Quality dashboard
Latest published snapshot: 2026-10-09T06:53:56.457282+00:00, commit
c2b07977, .NET SDK
10.0.401 on
Linux/x86_64. Source state:
clean commit.
Conformance
Complete programs run through three real hosts: emitted executable (oracle),
bare gsc emit-to-memory, and gsi script mode. Results compare exit code,
stdout, diagnostic IDs, and stderr byte-for-byte after line-ending
normalization.
| Metric | Current |
|---|---|
| Checked-in single-file golden programs | 144 |
| Multi-file emitted golden programs | 1 |
| Deterministic randomized programs | 32 |
| Programs compared across all drivers | 176 |
| Driver executions represented by that matrix | 528 |
| Scoped ILVerify suppressions | 2 |
Every pull request runs the conformance tests and ILVerify gate. Random cases
use fixed seeds so failures reproduce locally; expanding CaseCount grows the
standing corpus without changing the oracle.
Benchmarks
| Metric | Median |
|---|---|
gsc /help process startup | 59.81 ms |
| Compile four-program reference corpus | 3574.21 ms |
| G# steady-state workload | 59.41 ms |
| Equivalent C# steady-state workload | 62.7 ms |
| G#/C# runtime ratio | 0.948× |
Reference corpus:
Arithmetic.gs, Class.gs, PatternSwitch.gs, AsyncTask.gs.
Runtime comparison executes 25,000,000
integer-loop iterations and verifies both programs produce checksum
1199999814.
These are trend signals, not laboratory claims. GitHub-hosted runner load and process startup vary. Compare snapshots from the same OS/architecture and investigate sustained movement, not a single run.
Concurrency
The ten-pattern Go/G# comparison also presents a separately sourced concurrency-bench.yml snapshot with NativeAOT, JIT, Go, and full provenance. It is not the same measurement run as this dashboard's data.
No concurrency measurements in this snapshot. The paired benchmark runs nightly; a snapshot generated without it shows nothing here rather than a stale number.