Systems will fail.
Engineering decides how.
I build systems that remain predictable
when assumptions break.
Most outages do not come from missing features.
They come from hidden assumptions.
My work focuses on exposing those assumptions early
and enforcing structural guarantees so faults remain contained and recovery is clear.
• make invariants explicit
• detect ambiguous state transitions
• enforce fail-fast guarantees
• define clear recovery boundaries
NEAR MPC — Threshold Cryptography Infrastructure
Peer identity depended on implicit TLS ↔ participant binding.
Without explicit enforcement, node attribution could become ambiguous across the protocol network.
near/mpc#2250
Ironclaw — AI Agent Orchestration
Plugin removal did not revoke associated credential authority.
Credential mappings could remain active after uninstall, leaving permissions inconsistent with runtime state.
Model Context Protocol — Filesystem MCP Server
Read-only mode relied on client hints rather than server enforcement.
Filesystem write tools remained callable despite read-only intent.
failure-patterns → real-world investigations
failure-lab → controlled experiments