What direction does the author want to pursue?

[Dpananos]

What other types of simulations are you interested in adding? I could write a few in my spare time.

[statwonk]

@Dpananos thank you for the interest! My original need was to do simulations similar to a Frequentist "power study." I'm particularly interested in easy visual ways to explore the limits of experiments for the purposes of demonstrating how they fail, trade-offs and remedies to make them more robust.

The package is currently organized in one file with one function. It's a simple CRD of bernoulli experiments. It performs essentially a Bayesian one-tailed test. It seems like it could be made more generic by extracting this part of a function out? I think that'd let us de-couple the experiment definition from the way of summarizing the data. I have a visualization here, but I wonder if there's clearer ways to communicate calibration to folks that have little statistical expertise?

Because I believe the vast majority of experiments are CRD bernoullis, I'm as interested expanding on experimental design --- perhaps if the interface can be kept simple? At the same time, I could envision a CRD poisson experiment and perhaps making perform_experiment a higher-order function that accepts conjugate priors?

There's lots of possibilities. What matters most to me is:

• The package helps folks calibrate experiments
  • Learn about the trade-offs inherent in calibration (define calibration?)
  • Visualize failure cases (peeking? good vs. ill priors?)
  • Visualize success cases (priors protecting us from absurdity?)
• The interface is simple and easy to use
• Compositional in nature (tidy)

I hope that gives some material to start on, thank you for the interest! I'm definitely open to continuing to refine and flesh out the goal.

I'm interested to hear about what you (@Dpananos and others) hope / hoped to get out of a package like this?