We created a simple product that executes decisions made by POND AI on-chain.
This algorithm consists of a single decision-making resolver function and uses powerpool.finance as a DePin Keeper network
Firstly, the POND API is invoked to obtain the model metrics. Since the prediction is made at an hourly interval, and this interval is deemed short enough to allow a linear approximation of the metrics, the trend accuracy is chosen as the target (at least until POND implements some suggestions for metrics we have made).
If the current model's accuracy is too low, the results are immediately discarded, and the execution instance terminates.
If the accuracy is higher than a given threshold, then the POND API is again invoked to obtain the prediction for the non-stable token in the pair. The current price is also obtained in the same call.
If the projected change of the non-stable token in relative units is below a user-specified threshold, the results are again discarded (as too small a change may not warrant the risk of a swap or the costs thereof).