Notes on how real systems break — and how those failures can be contained.
Most investigations start with a simple question:
What assumption is this system relying on — and what happens if it breaks?
In practice, failures usually begin with small violations of invariants:
- hidden assumptions
- ambiguous state transitions
- adversarial or malformed inputs
- edge conditions under load
This repository records those investigations.
Each note follows roughly the same structure.
The purpose is simple: document these failures so they are not repeated, and make the underlying failure modes visible before they surface as incidents.
Examples come from different kinds of systems:
- distributed infrastructure
- cryptographic protocols
- permission and identity systems
- application lifecycle boundaries
- AI model behavior and evaluation
Despite their differences, many failures follow similar structural patterns.