Decisions and lessons from building nigo-protocol
What happens when a validator moves to the next round and discards finality evidence that arrives late? A reproduction and fix in nigo-protocol show why casting new votes and learning an existing decision are different operations.
What must a consensus node save so it remembers its own votes after a restart? A look at commit, durability confirmation, and signature broadcast in nigo-protocol's local safety records.
A responding server, a running consensus engine, and finalized transactions require different evidence. How nigo-protocol distinguishes normal idle time, delayed progress, and missing observations.
Block execution took 18% less time, but the full local processing path improved by only 2.4%; nigo-protocol's benchmark records show how validation costs, transaction dependencies, and measurement order change the performance comparison.
How a DAG represents the waiting required between transactions, how WAVE, READY_QUEUE, and WORK_FIRST_CONTINUATION schedule the same dependencies, and where these strategies apply to nigo-protocol's Direct Cell and Native Program paths.
A split-and-join example with 100 units of an asset explains how transaction independence, isolated execution, and ordered result application produce the same ledger as sequential execution.
Follow a flow verified with the same token contract and ethers, viem, and Web3j, from deployment to transaction results, to see which code and tools can be reused in nigo-protocol and which integration settings need adjustment.
Starting with verification data shared by several states, nigo-protocol's load tests show why stopping safely and completing automatic cleanup are different outcomes.
An allowance of 10 illustrates state preservation across revisions and RESET on program replacement, and why isolating records and preserving their meaning are different guarantees.
Follow a token transfer function from EVM state reads and calculations to StateCell changes, then examine how its execution responsibilities differ from Native Program validation and where parallel execution stops.