DAGs & Event Sourcing
Video & Resource Study Guide

The canonical talk. Greg Young coined CQRS. Traces the concept back to double-entry accounting ledgers — events are older than computers. Many companies require new hires to watch this. Includes the iconic slide of a ledger entry being erased (you can’t — that’s why events are immutable).

Fowler breaks down four distinct patterns that people conflate under “event-driven”: Event Notification, Event-Carried State Transfer, Event Sourcing, and CQRS. Essential for knowing which pattern solves which problem. Without this vocabulary, you’ll conflate things that should be kept separate.

THE foundational talk on functional core / imperative shell. Pure functions inside, I/O at the edges. This is literally what your core/ vs adapters/ separation implements. Bernhardt explains why this structure makes testing trivial, concurrency safe, and code changeable. This talk will feel like someone describing your repo’s architecture before it existed.

More technical depth than talk #1. Covers trade-offs, misconceptions, read models, object databases, accidental complexity, service buses, and the relationship between event sourcing and DDD. Watch after #1 to go deeper on the mechanics.

Visual walkthrough of every graph algorithm. The topological sort, DAG shortest/longest path, and cycle detection sections are exactly what you need for your product graph. Skip to those chapters. Includes working Java source code and slides.

Source code: github.com/williamfiset/Algorithms

Teaches every anti-pattern by building the worst possible system. You learn what NOT to do — often more valuable than best practices. Demonstrates that event sourcing is “not complex, it’s different.”

Live-codes an event store from scratch using PostgreSQL. Shows the table design, append operations, and projection mechanics. Directly relevant to your Supabase approach.

Code: github.com/oskardudycz/EventStoreInOneHour

Shows the practical migration path from a classical CRUD monolith to CQRS and event sourcing. Key insight: you do NOT need multiple databases or event sourcing from day one. Start with CQRS on a single database and add event sourcing later.

“Make illegal states unrepresentable” — use the type system to encode business rules so incorrect code cannot compile. Maps directly to your models.py and specs.py. No F# knowledge needed.

All talks: fsharpforfunandprofit.com/ddd

Retrospective from the person who coined CQRS — what worked, what didn’t, how thinking evolved over 10 years. Essential for avoiding mistakes the community has already learned from.

Search his channel for: “Topological Sort”, “Graph Traversals BFS DFS”. Exceptionally clear hand-drawn diagrams, slower-paced, deeper explanations. 500K+ students. Universally recommended on Reddit and LeetCode for graph algorithm fundamentals.

The gold standard academic treatment. Proves that a DAG has no back edges in DFS and that reverse finishing order gives a valid topological ordering. If you want to understand why the algorithm is correct, not just how it works, this is the lecture. Also covers edge classification: tree edges, back edges, forward edges, cross edges.

OCW: ocw.mit.edu — Lecture 14

Explains Git bottom-up as a DAG data structure. Commits are nodes, parent pointers are edges. Branches are just references into the DAG. Merging creates nodes with two parents. The best real-world DAG example — a system you already use daily.

Course: missing.csail.mit.edu — Version Control

LeetCode 207 and 210 — the classic DAG problems. “Is this graph a DAG?” (cycle detection) and “give me a topological ordering.” Clean Python walkthroughs. These are two sides of the same coin: if you can detect cycles, you can topologically sort, and vice versa.

Course: neetcode.io — Topological Sort

Key videos: “Topological Sort using DFS”, “Kahn’s Algorithm”, “Shortest Path in DAG.” The shortest path video demonstrates using topological sort to find shortest paths in O(V+E) — the practical application most tutorials skip. Code in Python, Java, C++.

Website: takeuforward.org — Graph Series

The single best ongoing YouTube channel for event sourcing + CQRS. Key videos to watch:

• “CRUD-Sourcing is Why Your Event Streams Are Bloated” — the #1 anti-pattern. If you model events as NameChanged instead of CustomerRelocated, your streams become meaningless noise.
• “Event Sourcing Do’s and Don’ts” — practical guardrails
• “Projections in Event Sourcing: Build ANY Model You Want!” — the read-model mechanism that makes ES practical for business systems
• “Event Sourcing vs Event Driven Architecture” — they are NOT the same thing
• “Greg Young Answers Your Event Sourcing Questions” — clears up common confusion

Key insight: “Query less; Event more.” Shows how to handle distributed system uncertainty using DDD + reactive patterns. Extends aggregate design into distributed contexts. Also read Vernon’s “Effective Aggregate Design” 3-part PDF series — the definitive guide.

The DDD originator explains bounded context types (bubble context, quaint context, patch-on-patch) and refines the vocabulary of DDD itself. Essential for your ERP domain modeling — knowing where to draw the boundaries between your tools.

Concise presentation from the Chief Scientist at Thoughtworks. Connects event sourcing to broader enterprise patterns from a higher architectural perspective. Good for zooming out after the deep dives.

Three pure functions: decide(command, state) → events, evolve(state, event) → state, and initialState. This IS your decision_logic.py pattern formalized. Both functions are pure — same inputs always produce same outputs.

Blog: thinkbeforecoding.com — Functional Event Sourcing Decider

How to model products, BOM hierarchies, and shared components as graphs. Nodes = entities (products, parts, customers), edges = relationships (CONTAINS, DEPENDS_ON). The US Army, Caterpillar, and Lockheed Martin use graph databases for BOM management. Your DAG product model is this pattern on Postgres.

Free course: graphacademy.neo4j.com — Modeling Fundamentals

BOM example: neo4j.com — Top 10 BOM Use Cases