From bf182daa8997ae4f3c2b9441a77f7b04a08f8c02 Mon Sep 17 00:00:00 2001
From: Michael Chus <mike@mchus.pro>
Date: Mon, 13 Apr 2026 23:43:12 +0300
Subject: [PATCH] Fix benchmark report methodology and rebuild gpu burn worker
 on toolchain changes

---
 audit/internal/platform/benchmark_report.go | 15 ++++++---------
 iso/builder/build.sh                        | 14 +++++++++++++-
 2 files changed, 19 insertions(+), 10 deletions(-)

diff --git a/audit/internal/platform/benchmark_report.go b/audit/internal/platform/benchmark_report.go
index 614846b..e07a8d6 100644
--- a/audit/internal/platform/benchmark_report.go
+++ b/audit/internal/platform/benchmark_report.go
@@ -81,8 +81,12 @@ func renderBenchmarkReportWithCharts(result NvidiaBenchmarkResult) string {
 		b.WriteString("\n")
 	}
 
-	// ── Scoring methodology ───────────────────────────────────────────────────
-	b.WriteString("## Scoring Methodology\n\n")
+	// ── Methodology ───────────────────────────────────────────────────────────
+	b.WriteString("## Methodology\n\n")
+	fmt.Fprintf(&b, "- Profile `%s` uses standardized baseline -> warmup -> steady-state -> interconnect -> cooldown phases.\n", result.BenchmarkProfile)
+	b.WriteString("- Single-GPU compute score comes from `bee-gpu-burn` on the cuBLASLt path when available.\n")
+	b.WriteString("- Thermal and power limits are inferred from NVIDIA clock-event counters plus sustained telemetry.\n")
+	b.WriteString("- `result.json` is the canonical machine-readable source for the run.\n\n")
 	b.WriteString("**Compute score** is derived from two phases:\n\n")
 	b.WriteString("- **Synthetic** — each precision type (fp8, fp16, fp32, fp64, fp4) runs alone for a dedicated window. ")
 	b.WriteString("Measures peak throughput with the full GPU dedicated to one kernel type. ")
@@ -286,13 +290,6 @@ func renderBenchmarkReportWithCharts(result NvidiaBenchmarkResult) string {
 		}
 	}
 
-	// ── Methodology ───────────────────────────────────────────────────────────
-	b.WriteString("## Methodology\n\n")
-	fmt.Fprintf(&b, "- Profile `%s` uses standardized baseline → warmup → steady-state → interconnect → cooldown phases.\n", result.BenchmarkProfile)
-	b.WriteString("- Single-GPU compute score from bee-gpu-burn cuBLASLt when available.\n")
-	b.WriteString("- Thermal and power limitations inferred from NVIDIA clock event reason counters and sustained telemetry.\n")
-	b.WriteString("- `result.json` is the canonical machine-readable source for this benchmark run.\n\n")
-
 	// ── Raw files ─────────────────────────────────────────────────────────────
 	b.WriteString("## Raw Files\n\n")
 	b.WriteString("- `result.json`\n- `report.md`\n- `summary.txt`\n- `verbose.log`\n")
diff --git a/iso/builder/build.sh b/iso/builder/build.sh
index a2160e3..ac89fcf 100755
--- a/iso/builder/build.sh
+++ b/iso/builder/build.sh
@@ -874,8 +874,20 @@ if [ "$BEE_GPU_VENDOR" = "nvidia" ]; then
     CUBLAS_CACHE="${DIST_DIR}/cublas-${CUBLAS_VERSION}+cuda${NCCL_CUDA_VERSION}"
 
     GPU_STRESS_NEED_BUILD=1
-    if [ -f "$GPU_BURN_WORKER_BIN" ] && [ "${BUILDER_DIR}/bee-gpu-stress.c" -ot "$GPU_BURN_WORKER_BIN" ]; then
+    if [ -f "$GPU_BURN_WORKER_BIN" ]; then
         GPU_STRESS_NEED_BUILD=0
+        for dep in \
+            "${BUILDER_DIR}/bee-gpu-stress.c" \
+            "${BUILDER_DIR}/VERSIONS"; do
+            if [ "$dep" -nt "$GPU_BURN_WORKER_BIN" ]; then
+                GPU_STRESS_NEED_BUILD=1
+                break
+            fi
+        done
+        if [ "$GPU_STRESS_NEED_BUILD" = "0" ] && \
+            find "${CUBLAS_CACHE}/include" "${CUBLAS_CACHE}/lib" -type f -newer "$GPU_BURN_WORKER_BIN" | grep -q .; then
+            GPU_STRESS_NEED_BUILD=1
+        fi
     fi
 
     if [ "$GPU_STRESS_NEED_BUILD" = "1" ]; then