artifacts: mtcollins1 ipmitool hpm check stdout (pre-reflash, BMC 0.32), the layout fixture for slice B2

[gunbai-bot]

Verbatim ipmitool hpm check stdout from mtcollins1's BMC, taken 2026-10-02 as the pre-reflash baseline. sha256 689624b261a32784a3a7d0373b1f05cfd7a46c1cff63e1ff761736cd9184a590.

This is the only full hpm-check table captured. It shows the BMC before the 0.45.3 reflash, so it serves as the parser's layout fixture and historical evidence, not the current reading. A post-reflash capture is owed once the BMC (currently unreachable) is back.

Consumer: slice B2's plan-bound firmware readback parser (calm-lynx-884, stacked on #13230) is the named first reader. This is a declared frontier under DESIGN §3c until B2 lands.

🤖 Generated with Claude Code

[briansrls]

Reviewed exact head cd5b30dd4eb054721b8327fc7d5d87fa909cdf3e.

APPROVE. This is exactly one raw artifact and does not claim a current firmware standing or add a decoder. I independently recomputed SHA-256 over the exact committed bytes (including the leading blank line, trailing spaces, and final blank lines): 689624b261a32784a3a7d0373b1f05cfd7a46c1cff63e1ff761736cd9184a590, matching the PR body. The artifact is 983 bytes / 23 newline-terminated lines and contains the complete target-information header, five component rows, and the payload-cold-reset marker. It contains no credential or token.

The historical/layout-only classification is honest. The body and filename say pre-reflash BMC 0.32 / 2026-10-02, and the bytes themselves show BOOT and APP active version 0.32. Nothing in this PR re-keys that observation onto the present 0.45.3 controller or fills slice B's current firmware readback. A data-only landing ahead of B2 is acceptable because the named reader remains a declared frontier and this artifact constructs no semantic standing by itself.

Non-blocking B2 guardrails:

• read and digest these exact bytes; do not normalize the table before provenance is established;
• bind the source/date/subject as self-observed historical provenance rather than deriving them from the filename (the stdout itself has no timestamp or host identity);
• parse component rows as raw HPM rows first. In particular, BIOS 0.00 and BOOTFW 0.00 must not become claims about the actual BIOS/boot-firmware versions merely because the table printed them;
• require a fresh post-reflash execution, bound to the attempt plan and current controller, before populating FirmwareReadBeforeActuation.

No data-artifact blocker remains. Land after ordinary exact-head CI is green.