Search for a command to run...
We present Document-Directed Computation (DDC), a computational architecture in which human-authored documents containing embedded semantic routing annotations direct the assembly, parameterization, and execution of computational processes without modification to program code. The core innovation is a three-way separation of concerns: domain experts author knowledge in structured documents, engineers author computational logic in programs, and semantic routing annotations embedded in the documents connect the two — such that modifying a document modifies the computation without altering any code. The architecture comprises a fail-closed graph resolver that validates document dependency graphs for acyclicity, referential integrity, duplicate exports, and contract compliance before permitting computation; an immutable, content-hashed snapshot mechanism ensuring deterministic reproducibility; a language-agnostic subprocess execution contract; and a cryptographic computation witness binding document state to computation results. We validate the architecture empirically through a production deployment spanning 855 domain-expert-authored documents, executing a five-stage computational pipeline. A 1,000-event stress test across 30 guest sizes, 8 event types, and 18 budget tiers produces zero failures with 83ms average execution time. To our knowledge, no prior system provides the combination of expert-authored documents directing computation through a fail-closed validated dependency graph with cryptographic computation witnesses. Patent Support: Patent 9 — Document-Directed Runtime (20 claims, 4 independent). USPTO App# 64/022,480, filed March 30, 2026.