Determinism before parallelism
When parallel execution caused divergent state roots, I made the execution path sequential by construction to restore reproducible replay.
Case 05 · Rust · EVM Layer 1
YouthChain
Four validators must reach the same state after every transaction.
I work on an EVM-compatible Layer 1 written in Rust, spanning DAG-based BFT consensus, BLS signatures, deterministic execution and the infrastructure that lets applications use the chain.
Result
Implemented and exercised the consensus, EVM execution and engagement pipeline on a four-validator testnet, including validator disruption and recovery. Production-readiness work remains separate.
My part
What I did
Validators must agree on more than the next block: they must execute the same history into the same state. Divergent replay can fork a network even when its consensus messages appear healthy.
01
Consensus engine
Co-designed and implemented DAG-based BFT consensus with BLS signature aggregation.
02
Execution correctness
Traced state-root mismatches to transaction execution and made replay deterministic by construction.
03
Protocol-to-app connection
Built protocol contracts, JSON-RPC integration and an off-chain sequencer with an engagement rollup.
How it worksSimplified walkthrough. No live transactions.
How a transaction becomes shared state
JSON-RPC + networking
An application submits a transaction through the chain interface. Nodes propagate the input for the consensus process.
Application → JSON-RPC → nodeApplications need a usable entry point into the protocol.Example state, not live data
- Transaction
- Received
- Consensus
- Pending
- State
- Previous state
FireDAG + BFT
Validators participate in DAG-based consensus, using BLS signatures to certify agreement.
DAG vertices + quorum certificationA local proposal is not yet an agreed chain history.Example state, not live data
- Transaction
- Ordered
- Consensus
- Certified
- State
- Previous state
Rust + EVM
Agreed transactions run through the execution layer. The deterministic path gives the same ordered history the same state transition.
Ordered transactions → EVM → state rootReplaying the same history must reproduce the same state.Example state, not live data
- Transaction
- Executed
- Consensus
- Agreed order
- State
- Computed root
State + application APIs
The resulting state becomes available to applications. Engagement batching connects mobile activity to the on-chain accounting path.
Committed state → receipts / application readsProtocol correctness must survive the boundary into the product.Example state, not live data
- Transaction
- Committed
- Consensus
- Agreed history
- State
- Queryable state
Trade-offs
Decisions I made
Exercise failure conditions
Validator disruption, recovery and state agreement are part of the testnet work; throughput alone does not establish protocol readiness.
Links and records
See the work
Architecture walkthrough
Consensus to application state
The interactive flow above describes the implemented system boundary and the determinism decision.
Development evidence
Four-validator testnet exercises
Internal testnet records cover load, validator interruption, recovery and execution consistency. Private infrastructure is not linked publicly.
Based on protocol implementation and four-validator testnet records.
Development-stage testnet. Mainnet readiness and production performance are outside this case’s claims.