2nd/10_Wiki/Topics/Domain_Programming/AI_and_ML/Reconciliation.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-reconciliation	Reconciliation	10_Wiki/Topics	verified	self	Data Reconciliation Conflict Resolution CRDT	none	A	0.9	applied	reconciliation crdt distributed finance sync		2026-05-10	pending	language framework typescript yjs

Reconciliation

매 한 줄

"매 두 source 의 truth 의 align". Reconciliation 의 financial (ledger ↔ bank statement match), data (master ↔ replica), distributed (CRDT conflict-free merge), AI (multi-agent state sync) 의 cross-domain pattern. 2026 의 local-first software 의 CRDT renaissance.

매 핵심

매 Financial reconciliation

• 3-way match: PO ↔ receipt ↔ invoice.
• Bank rec: GL ↔ bank statement 의 daily.
• Sub-ledger to GL: AR/AP detail 의 control account match.
• 2026 tools: BlackLine, FloQast, AI-driven (Trullion).

매 Data reconciliation

• CDC (Change Data Capture): Debezium, source-of-truth log.
• Lambda architecture: batch + stream — 매 reconcile 의 daily.
• Idempotency: 매 retry 의 safe — UUID dedup.
• Checksum: row-count + hash 의 source vs target.

매 CRDT (Conflict-free Replicated Data Type)

• G-Counter: grow-only counter (sum across replicas).
• PN-Counter: increment + decrement.
• OR-Set: observed-remove set — add wins.
• LWW-Register: last-write-wins by timestamp.
• RGA / Yjs Y.Text: collaborative text.
• Automerge: JSON CRDT, local-first.

매 Modern distributed

• Operational Transform (OT): Google Docs 의 legacy. Server-coord.
• CRDT: Notion, Linear, Figma (mixed). P2P-friendly.
• Vector clocks: causal ordering.
• Hybrid Logical Clock (HLC): physical + logical — CockroachDB.

매 응용

1. Collaborative editing (Yjs, Automerge).
2. Offline-first mobile app sync.
3. Multi-region database (DynamoDB Global, Cosmos DB).
4. Bank ledger reconciliation.
5. Multi-agent LLM shared scratchpad.

💻 패턴

Yjs collaborative text (TypeScript)

import * as Y from "yjs";
import { WebsocketProvider } from "y-websocket";

const doc = new Y.Doc();
const provider = new WebsocketProvider("wss://sync.example.com",
                                        "room-1", doc);
const text = doc.getText("content");

text.observe(event => console.log("delta:", event.delta));
text.insert(0, "Hello ");
// 매 다른 client 의 concurrent insert 의 auto-merge

Automerge JSON CRDT

import * as Automerge from "@automerge/automerge";

let doc1 = Automerge.from({ tasks: [] as string[] });
let doc2 = Automerge.clone(doc1);

doc1 = Automerge.change(doc1, d => d.tasks.push("write spec"));
doc2 = Automerge.change(doc2, d => d.tasks.push("review code"));

const merged = Automerge.merge(doc1, doc2);
// merged.tasks == ["write spec", "review code"] — 매 conflict X

G-Counter (CRDT)

class GCounter:
    def __init__(self, node_id: str):
        self.node = node_id
        self.counts: dict[str, int] = {}
    def inc(self, n: int = 1):
        self.counts[self.node] = self.counts.get(self.node, 0) + n
    def value(self) -> int:
        return sum(self.counts.values())
    def merge(self, other: "GCounter"):
        for k, v in other.counts.items():
            self.counts[k] = max(self.counts.get(k, 0), v)

Bank reconciliation pseudo

def reconcile(gl: list[Txn], bank: list[Txn]) -> dict:
    matched, gl_only, bank_only = [], [], []
    bank_idx = {(t.amount, t.date): t for t in bank}
    for g in gl:
        key = (g.amount, g.date)
        if key in bank_idx:
            matched.append((g, bank_idx.pop(key)))
        else:
            gl_only.append(g)
    bank_only = list(bank_idx.values())
    return {"matched": matched, "gl_only": gl_only, "bank_only": bank_only}

Idempotent upsert (Postgres)

INSERT INTO orders (id, amount, status)
VALUES ($1, $2, $3)
ON CONFLICT (id) DO UPDATE SET
  amount = EXCLUDED.amount,
  status = EXCLUDED.status,
  updated_at = NOW();

CDC with Debezium (Kafka Connect)

{
  "name": "pg-source",
  "config": {
    "connector.class": "io.debezium.connector.postgresql.PostgresConnector",
    "database.hostname": "db", "database.dbname": "app",
    "plugin.name": "pgoutput",
    "topic.prefix": "app",
    "table.include.list": "public.orders,public.payments"
  }
}

매 결정 기준

상황	Approach
Real-time collab text	Yjs (mature, fast)
Offline-first app	Automerge (rich JSON)
Counter across regions	PN-Counter / G-Counter
Centralized server OK	OT (simpler) or CRDT
Bank/finance	Manual + AI-assist (BlackLine)
DB replica sync	CDC (Debezium) + idempotency

기본값: Yjs 의 collab text, Automerge 의 structured offline-first, CDC 의 enterprise data sync.

🔗 Graph

• 부모: Distributed-Systems · Data-Engineering
• 변형: CRDT
• 응용: Local-First
• Adjacent: Eventual Consistency · CDC · Idempotency

🤖 LLM 활용

언제: 매 reconciliation rule 의 fuzzy match (description, amount tolerance), exception explanation, audit trail summary. 언제 X: deterministic match (use SQL JOIN), regulated final attestation (human sign-off).

❌ 안티패턴

• OT 의 P2P 의 force: 매 OT 의 server 의 require — P2P 의 CRDT 의 use.
• LWW 의 collab text: 매 char-level loss — RGA / Yjs 의 use.
• No vector clock 의 causal: 매 wall-clock skew → wrong order.
• No idempotency key: retry 의 duplicate → double charge.
• Reconcile manually 매 day: automate w/ exception queue.

🧪 검증 / 중복

• Verified (Yjs docs, Automerge paper, Shapiro CRDT survey 2011, Debezium docs).
• 신뢰도 A.

🕓 Changelog

날짜	변경
2026-05-08	Phase 1
2026-05-10	Manual cleanup — finance + CRDT + data sync unified