Governed Capture: Creation-Time, Per-Operation Ratification of AI Contributions in Collaborative Modelling

Generative AI can now draft conceptual models, but in collaborative settings a prior question goes unanswered: how should a machine’s contributions be admitted to a shared model for which a human team remains accountable? Existing human-in-the-loop modelling supervises a single analyst’s generated draft after the fact; it does not govern what enters a team-owned artefact at the moment of creation. We present governed capture, a creation-time, per-operation ratification pattern: the assistant never mutates the shared model directly but emits typed operations over the modelling grammar, each held pending until an accountable human applies, edits, or rejects it. We state the pattern’s contract, conditions of use, and failure modes, and instantiate it in VibeModeling, a real-time collaborative EventStorming environment with role-asymmetric access. We evaluate We design a formative evaluation around the pattern’s central risk — that per-operation ratification degrades to rubber-stamping — instrumenting the proposal lifecycle to distinguish active scrutiny from passive acceptance. We claim feasibility, not comparative benefit, and defer validated usability and a measure of ratification quality to a planned multi-department industrial study. The contribution is a candidate governance pattern for AI-assisted collaborative modelling, of which VibeModeling is the first worked instantiation.