benchmarks.md

Benchmarks

The public benchmark story is intentionally evidence-based. Benchmark changes must report the command, input set, machine class, and comparison baseline.

Current Baseline Checks

• Parser product-cutover checks compare parser output against the current product lane.
• Runtime loop checks track request-path latency for hover, definition, references, and completion.
• Fuzz checks cover parser, cascade, incremental, and transform safety targets.

Reporting Template

Command:
Inputs:
Machine:
Baseline:
Result:
Regression threshold:
Notes:

Do not treat a single synthetic benchmark as product readiness. Parser, cascade, transform, editor, and packaging paths each need their own evidence.