Object Aligner paper experiments: code and aggregated results

Abstract

Object Aligner (OA) is a Python library that scores two JSON objects by recursively aligning their trees: the Hungarian algorithm for unordered collections, sequence alignment for ordered ones, and partial credit at the granularity the schema declares. It is configured entirely through a compact set of JSON Schema extensions, supports cross-referenced records (graphs and hypergraphs) up to identifier renumbering, and produces a deterministic, auditable score suitable as a reward signal for LLM prompt-optimization frameworks. The package also provides attribution, repair suggestions, natural-language feedback and optional embedding-based semantic similarity for strings. Reference implementation for the paper Object Aligner: A Configurable JSON Schema Similarity Score for Graphs, Applied to LLM Prompt Optimization (arXiv:2607.01972).