A03

• Input: A procedural knowledge artifact.
• Operation: Encode its entry, actions, decisions, branching, iteration, evidence updates, and termination using a finite typed vocabulary and explicit composition rules.
• Output: A normalized, executable-or-inspectable procedure graph plus unresolved elements.
• Scope conditions: Artifacts that prescribe reasoning or decision behavior.
• Failure conditions: Vocabulary cannot represent a material operation; encoding adds unstated steps; independent encoders diverge.
• Predicted effect: Relative to free-form summaries, normalized graphs will increase cross-representation structural agreement and defect detection without losing task-critical steps.
• Measurement: Graph agreement, seeded-defect recall, omitted-step rate, and execution success.
• Implementation test: Independent encoders process prose and diagram variants; validators check type/composition constraints.
• Decision rights: Encoder maps; validator rejects invalid forms; adjudicator preserves ambiguity.
• Uncertainty handling: Unknown and alternative subgraphs remain explicit.
• Verification: Round-trip reconstruction and task simulation.
• Provenance: Nodes link to source spans.
• Implementation cost: Vocabulary training and encoding.
• Limitations: Existing evidence addresses expressiveness/repeatability, not incremental utility. Direct: layered encoding. Inference: comparative prediction.