Runtime performance baseline
The new language contracts have benchmarks for a typed record/function workload and IPv6 CIDR subtraction. Run them on the target machine:
go test ./core -run '^$' -bench 'Benchmark(Typed|IPv6)' -benchmemAn October 14, 2026 Windows amd64 development run on an AMD Ryzen 5 7535HS showed:
| Workload | Time/op | Bytes/op | Allocations/op |
|---|---|---|---|
| Prepared typed script, including fresh interpreter and 100 checked calls | 318 microseconds | 192,841 | 2,007 |
| Parse/prepare plus the same execution | 697 microseconds | 240,099 | 2,308 |
| IPv6 /0 minus one /128 | 28 microseconds | 35,113 | 521 |
These are short local measurements taken during development, not production latency guarantees or comparisons with other languages. Prepared execution avoids repeated parsing; it does not change the tree-walk evaluation model.
Profile the representative application before choosing a bytecode VM. Current benchmarks make allocation costs visible; interpreter initialization, environment maps and temporary Values are candidates for measured optimization. A VM, JIT, cross-language speed claim, and production throughput targets remain future work.