Parameter names are currently messy and unintuitive.

The project is becoming too complex to be left undocumented.

• Currently library is only being setting up considering a single-shot environment with y_desired=1
• Not all agents with non-favorable outcome are getting recourse

• Single-shot vs Multi-shot recourse (e.g., loan)
• Binary vs. continuous recourse
• Fixed vs. dynamic thresholds
• Fixed vs. growing population
• Desired outcome positive vs. negative

Setting to prioritize:

• Single-shot, growing population, dynamic threshold, binary (loan scenario)

Alternatively, the action set could be passed to the recourse method and a population object would no longer be necessary

• K -> number of people or an array for each time step
• Fixed number (absolute) or an array for each time step

The initial test suite is very superficial. More corner cases must be tested and more checks must be added.

The Population's representation should include info on this.

Determining the probability of an adapting agent to get a loan if they adapt in the binary format results is determined by:

1. The amount of new agents entering (calculating the probability a new agent to be above the threshold)
2. The amount of adapting agents

• Raise warning if the population becomes empty.
• Raise warning if it is not possible to find a counterfactual.

• Run for the binary setting
• Add probability of new agents to be above threshold
• Find situations where threshold is stable
• Plot success rate over time steps