
AI coding agents can change hundreds or thousands of lines of code in seconds. The problem is that traditional line-by-line diffs often make review harder: cosmetic refactors, moved code, formatting changes, and structural rewrites can overwhelm the small changes that actually affect behavior. Isomorph is a structural-forensics tool for AI-generated code changes. Its core question is simple: “An AI changed 1,000 lines. What actually changed?” Instead of sending an entire textual diff directly to an AI model, Isomorph first parses the original and modified code using Tree-sitter and performs deterministic structural analysis. It identifies unchanged, cosmetic, moved, inserted, deleted, updated, rewritten, and unknown regions, then applies rule-based risk classification to isolate behavior-sensitive changes. IBM Bob provides the contextual reasoning layer. After the deterministic engine isolates relevant structural evidence, Bob can use localized change information and repository context to explain why a change may matter and what should be verified. Isomorph then adds behavioral evidence through targeted test execution. The final result is a Change Proof containing structural evidence, risk information, contextual explanation, verification results, and a reproducible SHA-256 proof hash for the generated artifact. The target users are developers, engineering teams, and organizations adopting AI coding agents who need a faster and more evidence-driven way to review generated changes. What makes Isomorph different is the order of operations: deterministic analysis establishes the structural facts first, AI provides contextual interpretation second, and executable verification provides behavioral evidence. Isomorph does not claim universal semantic equivalence or perfect correctness. Instead, it turns a large, noisy AI-generated diff into a focused, evidence-backed verification workflow. The result is a simple shift: Diff noise → Structural signal → Risk → Evidence → Verification → Change Proof.
27 Sep 2026