Directed Information $\gamma$-covering: An Information-Theoretic Framework for Context Engineering

Abstract

We introduce \textbf{Directed Information $\gamma$-covering}, a simple but general framework for redundancy-aware context engineering. Directed information (DI), a causal analogue of mutual information, measures asymmetric predictiveness between chunks. If $\operatorname{DI}_{i \to j} \ge H(C_j) - \gamma$, then $C_i$ suffices to represent $C_j$ up to $\gamma$ bits. Building on this criterion, we formulate context selection as a $\gamma$-cover problem and propose a greedy algorithm with provable guarantees: it preserves query information within bounded slack, inherits $(1+\ln n)$ and $(1-1/e)$ approximations from submodular set cover, and enforces a diversity margin. Importantly, building the $\gamma$-cover is \emph{query-agnostic}: it incurs no online cost and can be computed once offline and amortized across all queries. Experiments on HotpotQA show that $\gamma$-covering consistently improves over BM25, a competitive baseline, and provides clear advantages in hard-decision regimes such as context compression and single-slot prompt selection. These results establish DI $\gamma$-covering as a principled, self-organizing backbone for modern LLM pipelines.