PAPER / ARXIV:2609.20446
Barzelay, U.; Azulai, O.; Friedman, I.
RESUMO
Coding agents and evolutionary code-search systems can improve implementations once a target evaluation criterion have been specified. Applying these methods to an existing software repository raises an earlier question: which implementation choices are worth investigating for a high-level engineering objective? We introduce optimization-opportunity discovery, the repository-level task of identifying candidate source regions, explaining how they relate to objective, and proposing possible changes. The task takes as input a repository, optional runtime evidence such as offline telemetry observations or profiles. It does not require the user to specify a defect, bottleneck, or code location. We present Spotlights, a system that performs this task through logical mapping, successive agent reviews, and research linking candidates to relevant techniques. We evaluate Spotlights across model serving, document retrieval, blockchain ordering, and processing. Across three cases, it recovers seven of nine expert-selected targets. In the reliability study, 70% of top ten candidates meet the stated correctness and severity thresholds. Across five repeated retrieval runs, 73.6% of occurrences match the source region in all runs. Spotlights also rediscovers a withheld retrieval optimization and connects it to a tiling technique. In a discovered change reduces end-to-end runtime by 10.6% while preserving measured output quality. These results establish optimization-opportunity discovery as distinct and empirically evaluable step between broad engineering objective and subsequent implementation validation.
NO MESMO MAPA