Code like an accountant: Designing data systems for accuracy, resilience and auditability
summary
Claims that double-entry invariants catch bugs, that item-level reconciliation and coarser cash reconciliation trade cost for signal, and that append-only storage plus per-transaction metadata make audits tractable. Shows tools and patterns: source-of-truth ETL, Kafka/Spark or log-based stores, Braintree reconciliation, and admin tooling for safe data fixes. Gives a real failure mode—Verizon overcharging ~15 million customers costing $30–90M—and describes incident response composition: engineering plus finance and support, and when refunds or credits are used.
AI-written summary · about
chapters (22)
- 0:00Start
- 3:55The History of Accounting
- 6:02Principle of Double Entry Bookkeeping
- 6:26The Principle of Double Entry
- 7:51Strategies for Monitoring Data
- 8:23Strategies for Monitoring Data
- 8:42Double Entry Bookkeeping
- 9:38Simplified Eventbrite Ticket Data Model
- 12:04Data Reconciliation
- 12:14Item Level Reconciliation
- 13:28Cash Reconciliation
- 14:24Cash Reconciliation
- 15:04Materiality
- 16:51Monitoring Systems Detected Data Inconsistency
- 17:29Who Should Be on Your Incident Response Teams
- 18:41Tools To Make Data Management Easy
- 19:31What Happens When Your Payment System Times Out from the Payment Gateway
- 21:49Bugs
- 24:24The Questions That Auditors Ask
- 25:12Append-Only Data Storage
- 25:54Automatic Backups
- 26:43Financial Life Cycle
related talks
52mHow to Account Money Correctly in a Database
39mData Corruption: Stop the Evil Tribbles
37mBuilding a powerful Double Entry Accounting system
52mTransactions: Myths, Surprises and Opportunities
48mFault Tolerance on the Cheap Making Systems That Probably Won't Fall Over
35mApplying Data Quality Monitoring
37mEngineering data quality
31mZero to data infrastructure