Testnet params

[danielattilasimon]

Let's agree on the parameters to be used in the testnet game.

I've started by copying the parameters of the FXS:cvxFXS pool for bLUSD:LUSD, as suggested by the Curve team on Telegram.

Apart from the Curve pool, the most important parameters for the game will be:

• How much we want to speed up the simulation (accrual of fake yield, bootstrap periods, etc)
• Fake APRs (BAMM & Curve)
• Initial accrual parameter (alpha)
• Target average bond age
• Chicken-in fee
• Chicken-in bootstrap
• Redemption bootstrap
• Shifting bootstrap

Of course we can tweak other parameters too, if we'd like.

[bingen]

My suggestion:

• How much we want to speed up the simulation (accrual of fake yield, bootstrap periods, etc): if we do 1 game hour -> 1 real day, then 5 game days would be 120 real days (~4 months), which sounds good. The only issue is that when we sleep we miss a whole week.
• BAMM APR: 7%
• Curve APR: 3%
• Initial accrual parameter (alpha): 10 days
• Target average bond age: 24 days
• Chicken-in fee: 10%
• Chicken-in bootstrap: 24 days
• Redemption bootstrap: 24 days more (48 days since the start)
• Shifting bootstrap: 72 days since the start

[danielattilasimon]

Interesting...

With that alpha and target average age (i.e. targetting average bonding time of 224 = 48 days), `lambda (1 - tau)has to be at least 1.44 for the optimal bonding time to be within 48 days (wheretauis the chicken-in fee). I'm usingrubber-chicken-ng` for rough calculations.

With the proposed tau = 0.1, this figures out to lambda > 1.6, otherwise it will take some time for the controller to reduce alpha to where bonding occurs with the targetted frequency. This may be a bit unrealistic given the pessimistic APR figures being proposed (are we playing hardmode? 😆).

For the sake of UX, I think it's better to potentially "undershoot" alpha (i.e. choose an alpha that may result in too fast accrual initally) rather than users having to wait ages for alpha to be set by the controller.

[bingen]

Definitely. I wanted to add comment saying that I’d rather leave alpha to you, as you are the controller expert. Anyway, for the record, yes, I was assuming lambda = 2 to do my napkin math for alpha, which is definitely high, but I would hope it can be realistic during the initial hype. But let’s agree on the rest of the values and then you can set the alpha that you think would work better with them.

[danielattilasimon]

Well, we can definitely try it with these parameters, and see what happens 😄 That would be interesting to see, for sure.

With such a high lambda, alpha = 10 is quite justified.