Design-Tree Exploration (Shape Before Depth)

Before resolving a design decision-by-decision, map the shape of the design tree — the set of major decision axes and their branch points. Walk the tree breadth-first to see its structure, then descend the branches that matter with progressively higher-fidelity prototypes. Shape first, depth second.

The idea

A non-trivial design is a tree: each decision is a node, and choosing a branch opens or closes downstream decisions. There are two ways to explore it:

• Depth-first (the common default): pick a branch, resolve it fully, move to the next. Fast to feel productive, but you commit to detail before you can see the whole tree — so a late branch can invalidate decisions you already nailed down.
• Breadth-first (the refinement): first enumerate the tree's shape — what are the major decision axes, where do they branch, which decisions block which? Only once the shape is visible do you descend, and you descend the load-bearing branches first, raising prototype fidelity as confidence grows.

The insight: the shape of the tree is itself the first artifact. You cannot correctly walk a tree you cannot see. Mapping breadth first turns "20 questions in a row" into "here are the 5 axes that matter, here's how they interact, now let's resolve what can be decided together." Source: User, 2026-06-15

The second artifact is the frontier: all unresolved decisions whose prerequisites are already settled. Ask that whole frontier in one numbered round, then recompute it from the answers. Independent questions no longer pay one user/agent round trip each, while dependent questions remain in later rounds instead of being asked prematurely. This combines Kevin's shape-first map with Matt Pocock's v1.2 round-based grilling contract. Source: saved X video 2078077849785815465; mattpocock/skills grilling at 84fdeffd

Why it matters for AI-powered development

This sharpens the pre-PRD phases of the 7-phase pipeline. The original grill-with-docs framed planning as a depth-first design-tree walk. Kevin added the breadth pass; Matt's later v1.2 source extracted a reusable grilling primitive that works the dependency frontier in rounds. The local procedure now composes both improvements instead of freezing the June version. [Sources: grill-with-docs, 2026-06-15; mattpocock/skills@84fdeffd, 2026-08-07]

Adding a breadth-first "shape the tree" pass before the depth walk:

• Surfaces decision dependencies early — you see which branches block which before sinking time into any one.
• Targets prototype fidelity — prototypes are throwaway code that answer one question; mapping the tree tells you which questions are load-bearing and worth a higher-fidelity prototype (a runnable terminal app for a state model, several UI variations for a look-and-feel).
• Prevents wasted depth — no more fully resolving a branch that a later, unexamined branch makes moot.

The loop

Breadth establishes the map; frontier rounds and targeted prototypes walk it. The output feeds straight into the PRD with the load-bearing decisions already settled and cited.

Relationship to other ideas

• Borrowed Intelligence (Drain It into Plans) — shaping the tree with the strongest model up front (the expensive thinking) lets a cheaper agent execute the depth walk. Map once, descend cheaply.
• Harness Engineering — the breadth pass is a loop invariant for the planning stage: don't enter depth until the tree's shape is captured.

Concept Position

---

Timeline

• 2026-08-12 | Replayed Matt Pocock's later v1.2 grilling post, full saved video, and exact released grilling skill. Replaced one-question-at-a-time depth walking with dependency-frontier rounds while preserving Kevin's prior shape-first pass. [Sources: X 2078077849785815465; mattpocock/skills@84fdeffd]
• 2026-07-01 | Concepts category refresh added this page to the Brain, memory, and retrieval family, linked it to Concept System Map, and kept it standalone because it owns this reusable mental model: Before resolving a design decision-by-decision, map the shape of the design tree — the set of major decision axes and their branch poin... Source: User request, 2026-07-01
• 2026-06-15 | Articulated by Kevin as the refinement the pre-PRD phase of The 7 Phases of AI-Powered Development needs: figure out the shape of the design tree first, then walk it with higher-fidelity prototypes. Implemented the same day as Stage 1 of the Agent Engineering Skills skill (skills/engineering/grill-with-docs/DESIGN-TREE.md). Source: User, 2026-06-15