An Experimental Comparison of Six AI Agent Session Reset Policies

I ran 36 coding-agent runs under six context-clearing policies. Clearing after every task cost 33.5% more than clearing every third task.