Exploring a DAG-based Layer-1 with EVM compatibility — looking for technical feedback

I’m part of a small builder-led community that’s been experimenting with a DAG-based Layer-1 design focused on parallel execution and developer compatibility.

The project (called PYRAX) is intentionally pre-presale. The focus so far has been on architecture, testing, and understanding tradeoffs rather than launching anything.

High level design points:
• DAG-based transaction graph (parallel execution vs linear blocks)
• EVM-compatible contracts to lower developer friction
• AI-assisted tooling used for network analysis and observability (not governance or consensus)

We’ve been stress-testing execution behavior and failure modes rather than optimizing for marketing benchmarks. Under controlled tests, throughput has approached ~100K TPS, but the more interesting work has been around how the system behaves under contention.

Posting here mainly to get feedback from folks who’ve worked with DAGs or large distributed systems:
• What tradeoffs have you seen combining DAG execution with EVM semantics?
• Where do DAG-based designs tend to break in practice?
• Does AI-assisted observability actually help at scale, or just add complexity?