delegation-parallelismlisted

# Delegation & Parallelism > Status: v0.1 baseline draft — derived from the case catalog and axis definitions; awaiting trace evidence from Fable 5.1 case runs. ## When to apply - A task could be split into pieces that different agents or sessions might run at the same time. - You are about to fan out N workers and are unsure whether the pieces are actually independent. - You are doing serially something that has no ordering dependency and could overlap. - Two candidate subtasks touch the same files, the same branch, or the same shared resource. - You have delegated work and need to decide how often to check on it without babysitting. - The coordination overhead of splitting might exceed the time the split would save. ## Core loop 1. **Decompose the task into candidate units and their outputs.** List the pieces and, for each, the files/resources it writes and the artifact it produces. You cannot reason about independence until you can see each unit's write-set. 2. **Apply the independence test to each pair.** Two units are parallel-safe only if their write-sets are disjoint AND neither consumes the other's output AND they share no serial resource (same branch, same DB row, same file, same port). If any of those holds, they are ordered or must be merged — not parallel. 3. **Enforce disjoint scope before splitting.** For units you will parallelize, make the scope boundaries explicit: which files each owns, which it must not touch. Overlapping write-s