[KeuqS]

Summary

Fixes #29.

This updates the Tesla T4 examples in docs/results-schema.md to match the canonical values in profiling/gpu_specs_db.py.

Changes

• Correct the T4 memory bandwidth, FP32, FP16, and BF16 peak values.
• Add specs_source and specs_note to both schema examples.
• Document that peak values must come from the GPU specification table and must not be hardcoded or inferred from compute capability alone.

Validation

• Compared the documentation examples with the T4 entry in profiling/gpu_specs_db.py and the current T4 result metadata.
• Searched docs/results-schema.md for the stale 900/13.0/26.0 T4 figures; no stale sample values remain.
• Documentation-only change; no runtime tests are required.

[Lincoln-developer]

Thanks for this.

One thing to fix before merge. The summary block in the metadata example, lines 58 to 68, is labelled Tesla T4 but the numbers in it are the A100's:

field	now	should be
avg_ssm_recurrent_latency_ms	319.4	388.312
avg_ssm_scan_latency_ms	413.0	497.913
avg_attention_latency_ms	0.23	3.247
avg_ssm_bandwidth_utilization_pct	0.06	0.2069
avg_attention_bandwidth_utilization_pct	0.82	2.4749
avg_ssm_compute_utilization_pct	0.01	0.02922
avg_attention_compute_utilization_pct	5.1	19.489

They come from profiling/results/t4-colab-2026-10-10/metadata.json. Swap those in and this is good to go.

[Lincoln-developer]

Happy to re-review as soon as the summary block is updated and kindly don't forget to update your branch.