BETO: New Protocol Formalizes What LLMs Don't Know in AI-Generated Software
Key Takeaways
- ▸BETO formalizes the gap between what humans declare and what LLMs assume, converting silent completion into traceable, blocking events
- ▸The protocol enforces three human gates (G-1, G-2, G-3) that maintain operator authority and prevent bypassing of authorization decisions
- ▸Available in three implementation modes: manual protocol, automated Python executor, and interactive Claude Skill for low-friction adoption
Summary
Alberto Ramírez has released BETO (version 4.2), an epistemic governance protocol designed to address a fundamental problem in AI-assisted software development: LLMs complete specifications by inventing undeclared details rather than acknowledging gaps in knowledge. The protocol formalizes the boundary between operator-declared requirements and model-assumed elements, preventing what Ramírez calls the "silent completion problem" that makes AI-generated software difficult to audit and verify.
BETO introduces three epistemic states for system elements (DECLARED, ASSUMED, NOT_STATED) and enforces an 11-step governance process with three human gates that maintain operator authority over topology, specification, and code materialization. Elements marked NOT_STATED cannot be silently resolved—they must either be formally declared or registered as known system limits. Every undeclared element generates a traceable BETO_GAP event, creating an auditable chain from requirement to implementation.
The framework is implemented across three execution paths: the BETO Protocol itself (the core governance process), the BETO Executor (an automated Python pipeline tested with Claude Sonnet and Qwen-Coder), and the BETO Skill (a Claude integration requiring no infrastructure). In empirical testing, three cycles produced 43 source files with 100% TRACE_VERIFIED delivery and zero silent completions, demonstrating the protocol's effectiveness at enforcing traceability in AI-assisted code generation.
- Empirical validation achieved 100% TRACE_VERIFIED delivery across 43 source files with zero unaudited completions
Editorial Opinion
BETO addresses a genuine and underappreciated problem in AI-assisted development: the conflation of declared intent with model-generated assumption. By making the epistemic gap explicit and non-bypassable, this protocol provides a much-needed mechanism for auditability in systems where LLMs play a material role in specification and code generation. The framework's emphasis on structural traceability over mathematical proof is pragmatic, though organizations will still need to validate operator declarations independently. The availability of a Claude Skill integration significantly lowers adoption friction.
