[KeuqS]

Summary

• Mark each README metric as Measured or Planned.
• Rename the latency row to reflect the one full forward pass that the runner actually times.
• Clarify that prefill/decode split, throughput, peak memory, and energy-per-token are not implemented.

Validation

• Cross-checked the metric names and outputs with profiling/benchmark_runner.py.
• Documentation-only change; no benchmark execution is required.

Closes #44

[Lincoln-developer]

Thank you for submitting this pull request.

[Lincoln-developer]

One note that is not a blocker. The branch is behind main, which has moved since you opened this. Updating the branch would re-run CI against current main and we merge this successfully.