2nd/10_Wiki/Topics/AI_and_ML/Solution.md

id, title, category, status, canonical_id, aliases, duplicate_of, source_trust_level, confidence_score, verification_status, tags, raw_sources, last_reinforced, github_commit, tech_stack

id	title	category	status	canonical_id	aliases	duplicate_of	source_trust_level	confidence_score	verification_status	tags	raw_sources	last_reinforced	github_commit	tech_stack
wiki-2026-0508-solution	Solution	10_Wiki/Topics	verified	self	Solutioning Solution Design Problem-Solution Mapping	none	A	0.9	applied	methodology design-thinking problem-solving solutioning		2026-05-10	pending	language framework methodology design-thinking

Solution

매 한 줄

"매 solution 의 problem 의 inverse 의 X — 매 fit 의 search". 매 problem statement → constraints → option-space → trade-off → committed solution. 매 design thinking + engineering rigor 의 fusion. 매 2026 modern PRD/RFC stack 매 LLM-aided option exploration 의 기본.

매 핵심

매 problem-vs-solution 의 분리

• Problem: 매 user pain, business gap, technical debt. 매 solution-agnostic description.
• Solution: 매 specific approach 의 implement. 매 multiple options 의 enumerate.
• Anti-pattern: 매 "we need X" framing — 매 solution 의 jumped-to. 매 problem 의 first articulate.

매 5-step canonical flow

1. Articulate: 매 problem 의 1-sentence + 매 measurable success criterion.
2. Constrain: 매 budget, deadline, team, tech-stack, risk tolerance.
3. Enumerate: 매 3+ options. 매 do-nothing baseline 의 always include.
4. Trade-off: 매 each option 의 cost/risk/value 의 score.
5. Commit + reverse-doc: 매 chosen option 의 RFC/ADR write. 매 rejected options 의 reason 도 기록.

매 응용

1. Tech RFC / ADR.
2. Product PRD.
3. Customer-discovery loop.
4. LLM-aided option generation.

💻 패턴

ADR template (Markdown)

# ADR-042: Switch to event-driven order pipeline

## Status
Accepted (2026-04-12)

## Context
Sync API call chain causes 2.3s p95 latency under peak load (12k rps).
Current monolith RPC stack cannot scale beyond 18k rps without sharding.

## Decision
Adopt Kafka-based event pipeline for order lifecycle (created → paid → shipped).

## Consequences
+ p95 drops to 400ms (validated in load test).
+ Decoupled services enable independent deploys.
- Operational complexity: Kafka cluster, schema registry, DLQ.
- 6-week migration with dual-write phase.

## Alternatives considered
1. Sharded monolith — rejected: 4mo migration, no future-proof.
2. gRPC streaming — rejected: still tightly coupled.
3. Do nothing — rejected: SLO breach by Q3.

Option matrix scoring

# option_matrix.py
from dataclasses import dataclass

@dataclass
class Option:
    name: str
    value: int      # 1-5 (impact)
    cost: int       # 1-5 (effort)
    risk: int       # 1-5 (uncertainty)

    @property
    def score(self) -> float:
        # Weighted: value heavy, risk penalizing
        return (self.value * 2.0) - (self.cost * 0.8) - (self.risk * 1.2)

options = [
    Option("event-pipeline", value=5, cost=4, risk=3),
    Option("sharded-monolith", value=3, cost=4, risk=2),
    Option("do-nothing", value=0, cost=0, risk=5),
]
ranked = sorted(options, key=lambda o: o.score, reverse=True)
for o in ranked:
    print(f"{o.name}: {o.score:.2f}")

Problem-statement template

**Who**: Mid-market SaaS ops engineers (50-500 employee orgs).
**What**: Cannot debug Kafka consumer lag without SSH-ing into broker.
**Why-now**: Compliance requires audit trail + zero-trust env (no SSH).
**Success**: 80% of lag incidents resolvable via dashboard alone within 2026 Q3.
**Non-goal**: Replacing existing Kafka cluster.

LLM-aided option enumeration

# enumerate_options.py
from anthropic import Anthropic

client = Anthropic()

def enumerate(problem: str, constraints: list[str]) -> list[dict]:
    msg = client.messages.create(
        model="claude-opus-4-7",
        max_tokens=2000,
        messages=[{
            "role": "user",
            "content": f"""Problem: {problem}
Constraints: {chr(10).join(f'- {c}' for c in constraints)}

Generate 4 distinct solution options. Include 1 'do-nothing' baseline
and 1 'wildcard' creative option. For each: name, summary, est_cost (1-5), est_risk (1-5)."""
        }]
    )
    return parse_options(msg.content[0].text)

Trade-off heuristic gate

def should_pursue(option) -> bool:
    if option.risk >= 5:
        return False  # too uncertain, prototype first
    if option.cost > option.value:
        return False  # net-negative
    return option.score > 0

Reverse-doc rejected options

## Rejected: GraphQL federation
**Why considered**: Frontend wants typed schema, backend wants composability.
**Why rejected**: Federation gateway adds 80ms median latency.
   Team has 0 GraphQL prod experience. 4mo onboarding curve.
**Revisit when**: Latency budget grows OR team gains GraphQL expert.

매 결정 기준

상황	Approach
Vague stakeholder ask	매 problem statement 5W 의 force
2+ viable options	매 ADR + matrix scoring
Reversible decision	매 ship-and-iterate (Type 2)
Irreversible decision	매 deep RFC + review (Type 1)
Unknown unknowns	매 spike/prototype 의 first

기본값: ADR + 3+ option enumeration + reverse-doc.

🔗 Graph

• 부모: Design Thinking
• 변형: RFC-Process · ADR

🤖 LLM 활용

언제: 매 option enumeration brainstorm, 매 ADR draft, 매 problem-statement refinement, 매 reverse-doc generation. 언제 X: 매 commit decision (human accountability), 매 stakeholder alignment (in-person needed).

❌ 안티패턴

• Solution-first: 매 "we need Kafka" 매 problem 의 articulate 없이.
• Single option: 매 alternatives 0개. 매 confirmation bias.
• No baseline: 매 do-nothing option 의 omit. 매 cost 의 hidden.
• No reverse-doc: 매 rejected options 의 oral history. 매 future-team 의 repeat.
• Solution worship: 매 framework 의 fetishize over outcomes.

🧪 검증 / 중복

• Verified (Lightweight ADR by Michael Nygard; Amazon "1-pager"; ThoughtWorks Tech Radar 2026).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — full content (5-step flow + ADR/option-matrix patterns)