Add slowdown temperature exceedance detector to benchmark

detectSlowdownTempExceedance scans steady-state metric rows per GPU and
emits a [WARNING] note + PARTIAL status if any sample >= SlowdownTempC.
Uses per-GPU threshold from nvidia-smi -q, fallback 80°C.

Distinct from p95-based TempHeadroomC check: catches even a single spike
above the slowdown threshold that would be smoothed out in aggregates.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

audit/internal/platform/benchmark.go

@@ -520,6 +520,13 @@ func (s *System) RunNvidiaBenchmark(ctx context.Context, baseDir string, opts Nv
 				gpuResult.Notes = append(gpuResult.Notes,
 					fmt.Sprintf("[HARD STOP] GPU %d: %s", idx, anomaly))
 			}
+			if warn := detectSlowdownTempExceedance(metricRows, idx, gpuResult.SlowdownTempC); warn != "" {
+				gpuResult.Notes = append(gpuResult.Notes,
+					fmt.Sprintf("[WARNING] GPU %d: %s", idx, warn))
+				if gpuResult.Status == "OK" {
+					gpuResult.Status = "PARTIAL"
+				}
+			}
 			if planErr != nil {
 				gpuResult.Status = classifySATErrorStatus(phaseLogs["mixed"], planErr)
 			} else if len(gpuResult.PrecisionFailures) > 0 {
@@ -1561,6 +1568,39 @@ func detectPowerAnomaly(rows []GPUMetricRow, gpuIndex int) string {
 	return ""
 }
 
+// detectSlowdownTempExceedance scans steady-state metric rows for a GPU and
+// returns a warning string if any temperature sample exceeded the GPU's
+// SlowdownTempC threshold. Uses fallback 80°C when SlowdownTempC is zero.
+// This is a real-time signal distinct from p95 stats — even a single spike
+// above the slowdown threshold is worth flagging.
+func detectSlowdownTempExceedance(rows []GPUMetricRow, gpuIndex int, slowdownTempC float64) string {
+	if slowdownTempC <= 0 {
+		slowdownTempC = 80
+	}
+	var maxTemp float64
+	var exceedCount int
+	for _, r := range rows {
+		if r.GPUIndex != gpuIndex {
+			continue
+		}
+		if !strings.Contains(r.Stage, "steady") {
+			continue
+		}
+		if r.TempC > maxTemp {
+			maxTemp = r.TempC
+		}
+		if r.TempC >= slowdownTempC {
+			exceedCount++
+		}
+	}
+	if exceedCount == 0 {
+		return ""
+	}
+	return fmt.Sprintf(
+		"temperature exceeded slowdown threshold (%.0f°C) in %d sample(s) during steady state — peak %.1f°C",
+		slowdownTempC, exceedCount, maxTemp)
+}
+
 func detectBenchmarkDegradationReasons(gpu BenchmarkGPUResult, normalizationStatus string) []string {
 	var reasons []string
 	runtimeUS := math.Max(1, gpu.Steady.DurationSec*1e6)