When `num_storage_nodes` is near a power of two, for example 17. `recovery_threshold` is set to 16 so we only have 1 redundancy. In...