Prior art: how humans decompose a task into (sub)stages

The "task → stages → sub-stages" structure, and its stopping rule, is studied across several traditions.

HTA — the most direct (Annett & Duncan, 1967)

Hierarchical Task Analysis recursively decomposes a high-level task into a hierarchy of subtasks (goal → subgoal → operation), each level carrying a "plan" that governs the sub-steps: > sequence · / either-or · + parallel · : order-not-critical. That is a stage + internal sub-stages + control flow.

The stopping rule. HTA stops decomposing by the P×C rule: stop when probability of failure × cost of failure drops to an acceptable level. Annett (1971) calls knowing when to stop "one of the most difficult features of task analysis" — decompose only where the expected cost of error still justifies it.

Sibling traditions (same structure)

• Newell & Simon — Human Problem Solving (1972): means-ends analysis, problem spaces, GPS — recursive goal→subgoal reduction. Plus Simon's near-decomposability (simon-ando) — the cognitive-science foundation.
• GOMS (Card, Moran & Newell, 1983): Goals/Operators/Methods/Selection; HTA often feeds it.
• HTN planning (Sacerdoti's NOAH 1975; Erol–Hendler–Nau, SHOP): task → subtask via "methods" — the AI formalization.
• WBS (project management, PMBOK): nested work packages; the "100% rule" and "8/80 rule" (8–80 h packages) are granularity heuristics.
• Activity Theory (Leontiev): activity → action → operation.
• Hierarchical control of behavior (Powers' PCT; Carver & Scheier): be-goal → do-goal → motor, recursively controlled.

Takeaway

Every tradition shares the structure (recursive goal→subgoal decomposition + a plan/method per level) and flags the same hard problem: the stopping rule. HTA's answer is P×C.