2nd/10_Wiki/Topics/Architecture/Bottom-Up-Approach.md

---
id: wiki-2026-0508-bottom-up-approach
title: Bottom Up Approach
category: 10_Wiki/Topics
status: verified
canonical_id: self
aliases: [Bottom-Up Design, Bottom-Up vs Top-Down, Composition-First Design]
duplicate_of: none
source_trust_level: A
confidence_score: 0.9
verification_status: applied
tags: [design, methodology, top-down, bottom-up, architecture]
raw_sources: []
last_reinforced: 2026-05-10
github_commit: pending
tech_stack:
  language: Polyglot
  framework: Methodology
---

# Bottom Up Approach

## 매 한 줄
> **"매 bottom-up = 매 작은 building block 부터 만들고 합쳐 system 으로 키우는 composition-first 접근."**. 매 top-down 이 spec → decompose 인 반면, bottom-up 은 primitive → compose. 매 2026 modern 실무는 hybrid (meet-in-the-middle) — 매 spike/prototype 은 bottom-up, architecture 는 top-down, 매 LLM-driven 합성에서 bottom-up 의 compositional reasoning 이 다시 부상.

## 매 핵심

### 매 distinction
- **Top-down**: 매 high-level spec → subsystem → module → function. 매 known-domain 적합.
- **Bottom-up**: 매 primitive utility → combinator → application. 매 unknown-domain / discovery 적합.
- **Meet-in-the-middle**: 매 양쪽 동시 → 중간에서 만남.
- **Outside-in (TDD)**: 매 acceptance test → unit — 매 top-down 변형.

### 매 strengths
- **Reusability**: 매 primitive 가 여러 system 에 공통.
- **Testability**: 매 작은 unit 부터 100% covered.
- **Discoverability**: 매 unknown territory 에서 "뭐가 가능한지" 발견.
- **Compositional**: 매 functional programming, Lisp, Forth 의 핵심.

### 매 weaknesses
- **No big picture**: 매 individual unit 우수해도 system 일관성 부재 가능.
- **YAGNI risk**: 매 안 쓰일 primitive 까지 만들 수 있음.
- **Integration debt**: 매 끝에 가서야 합쳐지는 surprise.

### 매 응용
1. **Functional libraries**: 매 Lodash, Ramda — combinator 가 primitive.
2. **Forth / Concatenative langs**: 매 word 정의 후 합성.
3. **Embedded firmware**: 매 driver → HAL → app.
4. **ML pipeline**: 매 ops → layers → model.
5. **Spike / discovery prototype**: 매 architecture 모르는 phase.

## 💻 패턴

### Functional combinator (bottom-up)
```haskell
-- primitives
add1 :: Int -> Int
add1 x = x + 1

double :: Int -> Int
double x = x * 2

-- composition (no top-level spec needed)
addThenDouble :: Int -> Int
addThenDouble = double . add1

-- application emerges by composing
process :: [Int] -> [Int]
process = map (double . add1) . filter (> 0)
```

### Lisp — define primitives, then compose
```lisp
(defun double (x) (* x 2))
(defun increment (x) (+ x 1))
(defun process (xs)
  (mapcar (lambda (x) (double (increment x)))
          (remove-if-not #'plusp xs)))
```

### Embedded — HAL up
```c
// 1. register-level primitive
void gpio_set_high(uint32_t pin) { GPIO_BSRR = 1u << pin; }

// 2. driver
void led_on(led_t led) { gpio_set_high(led.pin); }

// 3. application
void heartbeat(void) {
    while (1) { led_on(STATUS_LED); delay_ms(500); led_off(STATUS_LED); delay_ms(500); }
}
```

### React UI — primitive components up
```tsx
// 1. atoms
const Button = ({ children, ...props }) => <button className="btn" {...props}>{children}</button>;
const Input  = ({ ...props }) => <input className="input" {...props} />;

// 2. molecules
const SearchBox = () => (
  <div className="search">
    <Input placeholder="Search..." />
    <Button>Go</Button>
  </div>
);

// 3. page (emerges)
const HomePage = () => <header><Logo /><SearchBox /></header>;
```

### Spike-driven discovery
```python
# Day 1 — bottom-up exploration before architecture exists
def parse_log_line(s): ...           # primitive
def filter_errors(lines): ...        # combinator
def aggregate_by_minute(lines): ...
def render_chart(buckets): ...
# After 2 days you SEE the pipeline → THEN write architecture doc
```

### LLM compositional planning (modern bottom-up)
```python
# Claude tool-use bottom-up: define small tools, let model compose
tools = [
    {"name": "search_db", "description": "..."},
    {"name": "render_chart", "description": "..."},
    {"name": "send_email", "description": "..."},
]
# Model receives high-level task and composes a plan from primitives
client.messages.create(
    model="claude-opus-4-7",
    tools=tools,
    messages=[{"role": "user", "content": "Send a weekly sales summary"}],
)
```

### Forth — concatenative composition
```forth
: SQUARE  DUP * ;            \ primitive
: CUBE    DUP SQUARE * ;     \ compose
: AREA    SQUARE 3.14159 * ; \ compose with literal
```

## 매 결정 기준
| 상황 | Approach |
|---|---|
| Domain unknown / spike | Bottom-up |
| Spec frozen, deadline | Top-down |
| Reusable library | Bottom-up |
| End-to-end product | Hybrid (meet in middle) |
| TDD acceptance | Outside-in (top-down variant) |
| LLM tool-using agent | Bottom-up primitives, top-down task |

**기본값**: hybrid — 매 architecture skeleton (top-down) + 매 primitive layer (bottom-up) → 매 meet middle.

## 🔗 Graph
- 응용: [[Functional_Programming]] · [[Atomic_Design]]
- Adjacent: [[Composition]] · [[YAGNI]]

## 🤖 LLM 활용
**언제**: unknown domain spike, library design, primitive 조합 탐색, agentic tool composition.
**언제 X**: regulated/safety system 의 spec-driven part — 매 top-down 우선.

## ❌ 안티패턴
- **Pure bottom-up without integration plan**: 매 primitive 100개 + 통합 안 됨 = 0 가치.
- **Premature primitives**: 매 안 쓰일 utility 양산 (YAGNI 위반).
- **Bottom-up for legally-specified systems**: 매 spec drift → compliance fail.
- **Skipping integration tests**: 매 unit 모두 green 인데 system fail.

## 🧪 검증 / 중복
- Verified (Abelson&Sussman SICP 1985, Forth dictionary docs, "Out of the Tar Pit" 2006).
- 신뢰도 A.

## 🕓 Changelog
| 날짜 | 변경 |
|---|---|
| 2026-05-08 | Phase 1 |
| 2026-05-10 | Manual cleanup — bottom-up vs top-down, composition patterns, hybrid default |