EMBER Learns Which Evidence to Retain Pre-Query Under Fixed Token Budgets

EMBER is a learned retention policy for long-horizon agents that stores verbatim source excerpts paired with retrieval keys before any query is issued, operating under a fixed token budget. It trains via post-query outcome feedback to maximize evidence recoverability at read time, addressing the retrieval and rereading costs of large archived histories.