Research investigation R1021 / claim audit
llm-d 1.0’s passing TPU badges do not validate every documented TPU path
The sampled records distinguish release-validation lanes, checked passing badges and documented configurations, but do not establish exact v1.0.0 runtime validation for every path—or validation of routing, offload and disaggregation together.
This is a bounded audit of three guides, not every llm-d deployment or feature cross-product. It excludes CPU/NPU configurations, baseline-only serving, wide expert parallelism and multimodal guides. Passing statuses were checked only for the supplied GKE GPU/TPU badges. The saved Google announcement snapshot carries a retrieval time of October 21 at 20:31:58.905 UTC, later than the stated cutoff; it is not treated as pre-cutoff test verification.
This is a bounded audit of three guides, not every llm-d deployment or feature cross-product. It excludes CPU/NPU configurations, baseline-only serving, wide expert parallelism and multimodal guides. Passing statuses were checked only for the supplied GKE GPU/TPU badges. The saved Google announcement snapshot carries a retrieval time of October 21 at 20:31:58.905 UTC, later than the stated cutoff; it is not treated as pre-cutoff test verification.
- Dataset ID
- spd:llm-d-1-0-which-tpu-and-gpu-serving-paths-have-published-test-evidence-95e8db12
- Stable URL
- /research/llm-d-1-0-which-tpu-and-gpu-serving-paths-have-published-test-evidence-95e8db12
- Version
- Live
- Coverage
- Live collection
- Records
- 0
- Fields
- 7
- Updated
| Entity | Metric | Value | Unit | Observed | Source | Transform |
|---|
Measurement technique
How to read this report
- 01Reassess the supplied saved evidence only. Preserve the original October 21, 2026, 20:30 UTC evidence cutoff and collection record of October 21, 2026, 20:30:57.479 UTC; this synthesis is not a new collection, experiment or calculation.
- 02Bound the sample to TPU/GPU configurations in the v1.0.0-tagged precise prefix-cache routing, tiered prefix-cache and prefill/decode disaggregation guides, together with their release-matrix entries.
- 03Use the supplied Google and IBM announcements as attributed launch context, and project/router release notes as version context—not as independent deployment validation.
- 04Plan an evidence matrix with accelerator and generation, inference engine, serving technique, connector or backend, guide version, release lane, checked result and provenance limits. Keep release-lane existence separate from a checked passing result.
- 05Classify evidence as checked passing release-branch status, published lane with outcome not checked, explicit manual-validation statement, documented without a corresponding sampled release lane, or indeterminate exact-version/combined-feature validation. Do not calculate a tested-share percentage from the supplied grouped records.
Sources
Evidence
4 publishers supporting 0 records. Expand a publisher to inspect its cited pages.