diff --git a/README.md b/README.md
index 7143caa..7cdf1e2 100644
--- a/README.md
+++ b/README.md
@@ -82,6 +82,8 @@ See [examples](examples/README.md) for details.
 
 Measured on 4 SM100 GPUs for prefill and 64 for decode — each within one scale-up domain. To reproduce the benchmarks, see [run_sweep.sh](examples/sweep/run_sweep.sh).
 
+These are offline engine sweeps, not end-to-end serving measurements. Prefill throughput counts only non-cached input tokens. The decode plot measures steps per second, then converts both axes to tokens per second using an assumed draft acceptance of 75% with `nextn=3` (`1 + 3 × 0.75 = 3.25` tokens per step per request). The 75% value is supplied to the plotting script, not inferred from the sweep. See [benchmark methodology](examples/README.md#benchmark) for the measurement windows and token accounting.
+
 ## Evaluation benchmarks
 
 See [eval/README.md](eval/README.md).