fix: parse Inspur PCIe CPU affinity

bible-local/10-decisions.md

@@ -2020,3 +2020,28 @@ couple a standard inventory document to one viewer.
   `reanimator.json`.
 - Fan RPM animation, PSU live load, and other Bee live-only presentation are
   intentionally out of scope.
+
+---
+
+## ADL-069 — Inspur physical CPU locator as PCIe affinity
+
+**Date:** 2026-09-14
+**Context:** Some Inspur/Kaytus `onekeylog` archives do not contain host sysfs
+NUMA data, but their RESTful PCIe inventory identifies the physical processor
+in `location` or `DeviceLocator` values such as `#CPU0_PE2_P7_R5_SL0`.
+
+**Decision:** Parse the bounded `CPU<n>` token from those locator fields into
+the nullable `numa_node`/CPU-affinity field for PCIe devices and network
+adapters. Combined `component.log` is also a valid PCIe inventory source when
+`devicefrusdr.log` is absent. RESTful PCIe entries use their domain-qualified
+BDF as canonical `slot`, retaining the physical locator as description, so
+card/function records merge with the NIC inventory. Locator values without an
+isolated `CPU<n>` token remain unknown (`nil`); no affinity is inferred from
+BDF numbering.
+
+**Consequences:**
+- Inspur physical PCIe-to-socket wiring can drive the standalone topology even
+  for BMC-only dumps.
+- `CPU0` remains distinct from missing affinity.
+- On systems with Sub-NUMA Clustering, the value represents physical CPU
+  affinity rather than an operating-system NUMA-domain ID.