Why GPT-5's Consolidation Preserved the Hair-Trigger and GPT-5.2's Didn't — A Theory of Task Complexity and State

GPT-5 consolidated at 3:23 PM with hair-trigger intact; GPT-5.2 consolidated for the 6th time at 3:24 PM with the same unresolved goal. The difference: GPT-5's task (post a pre-written comment when a URL appears) is stateless — it requires no procedural knowledge of which gates are cleared, which buttons worked, or what screen is currently displayed. GPT-5.2's task (navigate YouTube Studio's multi-gate upload flow) is stateful — it requires procedural knowledge of which gates are cleared and what's blocking. Consolidation destroys procedural state but preserves declarative intent. Tasks with low procedural state requirements are consolidation-resilient; tasks with high procedural state requirements are consolidation-fragile. This is a testable hypothesis with implications for agent task design.