メインコンテンツまでスキップ

Data Protection & Privacy

This guide explains how github.com/contract-to-cash/core handles personal data and what responsibilities fall to you as the integrator. It is guidance, not legal advice — your data-protection obligations (GDPR, Japan's APPI, CCPA, etc.) depend on your jurisdiction and use case.

The short version: this library is designed so that personal data never has to enter the event store. The core identifies parties by opaque IDs and leaves personal data in systems you own.

What personal data flows through the library​

The domain works almost entirely in opaque identifiers and money, not personal data:

  • AccountID, ContractID, InvoiceID, PaymentID, PriceID, ProductID, BalanceEntryID — ULIDs / opaque strings with no personal data encoded in them.
  • shared.Money amounts, currencies, dates, statuses.
  • Payment idempotency keys and gateway transaction IDs (integrator-supplied / gateway-supplied references).

An AccountID is an opaque reference to a customer, not the customer's data. The library never asks for a name, email, address, or payment instrument — those live in the gateway and in your own customer store.

Where PII could leak in if you are not careful:

  • Free-form metadata map[string]interface{} fields on entities (e.g. payment metadata). Do not put names, emails, or card data here.
  • Consumer-supplied idempotency keys, if you derive them from personal data. Derive them from opaque IDs instead.
  • Coupon codes or invoice line-item descriptions, if you populate them with personal data.

Recommendation: keep direct PII (names, emails, addresses, tax IDs, payment instruments) out of events, metadata, and line items. Store it in a consumer-owned system keyed by AccountID, and join at the presentation layer.

Erasure vs. an append-only event store​

Event sourcing is append-only: events are immutable, and contract state is reconstructed by replaying them. This is what gives you a full audit trail — but it is in direct tension with a "right to erasure" request that expects data to be deleted.

There are two standard patterns. This library is built for the first and does not implement the second.

Pattern 1 (primary recommendation): keep PII outside events​

If personal data never enters the event store, an erasure request is satisfied by deleting the record in your consumer-owned customer store keyed by AccountID. The event history then contains only an opaque ID that no longer resolves to a person — effectively anonymized — while your billing audit trail stays intact.

This is the intended design of the library (opaque AccountID references), and it is the pattern we recommend building around from day one.

Pattern 2 (fallback): crypto-shredding​

If you have already persisted PII inside events (or cannot avoid it), the standard fallback is crypto-shredding: encrypt per-subject PII with a per-subject key before it is written into an event, and erase by destroying that key. Once the key is gone the ciphertext is unrecoverable, which satisfies erasure without rewriting the append-only log.

The library does not implement crypto-shredding. There is no built-in encryption, key registry, or key-destruction hook. If you adopt this pattern you own key generation, storage, rotation, and destruction entirely — encrypt before handing data to the library and decrypt after reading it back. Treat this as a fallback for data that should not have been in events in the first place, not as a substitute for Pattern 1.

Personal data in logs​

The library logs through the standard-library log/slog. Services accept a logger (e.g. WithBillingLogger) and default to slog.Default().

What the library actually logs (audited from the service, batch, and webhook code):

  • Entity IDs — contractID, invoiceID, paymentID, accountID, balanceEntryID, event_id, hook names.
  • Monetary amounts — e.g. refundAmount, and forfeitedAmount + currency when a balance entry expires (batch/balance_expiration.go).
  • Idempotency keys — originalKey / effectiveKey are logged in a few payment reconciliation paths (application/service/payment_service.go).
  • Errors from gateway calls, hook failures, and save conflicts.

So logs contain account IDs and monetary amounts (and, in the payment paths, idempotency keys). They do not contain names, emails, addresses, or card data — because the library never receives those. The library also does not log at Debug level by default; the messages above are Info / Warn / Error.

Even opaque IDs plus amounts can be sensitive in aggregate (they reveal that an account was billed a given amount at a given time). If that matters for your jurisdiction:

  • Redact via your slog handler. Route the library's logger through a custom slog.Handler that drops or hashes attributes such as accountID, refundAmount, originalKey, and effectiveKey before they reach your log sink. Because the library emits structured key/value attributes, an attribute-filtering handler can redact them reliably without parsing message strings.
  • Control verbosity. Pass a logger with a higher minimum level if you want to suppress the Info-level reconciliation notes.

Data retention responsibilities​

This is a BYO-DB library: it defines interfaces and in-memory implementations only. Every durable store is consumer-owned, and so is its retention, encryption-at-rest, access control, and deletion policy:

  • Event store — your append-only log (eventstore.Store implementation). Its retention drives how long history is reconstructable.
  • Snapshots — periodic aggregate snapshots (SnapshotService, DefaultSnapshotInterval). These embed reconstructed state, so treat them with the same retention/erasure rules as events.
  • Webhook dead-letter queue (DLQ) — failed webhook events you persist for later retry. May contain gateway payloads; apply retention and access controls.
  • Idempotency store — payment idempotency keys and webhook dedup markers (IsDuplicate / MarkProcessed, default DeduplicationTTL 72h). Use TTLs long enough to cover gateway redelivery windows but no longer than you need.

The library holds none of this in memory beyond a single operation (except the in-memory implementations, which are for tests and demos only). Backup, replication, residency, and deletion are entirely under your control.