Evaluate with the information available then
A task created earlier can have its solution published online later. Historical web context helps you examine what an agent could have found using the information available at the task’s cutoff date.
Search the web at a point in time.
Time machine lets you query the web as it was at a chosen date. Explore historical web context for reinforcement learning (RL), agent evaluation, and backtesting. Request access to discuss your dates and data requirements.
01
Choose a point in time
02
Retrieve historical context
03
Evaluate with hindsight
Historical context for meaningful comparisons
Control the web context during RL
Training tasks can become easier when their answers appear online during a run. Use a historical cutoff to examine what a search agent could have found before those solutions were published.
Backtest with the right context
Investigate forecasts and research decisions using information available at the time. Separate the evidence a model could have used then from what became known afterward.