Server-side calculation of agent's performance metrics

[harpomaxx]

Currently, metrics such as win_rate, steps, detection rate, etc are calculated on the client side, which makes them prone to reporting errors. It is suggested to move the calculation of these metrics to the server side, where the current trajectories can be used as a starting point for these calculations. This would improve the accuracy and reliability of the metrics reported.

Proposal:

• Implement server-side processing to calculate key metrics (win_rate, steps, detection rate, etc)
• Replace or validate client-reported metrics using these server-side computed values.

Reasoning:

• Clients may report incorrect or manipulated values, undermining the integrity of the metrics.
• Centralizing calculations improves reliability and trust in reported statistics.

Related context:

• The current trajectories data might already include all necessary information to enable these calculations.

[ondrej-lukas]

The environment provides (upon the agent's request) the previous trajectory when resetting the game. As of right now, it does not store the historical plays and therefore can't compute any statistics apart from the single episode currently in memory.

[ondrej-lukas]

How about creating a utility function to load a set of trajectories from the standardized format from a file and compute the stats from those?