2nd/10_Wiki/Topics/AI_and_ML/Schema.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-schema	Schema	10_Wiki/Topics	verified	self	Data Schema Schema Definition	none	A	0.9	applied	schema validation zod json-schema database		2026-05-10	pending	language framework TypeScript Zod 4

Schema

매 한 줄

"매 schema 는 data 의 contract — 매 runtime + compile-time 양쪽에서 enforce". 매 origin 은 1970s relational DB DDL; 매 modern state 는 Zod 4 (TS-first, runtime + types), JSON Schema 2020-12, Schema.org (web), Avro/Protobuf (wire), Postgres 17 declarative migration.

매 핵심

매 schema layer (매 stack 별)

• DB schema: DDL (Postgres CREATE TABLE), constraints, indexes.
• Wire schema: Protobuf, Avro, JSON Schema — 매 inter-service contract.
• App-runtime schema: Zod, Pydantic, Valibot — 매 API boundary 의 validation.
• Web schema: Schema.org JSON-LD — 매 SEO + semantic web.
• AI schema: 매 LLM structured output (OpenAI structured outputs, Anthropic tool use schema).

매 Zod 4 의 modern 위치 (2026)

• 매 single source of truth → infer TS type + runtime validate + JSON Schema 변환.
• v4 (2026) 의 major: pure-ESM, smaller bundle, sync error throw, .brand() first-class.

매 응용

1. API request/response validation (Hono, tRPC, Next.js Route Handler).
2. Form validation (React Hook Form + Zod resolver).
3. LLM structured output (Anthropic tool schema, function calling).
4. Config validation (env vars at boot).

💻 패턴

매 Zod 4 schema → TS type infer

import { z } from "zod";

export const UserSchema = z.object({
  id: z.string().uuid(),
  email: z.string().email(),
  age: z.number().int().min(0).max(150),
  role: z.enum(["admin", "user", "guest"]),
  createdAt: z.date(),
});

export type User = z.infer<typeof UserSchema>;
// → { id: string; email: string; age: number; role: "admin"|"user"|"guest"; createdAt: Date }

const parsed = UserSchema.parse(rawJson);  // 매 throw on invalid
const safe = UserSchema.safeParse(rawJson); // 매 { success, data | error }

매 Zod → JSON Schema (LLM structured output)

import { z } from "zod";
import { zodToJsonSchema } from "zod-to-json-schema";

const ExtractedInvoice = z.object({
  vendor: z.string(),
  total: z.number(),
  lineItems: z.array(z.object({
    desc: z.string(),
    qty: z.number().int(),
    unitPrice: z.number(),
  })),
});

const jsonSchema = zodToJsonSchema(ExtractedInvoice);
// 매 Anthropic tool input_schema 에 그대로 사용

매 Anthropic tool use (Claude Opus 4.7 structured output)

import anthropic
client = anthropic.Anthropic()

resp = client.messages.create(
    model="claude-opus-4-7",
    max_tokens=2048,
    tools=[{
        "name": "extract_invoice",
        "description": "Extract structured invoice data",
        "input_schema": {
            "type": "object",
            "properties": {
                "vendor": {"type": "string"},
                "total": {"type": "number"},
                "line_items": {
                    "type": "array",
                    "items": {
                        "type": "object",
                        "properties": {
                            "desc": {"type": "string"},
                            "qty": {"type": "integer"},
                            "unit_price": {"type": "number"},
                        },
                        "required": ["desc", "qty", "unit_price"],
                    },
                },
            },
            "required": ["vendor", "total", "line_items"],
        },
    }],
    tool_choice={"type": "tool", "name": "extract_invoice"},
    messages=[{"role": "user", "content": invoice_text}],
)
# resp.content[0].input → 매 schema-validated dict

매 Postgres 17 declarative schema (with constraints)

CREATE TABLE users (
  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  email TEXT NOT NULL UNIQUE CHECK (email ~ '^[^@]+@[^@]+$'),
  age INT NOT NULL CHECK (age >= 0 AND age <= 150),
  role TEXT NOT NULL CHECK (role IN ('admin','user','guest')),
  created_at TIMESTAMPTZ NOT NULL DEFAULT NOW()
);

CREATE INDEX idx_users_email_lower ON users (LOWER(email));
CREATE INDEX idx_users_role_created ON users (role, created_at DESC);

매 Schema.org JSON-LD (SEO, 매 Article)

<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Article",
  "headline": "Modern Schema Validation in 2026",
  "author": {"@type": "Person", "name": "Jane Doe"},
  "datePublished": "2026-05-10",
  "image": "https://example.com/cover.jpg"
}
</script>

매 Pydantic v2 (Python, 매 FastAPI 와 함께)

from pydantic import BaseModel, EmailStr, Field
from typing import Literal

class User(BaseModel):
    id: str
    email: EmailStr
    age: int = Field(ge=0, le=150)
    role: Literal["admin", "user", "guest"]

# FastAPI route — 매 자동 OpenAPI + validation
@app.post("/users")
def create_user(user: User) -> User:
    return user

매 schema migration (Drizzle, TS 2026)

// schema.ts — 매 source of truth
import { pgTable, uuid, text, integer, timestamp } from "drizzle-orm/pg-core";

export const users = pgTable("users", {
  id: uuid("id").primaryKey().defaultRandom(),
  email: text("email").notNull().unique(),
  age: integer("age").notNull(),
  role: text("role", { enum: ["admin","user","guest"] }).notNull(),
  createdAt: timestamp("created_at").notNull().defaultNow(),
});

// $ drizzle-kit generate  → 매 SQL migration auto
// $ drizzle-kit migrate

매 결정 기준

상황	Approach
매 TS API boundary	Zod 4
매 Python API	Pydantic v2
매 cross-language wire	Protobuf / Avro
매 LLM structured output	JSON Schema (via Zod/Pydantic)
매 SEO web page	Schema.org JSON-LD
매 RDB	Postgres DDL + Drizzle/Prisma migration

기본값: TS 면 Zod 4 (single source → type + runtime + JSON Schema).

🔗 Graph

• 부모: TypeScript 타입 시스템 (TypeScript Type System) · Validation
• 변형: Zod · JSON Schema · Protobuf
• 응용: API Design
• Adjacent: OpenAPI

🤖 LLM 활용

언제: 매 LLM structured output 강제 (tool use input_schema). 매 unstructured text → typed object extraction. 언제 X: 매 schema 자체의 design — 매 domain modeling 은 human 의 일.

❌ 안티패턴

• z.any() everywhere: 매 schema 의 의미 X.
• Schema drift: API schema 와 DB schema 가 따로 → 매 single source 필요.
• Over-validation in hot path: 매 매 request 마다 deep validate → 매 boundary 만 validate.
• Stringly-typed enums: 매 z.string() for role → 매 z.enum() 으로 narrow.

🧪 검증 / 중복

• Verified (Zod docs v4, JSON Schema 2020-12 spec, Schema.org, Postgres 17 docs).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — Zod 4 + Pydantic v2 + LLM tool schema + Drizzle