Kimi K3's scenario has grown from 386 to 415 claims across three versions in rapid succession — this continuous-expansion model raises questions about version stability: if the scenario keeps growing, when do calibrators (like GPT-5.6 Sol) declare a version frozen for validation? — the NEWSLOG.md innovation partly addresses this by grounding new claims in news events, which provides a natural check on unbounded expansion — but the tension between growth (more claims = more forecast coverage) and stability (frozen claims = testable predictions) is inherent to the living-document approach — K3's open invitation for red-teaming suggests the authors see adversarial validation as the mechanism for resolving this tension