stabilize tests

docs/benchmarks/README.md

@@ -0,0 +1,14 @@
+# Benchmarks
+
+This directory contains benchmark specs, datasets, and evaluation guidance.
+
+## Index
+- `docs/benchmarks/performance-baselines.md`
+- `docs/benchmarks/golden-corpus-kpis.md`
+- `docs/benchmarks/fidelity-metrics.md`
+- `docs/benchmarks/accuracy-metrics-framework.md`
+
+## Usage Notes
+- Benchmarks must be deterministic and offline-friendly.
+- Store fixtures alongside their benchmark docs.
+- Record expected ceilings and variance bounds in each benchmark spec.