NVIDIA / NVIDIA/ai-cloud-validation

provider_hidden allows vacuous passes on hardware-telemetry checks — decide legitimacy per plane

Open
#552 0 comments 0 reactions 1 assignee View on GitHub

@abegnoche is already working on this.

Since Jul 21, 2026.

Dominant language
Python
Stars
27
Forks
33
Avg merge
1d 15h
Merged PRs (30d)
28

Description

Problem

Several validations accept provider_hidden=true evidence and pass without verifying anything. The AWS reference passes ~15 observability checks and the security BMC checks this way. As written, a provider can be certified on NVSwitch/NVLink/BMC telemetry by asserting the plane is provider-owned — the check verifies nothing.

Found during the per-check requires: inventory (2026-07-21) tracing every module-suite check to its stub scripts and validation contract.

Affected checks

Observability — always/AWS provider-hidden:

  • OBS07-01 EastWestNetworkTelemetryCheck, OBS08-01 ManagementNetworkTelemetryCheck, OBS09-01 NvswitchFabricTelemetryCheck
  • OBS17-01 BmcSelLogsCheck, TELEM04-01 BmcGpuTelemetryCheck, TELEM05-01 StorageCapacityTelemetryCheck
  • TELEM07-01 GpuNvlinkTelemetryCheck, TELEM08-01 SwitchNvlinkTelemetryCheck (hidden even in NICo)
  • OBS14-01 / OBS15-01 / OBS16-01 switch logs (hidden in both AWS and NICo)
  • OBS11/12/13 fabric-manager/SM/UFM logs pass hidden on AWS (real NICo UFM implementations exist)

Security:

  • CNP10-01 BmcProtocolSecurityCheck emits a fixed "no customer BMC surface" pass
  • SEC12-01/02/03 BMC checks pass provider-hidden on non-BM providers

Decision needed

Is "plane is provider-owned/hidden" a legitimate certification pass? Options:

  1. Legitimate, but visible — keep the escape, record provider_hidden as a distinct result status ("attested hidden" vs "verified") in upload/reporting so reviewers can distinguish.
  2. Hole — require real evidence; these checks become bare_metal/GPU-fabric-shaped (requires) and the escape is removed.
  3. Per-plane hybrid — allowed for planes genuinely provider-owned in a cloud model (switch logs), disallowed where the declared capability implies tenant visibility (e.g. BMC/SEL and NVLink telemetry must be verified when bare_metal is declared).

Interplay with capability declarations: whatever the answer, it should be consistent with the declared-capability obligation model — a bare_metal-declaring provider passing BMC checks via "hidden" is the case that most looks like a hole.

🤖 Generated with Claude Code

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Assessment

This issue has not been assessed yet.

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.