Add NVMe block format tool

New in chart:
- event_logs and platform_config sections in viewer
- Storage columns: logical_block_size_bytes, physical_block_size_bytes,
  metadata_bytes_per_block
- Compact status/severity icons, severity filtering for event logs
- Fixed JS MIME type and base stylesheet

bee audit schema already has all required fields; no schema changes needed.

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

audit/internal/collector/board.go

@@ -3,6 +3,7 @@ package collector
 import (
 	"bee/audit/internal/schema"
 	"bufio"
+	"context"
 	"log/slog"
 	"os"
 	"os/exec"
@@ -17,14 +18,6 @@ var execDmidecode = func(typeNum string) (string, error) {
 	return string(out), nil
 }
 
-var execIpmitool = func(args ...string) (string, error) {
-	out, err := exec.Command("ipmitool", args...).Output()
-	if err != nil {
-		return "", err
-	}
-	return string(out), nil
-}
-
 // collectBoard runs dmidecode for types 0, 1, 2 and returns the board record
 // plus the BIOS firmware entry. Any failure is logged and returns zero values.
 func collectBoard() (schema.HardwareBoard, []schema.HardwareFirmwareRecord) {
@@ -80,19 +73,23 @@ func parseBoard(type1, type2 string) schema.HardwareBoard {
 
 // collectBMCFirmware collects BMC firmware version via ipmitool mc info.
 // Returns nil if ipmitool is missing, /dev/ipmi0 is absent, or any error occurs.
-func collectBMCFirmware() []schema.HardwareFirmwareRecord {
+func collectBMCFirmware(manufacturer string) []schema.HardwareFirmwareRecord {
 	if _, err := exec.LookPath("ipmitool"); err != nil {
 		return nil
 	}
 	if _, err := os.Stat("/dev/ipmi0"); err != nil {
 		return nil
 	}
-	out, err := execIpmitool("mc", "info")
+	profile := selectIPMIProfile(manufacturer)
+	ctx, cancel := context.WithTimeout(context.Background(), profile.mcInfoTimeout)
+	defer cancel()
+	cmd := exec.CommandContext(ctx, "ipmitool", "mc", "info")
+	raw, err := cmd.Output()
 	if err != nil {
 		slog.Info("bmc: ipmitool mc info unavailable", "err", err)
 		return nil
 	}
-	version := parseBMCFirmwareRevision(out)
+	version := parseBMCFirmwareRevision(string(raw))
 	if version == "" {
 		return nil
 	}

audit/internal/collector/collector.go

@@ -23,7 +23,7 @@ func Run(_ runtimeenv.Mode) schema.HardwareIngestRequest {
 	board, biosFW := collectBoard()
 	snap.Board = board
 	snap.Firmware = append(snap.Firmware, biosFW...)
-	snap.Firmware = append(snap.Firmware, collectBMCFirmware()...)
+	snap.Firmware = append(snap.Firmware, collectBMCFirmware(derefString(snap.Board.Manufacturer))...)
 
 	snap.CPUs = collectCPUs()
 
@@ -45,7 +45,8 @@ func Run(_ runtimeenv.Mode) schema.HardwareIngestRequest {
 	snap.PCIeDevices = enrichPCIeWithRAIDTelemetry(snap.PCIeDevices)
 	snap.Storage = enrichStorageWithVROC(snap.Storage, snap.PCIeDevices)
 	snap.Storage = appendUniqueStorage(snap.Storage, collectRAIDStorage(snap.PCIeDevices))
-	snap.PowerSupplies = collectPSUs()
+	snap.VROCLicense = collectVROCLicense(snap.PCIeDevices)
+	snap.PowerSupplies = collectPSUs(derefString(snap.Board.Manufacturer))
 	snap.PowerSupplies = enrichPSUsWithTelemetry(snap.PowerSupplies, sensorDoc)
 	snap.Sensors = buildSensorsFromDoc(sensorDoc)
 	finalizeSnapshot(&snap, collectedAt)

audit/internal/collector/ipmi_profile.go

@@ -0,0 +1,92 @@
+package collector
+
+// Package-level IPMI tuning profiles.
+//
+// Each profile is matched by board manufacturer (already known before PSU
+// collection runs). The profile drives two things:
+//   - Per-command timeouts — prevents infinite hangs on slow BMCs.
+//   - FRU early-exit — streaming parser stops reading once all PSU entries
+//     are found, avoiding the tail of non-PSU FRU records.
+//
+// To add a new vendor: append to ipmiProfiles. The first matching entry wins.
+
+import (
+	"strings"
+	"time"
+)
+
+// ipmiProfile holds tuning parameters for one or more board manufacturers.
+type ipmiProfile struct {
+	// name is shown in log messages.
+	name string
+	// manufacturers is a list of lowercase substrings matched against the
+	// board manufacturer string from dmidecode type 1.
+	manufacturers []string
+	// fruTimeout is the hard deadline for the entire `ipmitool fru print`
+	// command. Zero means no timeout (not recommended).
+	fruTimeout time.Duration
+	// sdrTimeout is the hard deadline for `ipmitool sdr`.
+	sdrTimeout time.Duration
+	// mcInfoTimeout is the hard deadline for `ipmitool mc info`.
+	mcInfoTimeout time.Duration
+	// fruEarlyExit instructs the streaming FRU parser to stop reading
+	// after it has found at least one PSU entry and the current block is
+	// complete. Useful on servers with many non-PSU FRU devices.
+	fruEarlyExit bool
+}
+
+// ipmiProfiles is the ordered list of profiles. First match wins.
+var ipmiProfiles = []ipmiProfile{
+	{
+		// Lenovo XCC-based servers (ThinkSystem SR6xx / SR8xx / ST series).
+		// SR650 V3 has 54 FRU devices; each IPMI read takes ~2 s, so the
+		// full `fru print` scan takes ~108 s on a loaded BMC. Enable early
+		// exit so collection stops once PSU records are found.
+		name:          "lenovo",
+		manufacturers: []string{"lenovo"},
+		fruTimeout:    90 * time.Second,
+		sdrTimeout:    45 * time.Second,
+		mcInfoTimeout: 15 * time.Second,
+		fruEarlyExit:  true,
+	},
+	{
+		// HPE iLO-based servers (ProLiant DL/ML/BL).
+		name:          "hpe",
+		manufacturers: []string{"hp", "hewlett packard"},
+		fruTimeout:    60 * time.Second,
+		sdrTimeout:    30 * time.Second,
+		mcInfoTimeout: 10 * time.Second,
+		fruEarlyExit:  false,
+	},
+	{
+		// Dell iDRAC-based servers.
+		name:          "dell",
+		manufacturers: []string{"dell"},
+		fruTimeout:    60 * time.Second,
+		sdrTimeout:    30 * time.Second,
+		mcInfoTimeout: 10 * time.Second,
+		fruEarlyExit:  false,
+	},
+}
+
+// defaultIPMIProfile is used when no vendor profile matches.
+var defaultIPMIProfile = ipmiProfile{
+	name:          "default",
+	fruTimeout:    60 * time.Second,
+	sdrTimeout:    30 * time.Second,
+	mcInfoTimeout: 10 * time.Second,
+	fruEarlyExit:  false,
+}
+
+// selectIPMIProfile returns the profile for the given board manufacturer.
+func selectIPMIProfile(manufacturer string) ipmiProfile {
+	mfgLower := strings.ToLower(strings.TrimSpace(manufacturer))
+	for _, p := range ipmiProfiles {
+		for _, m := range p.manufacturers {
+			if strings.Contains(mfgLower, m) {
+				return p
+			}
+		}
+	}
+	return defaultIPMIProfile
+}

audit/internal/collector/psu.go

@@ -2,6 +2,8 @@ package collector
 
 import (
 	"bee/audit/internal/schema"
+	"bufio"
+	"context"
 	"log/slog"
 	"os/exec"
 	"regexp"
@@ -10,16 +12,29 @@ import (
 	"strings"
 )
 
-func collectPSUs() []schema.HardwarePowerSupply {
+func collectPSUs(manufacturer string) []schema.HardwarePowerSupply {
+	profile := selectIPMIProfile(manufacturer)
+
 	var psus []schema.HardwarePowerSupply
-	if out, err := exec.Command("ipmitool", "fru", "print").Output(); err == nil {
-		psus = parseFRU(string(out))
+	fruCtx, fruCancel := context.WithTimeout(context.Background(), profile.fruTimeout)
+	defer fruCancel()
+
+	if profile.fruEarlyExit {
+		psus = collectFRUEarlyExit(fruCtx)
 	} else {
-		slog.Info("psu: fru unavailable", "err", err)
+		cmd := exec.CommandContext(fruCtx, "ipmitool", "fru", "print")
+		if out, err := cmd.Output(); err == nil {
+			psus = parseFRU(string(out))
+		} else {
+			slog.Info("psu: fru unavailable", "err", err)
+		}
 	}
 
 	sdrData := map[int]psuSDR{}
-	if sdrOut, err := exec.Command("ipmitool", "sdr").Output(); err == nil {
+	sdrCtx, sdrCancel := context.WithTimeout(context.Background(), profile.sdrTimeout)
+	defer sdrCancel()
+	cmd := exec.CommandContext(sdrCtx, "ipmitool", "sdr")
+	if sdrOut, err := cmd.Output(); err == nil {
 		sdrData = parsePSUSDR(string(sdrOut))
 		if len(psus) == 0 {
 			psus = synthesizePSUsFromSDR(sdrData)
@@ -30,7 +45,66 @@ func collectPSUs() []schema.HardwarePowerSupply {
 		slog.Info("psu: ipmitool unavailable, skipping", "err", err)
 		return nil
 	}
-	slog.Info("psu: collected", "count", len(psus))
+	slog.Info("psu: collected", "count", len(psus), "profile", profile.name)
+	return psus
+}
+
+// collectFRUEarlyExit streams ipmitool fru print line-by-line and stops reading
+// as soon as it has found all PSU blocks and the next block is not a PSU.
+// This avoids scanning all 50+ non-PSU FRU devices on Lenovo XCC servers.
+func collectFRUEarlyExit(ctx context.Context) []schema.HardwarePowerSupply {
+	cmd := exec.CommandContext(ctx, "ipmitool", "fru", "print")
+	pipe, err := cmd.StdoutPipe()
+	if err != nil {
+		slog.Info("psu: fru pipe unavailable", "err", err)
+		return nil
+	}
+	if err := cmd.Start(); err != nil {
+		slog.Info("psu: fru start failed", "err", err)
+		return nil
+	}
+
+	var psus []schema.HardwarePowerSupply
+	var currentBlock strings.Builder
+	slot := 0
+	psuFound := false
+	stoppedEarly := false
+
+	scanner := bufio.NewScanner(pipe)
+	for scanner.Scan() {
+		line := scanner.Text()
+
+		if strings.HasPrefix(line, "FRU Device Description") {
+			if currentBlock.Len() > 0 {
+				if psu, ok := parseFRUBlock(currentBlock.String(), slot); ok {
+					psus = append(psus, psu)
+					psuFound = true
+					slot++
+				}
+				currentBlock.Reset()
+			}
+			// Stop once we've collected PSUs and hit a non-PSU block header.
+			if psuFound && !isPSUHeader(strings.ToLower(line)) {
+				stoppedEarly = true
+				break
+			}
+		}
+		currentBlock.WriteString(line)
+		currentBlock.WriteByte('\n')
+	}
+
+	if !stoppedEarly && currentBlock.Len() > 0 {
+		if psu, ok := parseFRUBlock(currentBlock.String(), slot); ok {
+			psus = append(psus, psu)
+		}
+	}
+
+	// Kill the process immediately on early exit rather than waiting for context timeout.
+	if cmd.Process != nil {
+		cmd.Process.Kill() //nolint:errcheck
+	}
+	cmd.Wait() //nolint:errcheck
+	slog.Info("psu: fru early-exit complete", "psus_found", len(psus), "stopped_early", stoppedEarly)
 	return psus
 }
 

audit/internal/collector/raid.go

@@ -733,6 +733,37 @@ func parseMDStatArrays(raw string) []mdArray {
 	return arrays
 }
 
+// collectVROCLicense runs mdadm --detail-platform and extracts the License field.
+// Returns nil when VROC is absent or the platform does not report a license.
+func collectVROCLicense(pcie []schema.HardwarePCIeDevice) *string {
+	if !hasVROCController(pcie) {
+		return nil
+	}
+	out, err := raidToolQuery("mdadm", "--detail-platform")
+	if err != nil {
+		slog.Info("vroc: mdadm --detail-platform unavailable", "err", err)
+		return nil
+	}
+	return parseMDAdmPlatformLicense(string(out))
+}
+
+func parseMDAdmPlatformLicense(raw string) *string {
+	for _, line := range strings.Split(raw, "\n") {
+		trimmed := strings.TrimSpace(line)
+		if !strings.HasPrefix(strings.ToLower(trimmed), "license") {
+			continue
+		}
+		if idx := strings.Index(trimmed, ":"); idx >= 0 {
+			val := strings.TrimSpace(trimmed[idx+1:])
+			if val != "" {
+				v := strings.ToLower(val)
+				return &v
+			}
+		}
+	}
+	return nil
+}
+
 func queryDeviceSerial(devPath string) string {
 	if out, err := exec.Command("nvme", "id-ctrl", devPath, "-o", "json").Output(); err == nil {
 		var ctrl nvmeIDCtrl

audit/internal/collector/vroc_test.go

@@ -28,6 +28,35 @@ md125 : active raid1 nvme2n1[0] nvme3n1[1]
 	}
 }
 
+func TestParseMDAdmPlatformLicense(t *testing.T) {
+	premium := `Platform : Intel(R) Virtual RAID on CPU
+Version  : 1.3.0.1138
+RAID Levels : raid0 raid1 raid5 raid10
+Total Disks : 4
+License  : Premium
+`
+	got := parseMDAdmPlatformLicense(premium)
+	if got == nil || *got != "premium" {
+		t.Fatalf("expected 'premium', got %v", got)
+	}
+
+	standard := `Platform : Intel(R) Virtual RAID on CPU
+License  : Standard
+`
+	got = parseMDAdmPlatformLicense(standard)
+	if got == nil || *got != "standard" {
+		t.Fatalf("expected 'standard', got %v", got)
+	}
+
+	noLicense := `Platform : Intel(R) Virtual RAID on CPU
+Version  : 1.0.0
+`
+	got = parseMDAdmPlatformLicense(noLicense)
+	if got != nil {
+		t.Fatalf("expected nil, got %v", *got)
+	}
+}
+
 func TestHasVROCController(t *testing.T) {
 	intel := vendorIntel
 	model := "Volume Management Device NVMe RAID Controller"

audit/internal/schema/hardware.go

@@ -66,6 +66,7 @@ type HardwareSnapshot struct {
 	PowerSupplies []HardwarePowerSupply    `json:"power_supplies,omitempty"`
 	Sensors       *HardwareSensors         `json:"sensors,omitempty"`
 	EventLogs     []HardwareEventLog       `json:"event_logs,omitempty"`
+	VROCLicense   *string                  `json:"vroc_license,omitempty"`
 }
 
 type HardwareHealthSummary struct {

audit/internal/webui/api.go

@@ -125,6 +125,8 @@ func defaultTaskPriority(target string, params taskParams) int {
 		return taskPriorityInstall
 	case "install-to-ram":
 		return taskPriorityInstallToRAM
+	case "nvme-format":
+		return taskPriorityInstall
 	case "audit":
 		return taskPriorityAudit
 	case "nvidia-bench-perf", "nvidia-bench-power", "nvidia-bench-autotune":
@@ -1295,7 +1297,7 @@ func (h *handler) handleAPIInstallToRAM(w http.ResponseWriter, r *http.Request)
 var standardTools = []string{
 	"dmidecode", "smartctl", "nvme", "lspci", "ipmitool",
 	"nvidia-smi", "dcgmi", "nv-hostengine", "memtester", "stress-ng", "nvtop",
-	"mstflint", "qrencode",
+	"mstflint",
 }
 
 func (h *handler) handleAPIToolsCheck(w http.ResponseWriter, r *http.Request) {

audit/internal/webui/api_test.go

@@ -85,6 +85,27 @@ func TestHandleAPIBlackboxStatusReturnsPersistedState(t *testing.T) {
 	}
 }
 
+func TestParseNVMeFormatModes(t *testing.T) {
+	raw := `
+lbaf  0 : ms:0   lbads:9  rp:0x2 (in use)
+lbaf  1 : ms:8   lbads:9  rp:0x1
+lbaf  2 : ms:0   lbads:12 rp:0
+`
+	modes := parseNVMeFormatModes(raw)
+	if len(modes) != 3 {
+		t.Fatalf("modes=%#v want 3 modes", modes)
+	}
+	if modes[0].Mode != 0 || modes[0].DataBytes != 512 || modes[0].MetadataBytes != 0 || !modes[0].InUse {
+		t.Fatalf("mode 0=%#v", modes[0])
+	}
+	if modes[1].Label != "MODE 1 (512+8)" {
+		t.Fatalf("mode 1 label=%q", modes[1].Label)
+	}
+	if modes[2].DataBytes != 4096 || modes[2].MetadataBytes != 0 {
+		t.Fatalf("mode 2=%#v", modes[2])
+	}
+}
+
 func TestHandleAPIBenchmarkNvidiaRunQueuesSelectedGPUs(t *testing.T) {
 	globalQueue.mu.Lock()
 	originalTasks := globalQueue.tasks

audit/internal/webui/nvme_format.go

@@ -0,0 +1,368 @@
+package webui
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"net/http"
+	"os/exec"
+	"path/filepath"
+	"regexp"
+	"sort"
+	"strconv"
+	"strings"
+	"time"
+)
+
+type nvmeFormatMode struct {
+	Mode          int    `json:"mode"`
+	DataBytes     int64  `json:"data_bytes"`
+	MetadataBytes int64  `json:"metadata_bytes"`
+	InUse         bool   `json:"in_use"`
+	Label         string `json:"label"`
+}
+
+type nvmeFormatDisk struct {
+	Device        string           `json:"device"`
+	Model         string           `json:"model,omitempty"`
+	Serial        string           `json:"serial,omitempty"`
+	Size          string           `json:"size,omitempty"`
+	CurrentMode   int              `json:"current_mode"`
+	CurrentFormat string           `json:"current_format"`
+	Modes         []nvmeFormatMode `json:"modes"`
+	Error         string           `json:"error,omitempty"`
+}
+
+type nvmeListJSON struct {
+	Devices []struct {
+		DevicePath   string `json:"DevicePath"`
+		ModelNumber  string `json:"ModelNumber"`
+		SerialNumber string `json:"SerialNumber"`
+		PhysicalSize int64  `json:"PhysicalSize"`
+	} `json:"Devices"`
+}
+
+var (
+	nvmeFormatDeviceRE     = regexp.MustCompile(`^/dev/nvme[0-9]+n[0-9]+$`)
+	nvmeLBAFCompactLineRE  = regexp.MustCompile(`(?im)^\s*lbaf\s+(\d+)\s*:\s*ms:(\d+)\s+lbads:(\d+).*$`)
+	nvmeLBAFVerboseLineRE  = regexp.MustCompile(`(?im)^\s*LBA Format\s+(\d+)\s*:\s*Metadata Size:\s*(\d+)\s+bytes\s*-\s*Data Size:\s*(\d+)\s+bytes.*$`)
+	nvmeCommandContext     = exec.CommandContext
+	nvmeListFormatsTimeout = 20 * time.Second
+)
+
+func listNVMeFormatDisks(ctx context.Context) ([]nvmeFormatDisk, error) {
+	ctx, cancel := context.WithTimeout(ctx, nvmeListFormatsTimeout)
+	defer cancel()
+	out, err := nvmeCommandContext(ctx, "nvme", "list", "-o", "json").Output()
+	if err != nil {
+		return nil, err
+	}
+	var root nvmeListJSON
+	if err := json.Unmarshal(out, &root); err != nil {
+		return nil, err
+	}
+	disks := make([]nvmeFormatDisk, 0, len(root.Devices))
+	seen := map[string]struct{}{}
+	for _, dev := range root.Devices {
+		path := strings.TrimSpace(dev.DevicePath)
+		if !nvmeFormatDeviceRE.MatchString(path) {
+			continue
+		}
+		if _, ok := seen[path]; ok {
+			continue
+		}
+		seen[path] = struct{}{}
+		disk := nvmeFormatDisk{
+			Device:      path,
+			Model:       strings.TrimSpace(dev.ModelNumber),
+			Serial:      strings.TrimSpace(dev.SerialNumber),
+			Size:        formatNVMeBytes(dev.PhysicalSize),
+			CurrentMode: -1,
+		}
+		modes, parseErr := readNVMeFormatModes(ctx, path)
+		if parseErr != nil {
+			disk.Error = parseErr.Error()
+		}
+		disk.Modes = modes
+		for _, mode := range modes {
+			if mode.InUse {
+				disk.CurrentMode = mode.Mode
+				disk.CurrentFormat = formatNVMeBlock(mode.DataBytes, mode.MetadataBytes)
+				break
+			}
+		}
+		disks = append(disks, disk)
+	}
+	sort.Slice(disks, func(i, j int) bool { return disks[i].Device < disks[j].Device })
+	return disks, nil
+}
+
+func readNVMeFormatModes(ctx context.Context, device string) ([]nvmeFormatMode, error) {
+	if !nvmeFormatDeviceRE.MatchString(device) {
+		return nil, fmt.Errorf("invalid NVMe device")
+	}
+	out, err := nvmeCommandContext(ctx, "nvme", "id-ns", device, "-H").CombinedOutput()
+	if err != nil {
+		msg := strings.TrimSpace(string(out))
+		if msg == "" {
+			msg = err.Error()
+		}
+		return nil, fmt.Errorf("%s", msg)
+	}
+	modes := parseNVMeFormatModes(string(out))
+	if len(modes) == 0 {
+		return nil, fmt.Errorf("no LBA format modes found")
+	}
+	return modes, nil
+}
+
+func parseNVMeFormatModes(raw string) []nvmeFormatMode {
+	byMode := map[int]nvmeFormatMode{}
+	for _, m := range nvmeLBAFCompactLineRE.FindAllStringSubmatch(raw, -1) {
+		mode, errMode := strconv.Atoi(m[1])
+		metadata, errMS := strconv.ParseInt(m[2], 10, 64)
+		lbads, errLBADS := strconv.Atoi(m[3])
+		if errMode != nil || errMS != nil || errLBADS != nil || lbads < 0 || lbads >= 63 {
+			continue
+		}
+		data := int64(1) << lbads
+		line := m[0]
+		byMode[mode] = nvmeFormatMode{
+			Mode:          mode,
+			DataBytes:     data,
+			MetadataBytes: metadata,
+			InUse:         strings.Contains(strings.ToLower(line), "in use"),
+			Label:         fmt.Sprintf("MODE %d (%s)", mode, formatNVMeBlock(data, metadata)),
+		}
+	}
+	for _, m := range nvmeLBAFVerboseLineRE.FindAllStringSubmatch(raw, -1) {
+		mode, errMode := strconv.Atoi(m[1])
+		metadata, errMS := strconv.ParseInt(m[2], 10, 64)
+		data, errData := strconv.ParseInt(m[3], 10, 64)
+		if errMode != nil || errMS != nil || errData != nil || data <= 0 {
+			continue
+		}
+		line := m[0]
+		byMode[mode] = nvmeFormatMode{
+			Mode:          mode,
+			DataBytes:     data,
+			MetadataBytes: metadata,
+			InUse:         strings.Contains(strings.ToLower(line), "in use"),
+			Label:         fmt.Sprintf("MODE %d (%s)", mode, formatNVMeBlock(data, metadata)),
+		}
+	}
+	modes := make([]nvmeFormatMode, 0, len(byMode))
+	for _, mode := range byMode {
+		modes = append(modes, mode)
+	}
+	sort.Slice(modes, func(i, j int) bool { return modes[i].Mode < modes[j].Mode })
+	return modes
+}
+
+func runNVMeFormatTask(ctx context.Context, j *jobState, device string, lbaf int) error {
+	if !nvmeFormatDeviceRE.MatchString(device) {
+		return fmt.Errorf("invalid NVMe device")
+	}
+	modes, err := readNVMeFormatModes(ctx, device)
+	if err != nil {
+		return err
+	}
+	var selected nvmeFormatMode
+	found := false
+	for _, mode := range modes {
+		if mode.Mode == lbaf {
+			selected = mode
+			found = true
+			break
+		}
+	}
+	if !found {
+		return fmt.Errorf("MODE %d is not available on %s", lbaf, device)
+	}
+	ms := 0
+	if selected.MetadataBytes > 0 {
+		ms = 1
+	}
+	j.append(fmt.Sprintf("Formatting %s to %s with --lbaf=%d --ms=%d --force", device, formatNVMeBlock(selected.DataBytes, selected.MetadataBytes), selected.Mode, ms))
+	cmd := nvmeCommandContext(ctx, "nvme", "format", device, fmt.Sprintf("--lbaf=%d", selected.Mode), fmt.Sprintf("--ms=%d", ms), "--force")
+	return streamCmdJob(j, cmd)
+}
+
+func (h *handler) handleAPINVMeFormats(w http.ResponseWriter, r *http.Request) {
+	disks, err := listNVMeFormatDisks(r.Context())
+	if err != nil {
+		writeError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	writeJSON(w, disks)
+}
+
+func (h *handler) handleAPINVMeFormatRun(w http.ResponseWriter, r *http.Request) {
+	var req struct {
+		Device string `json:"device"`
+		LBAF   int    `json:"lbaf"`
+	}
+	if err := json.NewDecoder(r.Body).Decode(&req); err != nil {
+		writeError(w, http.StatusBadRequest, "invalid request body")
+		return
+	}
+	if !nvmeFormatDeviceRE.MatchString(req.Device) {
+		writeError(w, http.StatusBadRequest, "invalid NVMe device")
+		return
+	}
+	disks, err := listNVMeFormatDisks(r.Context())
+	if err != nil {
+		writeError(w, http.StatusInternalServerError, err.Error())
+		return
+	}
+	var label string
+	allowed := false
+	for _, disk := range disks {
+		if disk.Device != req.Device {
+			continue
+		}
+		for _, mode := range disk.Modes {
+			if mode.Mode == req.LBAF {
+				allowed = true
+				label = mode.Label
+				break
+			}
+		}
+	}
+	if !allowed {
+		writeError(w, http.StatusBadRequest, "LBA format mode is not available for this device")
+		return
+	}
+	name := fmt.Sprintf("NVMe Format %s to %s", filepath.Base(req.Device), label)
+	t := &Task{
+		ID:        newJobID("nvme-format"),
+		Name:      name,
+		Target:    "nvme-format",
+		Priority:  defaultTaskPriority("nvme-format", taskParams{}),
+		Status:    TaskPending,
+		CreatedAt: time.Now(),
+		params: taskParams{
+			Device: req.Device,
+			LBAF:   req.LBAF,
+		},
+	}
+	globalQueue.enqueue(t)
+	writeJSON(w, map[string]string{"task_id": t.ID, "job_id": t.ID})
+}
+
+func formatNVMeBlock(dataBytes, metadataBytes int64) string {
+	return strconv.FormatInt(dataBytes, 10) + "+" + strconv.FormatInt(metadataBytes, 10)
+}
+
+func formatNVMeBytes(n int64) string {
+	if n <= 0 {
+		return ""
+	}
+	units := []string{"B", "KB", "MB", "GB", "TB", "PB"}
+	v := float64(n)
+	unit := 0
+	for v >= 1000 && unit < len(units)-1 {
+		v /= 1000
+		unit++
+	}
+	if unit == 0 {
+		return fmt.Sprintf("%d B", n)
+	}
+	return fmt.Sprintf("%.1f %s", v, units[unit])
+}
+
+func renderNVMeFormatInline() string {
+	return `<div id="nvme-format-status" style="font-size:13px;color:var(--muted);margin-bottom:12px">Loading NVMe disks...</div>
+<div id="nvme-format-table"><p style="color:var(--muted);font-size:13px">Loading...</p></div>
+<script>
+function nvmeFormatEsc(s) {
+  return String(s == null ? '' : s).replace(/[&<>"']/g, function(c) {
+    return {'&':'&amp;','<':'&lt;','>':'&gt;','"':'&quot;',"'":'&#39;'}[c];
+  });
+}
+function loadNVMeFormats() {
+  var status = document.getElementById('nvme-format-status');
+  var table = document.getElementById('nvme-format-table');
+  status.textContent = 'Loading NVMe disks...';
+  status.style.color = 'var(--muted)';
+  table.innerHTML = '<p style="color:var(--muted);font-size:13px">Loading...</p>';
+  fetch('/api/tools/nvme-formats').then(function(r) { return r.json().then(function(d) { if (!r.ok) throw new Error(d.error || ('HTTP ' + r.status)); return d; }); }).then(function(disks) {
+    window._nvmeFormatDisks = Array.isArray(disks) ? disks : [];
+    if (!window._nvmeFormatDisks.length) {
+      status.textContent = 'No NVMe disks found.';
+      table.innerHTML = '';
+      return;
+    }
+    status.textContent = window._nvmeFormatDisks.length + ' NVMe disk(s) found.';
+    var rows = window._nvmeFormatDisks.map(function(d, idx) {
+      var current = d.current_format ? (d.current_format + ' / MODE ' + d.current_mode) : 'unknown';
+      var detail = [d.model || '', d.serial || '', d.size || ''].filter(Boolean).join(' | ');
+      var options = (d.modes || []).map(function(m) {
+        return '<option value="' + m.mode + '"' + (m.in_use ? ' selected' : '') + '>' + nvmeFormatEsc(m.label) + '</option>';
+      }).join('');
+      var disabled = options ? '' : ' disabled';
+      var err = d.error ? '<div style="font-size:12px;color:var(--crit-fg,#9f3a38);margin-top:4px">' + nvmeFormatEsc(d.error) + '</div>' : '';
+      return '<tr>'
+        + '<td style="font-family:monospace;white-space:nowrap">' + nvmeFormatEsc(d.device) + (detail ? '<div style="font-family:inherit;font-size:12px;color:var(--muted)">' + nvmeFormatEsc(detail) + '</div>' : '') + '</td>'
+        + '<td style="white-space:nowrap">' + nvmeFormatEsc(current) + err + '</td>'
+        + '<td style="white-space:nowrap"><select id="nvme-format-select-' + idx + '"' + disabled + '>' + options + '</select></td>'
+        + '<td style="white-space:nowrap"><button class="btn btn-sm btn-primary" onclick="nvmeFormatRun(' + idx + ', this)"' + disabled + '>Apply</button><div class="nvme-format-row-msg" style="margin-top:6px;font-size:12px;color:var(--muted)"></div></td>'
+        + '</tr>';
+    }).join('');
+    table.innerHTML = '<table><tr><th>Disk</th><th>Current block / mode</th><th>New mode</th><th>Action</th></tr>' + rows + '</table>';
+  }).catch(function(e) {
+    status.textContent = 'Error loading NVMe disks: ' + e.message;
+    status.style.color = 'var(--crit-fg,#9f3a38)';
+    table.innerHTML = '';
+  });
+}
+function nvmeWaitTaskDone(taskID, rowMsg) {
+  var timer = setInterval(function() {
+    fetch('/api/tasks').then(function(r) { return r.json(); }).then(function(tasks) {
+      var task = (tasks || []).find(function(t) { return t.id === taskID; });
+      if (!task) return;
+      if (task.status === 'done' || task.status === 'failed' || task.status === 'cancelled') {
+        clearInterval(timer);
+        rowMsg.textContent = 'Task ' + taskID + ': ' + task.status + (task.error ? ' - ' + task.error : '');
+        rowMsg.style.color = task.status === 'done' ? 'var(--ok,green)' : 'var(--crit-fg,#9f3a38)';
+        loadNVMeFormats();
+      }
+    }).catch(function(){});
+  }, 1500);
+}
+function nvmeFormatRun(idx, btn) {
+  var disk = (window._nvmeFormatDisks || [])[idx];
+  var select = document.getElementById('nvme-format-select-' + idx);
+  var row = btn.closest('td');
+  var rowMsg = row.querySelector('.nvme-format-row-msg');
+  if (!disk || !select) return;
+  var lbaf = parseInt(select.value, 10);
+  var mode = (disk.modes || []).find(function(m) { return m.mode === lbaf; });
+  if (!mode) return;
+  if (!window.confirm('Format ' + disk.device + ' to ' + mode.label + '? This erases data on the namespace.')) return;
+  btn.disabled = true;
+  rowMsg.style.color = 'var(--muted)';
+  rowMsg.textContent = 'Queued...';
+  fetch('/api/tools/nvme-format/run', {
+    method:'POST',
+    headers:{'Content-Type':'application/json'},
+    body:JSON.stringify({device: disk.device, lbaf: lbaf})
+  }).then(function(r) { return r.json().then(function(d) { if (!r.ok) throw new Error(d.error || ('HTTP ' + r.status)); return d; }); }).then(function(d) {
+    rowMsg.textContent = 'Task ' + d.task_id + ' queued.';
+    nvmeWaitTaskDone(d.task_id, rowMsg);
+  }).catch(function(e) {
+    rowMsg.style.color = 'var(--crit-fg,#9f3a38)';
+    rowMsg.textContent = 'Error: ' + e.message;
+  }).finally(function() {
+    btn.disabled = false;
+  });
+}
+loadNVMeFormats();
+</script>`
+}
+
+func renderNVMeFormatCard() string {
+	return `<div class="card"><div class="card-head">NVMe Block Format <button class="btn btn-sm btn-secondary" onclick="loadNVMeFormats()" style="margin-left:auto">&#8635; Refresh</button></div><div class="card-body">` +
+		`<p style="font-size:13px;color:var(--muted);margin-bottom:12px">Lists NVMe namespaces and changes their LBA format through a queued task.</p>` +
+		renderNVMeFormatInline() + `</div></div>`
+}

audit/internal/webui/page_export_tools.go

@@ -475,6 +475,7 @@ function installToRAM() {
 <div class="card"><div class="card-head">Services</div><div class="card-body">` +
 		renderServicesInline() + `</div></div>
 
+` + renderNVMeFormatCard() + `
 
 <script>
 function checkTools() {

audit/internal/webui/server.go

@@ -307,6 +307,8 @@ func NewHandler(opts HandlerOptions) http.Handler {
 
 	// Tools
 	mux.HandleFunc("GET /api/tools/check", h.handleAPIToolsCheck)
+	mux.HandleFunc("GET /api/tools/nvme-formats", h.handleAPINVMeFormats)
+	mux.HandleFunc("POST /api/tools/nvme-format/run", h.handleAPINVMeFormatRun)
 
 	// GPU presence / tools
 	mux.HandleFunc("GET /api/gpu/presence", h.handleAPIGPUPresence)

audit/internal/webui/server_test.go

@@ -677,6 +677,12 @@ func TestToolsPageRendersNvidiaSelfHealSection(t *testing.T) {
 	if !strings.Contains(body, `/api/blackbox/status`) {
 		t.Fatalf("tools page missing black-box status api usage: %s", body)
 	}
+	if !strings.Contains(body, `NVMe Block Format`) {
+		t.Fatalf("tools page missing nvme block format section: %s", body)
+	}
+	if !strings.Contains(body, `/api/tools/nvme-formats`) || !strings.Contains(body, `/api/tools/nvme-format/run`) {
+		t.Fatalf("tools page missing nvme format api usage: %s", body)
+	}
 }
 
 func TestBenchmarkPageRendersGPUSelectionControls(t *testing.T) {

audit/internal/webui/task_runner.go

@@ -376,6 +376,12 @@ func executeTaskWithOptions(opts *HandlerOptions, t *Task, j *jobState, ctx cont
 			break
 		}
 		err = a.RunInstallToRAM(ctx, j.append)
+	case "nvme-format":
+		if strings.TrimSpace(t.params.Device) == "" {
+			err = fmt.Errorf("device is required")
+			break
+		}
+		err = runNVMeFormatTask(ctx, j, t.params.Device, t.params.LBAF)
 	default:
 		j.append("ERROR: unknown target: " + t.Target)
 		j.finish("unknown target")

audit/internal/webui/tasks.go

@@ -57,6 +57,7 @@ var taskNames = map[string]string{
 	"support-bundle":         "Support Bundle",
 	"install":                "Install to Disk",
 	"install-to-ram":         "Install to RAM",
+	"nvme-format":            "NVMe Block Format Change",
 }
 
 // burnNames maps target → human-readable name when a burn profile is set.
@@ -137,6 +138,7 @@ type taskParams struct {
 	RampRunID          string   `json:"ramp_run_id,omitempty"`
 	DisplayName        string   `json:"display_name,omitempty"`
 	Device             string   `json:"device,omitempty"` // for install
+	LBAF               int      `json:"lbaf,omitempty"`
 	PlatformComponents []string `json:"platform_components,omitempty"`
 }
 

bible-local/README.md

@@ -9,5 +9,62 @@ Generic engineering rules live in `bible/rules/patterns/`.
 |---|---|
 | `architecture/system-overview.md` | What bee does, scope, tech stack |
 | `architecture/runtime-flows.md` | Boot sequence, audit flow, service order |
+| `docs/customer-gpu-test-methodology.md` | Customer-facing GPU PCIe Validate / Validate -> Stress test list |
 | `docs/hardware-ingest-contract.md` | Current Reanimator hardware ingest JSON contract |
+| `docs/validate-vs-burn.md` | Validate and Validate -> Stress hardware test policy |
 | `decisions/` | Architectural decision log, including read-only submodule policy |
+
+## Validate Test Matrix
+
+### Validate
+
+- CPU check
+  - `lscpu`
+  - `sensors`
+  - `stress-ng`
+- Memory check
+  - `free`
+  - `timeout <timeout_sec> memtester`
+  - `free`
+- NVMe storage check
+  - `nvme id-ctrl`
+  - `nvme smart-log`
+  - `nvme device-self-test`
+- SATA/SAS storage check
+  - `smartctl -H -A`
+  - `smartctl -t short`
+- Basic NVIDIA GPU check
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dmidecode -t baseboard`
+  - `dmidecode -t system`
+  - `dcgmi diag -r 2`
+- Inter-GPU communication check
+  - `all_reduce_perf`
+- GPU bandwidth check
+  - `dcgmi diag -r nvbandwidth`
+
+### Validate -> Stress
+
+- Extended NVIDIA GPU check
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dmidecode -t baseboard`
+  - `dmidecode -t system`
+  - `dcgmi diag -r 3`
+- NVIDIA targeted stress
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dcgmi diag -r targeted_stress`
+- NVIDIA targeted power
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dcgmi diag -r targeted_power`
+- NVIDIA pulse test
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dcgmi diag -r pulse_test`
+- Inter-GPU communication check
+  - `all_reduce_perf`
+- GPU bandwidth check
+  - `dcgmi diag -r nvbandwidth`

bible-local/architecture/runtime-flows.md

@@ -149,7 +149,6 @@ Current validation state:
   6. psu collector     (ipmitool fru + sdr — silent if no /dev/ipmi0)
   7. nvidia enrichment (nvidia-smi — skipped if binary absent or driver not loaded)
   8. output JSON → /var/log/bee-audit.json
-  9. QR summary to stdout (qrencode if available)
 ```
 
 Every collector returns `nil, nil` on tool-not-found. Errors are logged, never fatal.

bible-local/docs/customer-gpu-test-methodology.md

@@ -0,0 +1,54 @@
+# GPU PCIe Test Methodology
+
+## Validate
+
+- CPU check
+  - `lscpu`
+  - `sensors`
+  - `stress-ng`
+- Memory check
+  - `free`
+  - `timeout <timeout_sec> memtester`
+  - `free`
+- NVMe storage check
+  - `nvme id-ctrl`
+  - `nvme smart-log`
+  - `nvme device-self-test`
+- SATA/SAS storage check
+  - `smartctl -H -A`
+  - `smartctl -t short`
+- Basic NVIDIA GPU check
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dmidecode -t baseboard`
+  - `dmidecode -t system`
+  - `dcgmi diag -r 2`
+- Inter-GPU communication check
+  - `all_reduce_perf`
+- GPU bandwidth check
+  - `dcgmi diag -r nvbandwidth`
+
+## Validate -> Stress
+
+- Extended NVIDIA GPU check
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dmidecode -t baseboard`
+  - `dmidecode -t system`
+  - `dcgmi diag -r 3`
+- NVIDIA targeted stress
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dcgmi diag -r targeted_stress`
+- NVIDIA targeted power
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dcgmi diag -r targeted_power`
+- NVIDIA pulse test
+  - `nvidia-smi -pm 1`
+  - `nvidia-smi -q`
+  - `dcgmi diag -r pulse_test`
+- Inter-GPU communication check
+  - `all_reduce_perf`
+- GPU bandwidth check
+  - `dcgmi diag -r nvbandwidth`

git-bible/grub-bitmap-error.md

@@ -0,0 +1,134 @@
+# GRUB bitmap error: null src bitmap in grub_video_bitmap_create_scaled
+
+## Symptom
+
+```
+error: null src bitmap in grub_video_bitmap_create_scaled.
+Press any key to continue...
+```
+
+Appears on boot before the GRUB menu renders. The menu still appears after pressing a key,
+but without the bee logo. Reproduced on real hardware (Lenovo SR650 V3, ASUS GPU servers).
+
+## Root cause model
+
+`grub_video_bitmap_create_scaled` receives a null `src` pointer, meaning the PNG loader
+returned null for `bee-logo.png`. GRUB calls this function even when no explicit
+`width`/`height` are set in `theme.txt` — it is invoked any time an image component is
+rendered, passing the image's natural dimensions as the target size.
+
+The PNG file is referenced as `file = "bee-logo.png"` (relative to theme dir).
+GRUB resolves this to `/boot/grub/live-theme/bee-logo.png`.
+
+## Attempts that did NOT fix the error
+
+### Attempt 1 — add explicit `width`/`height` to image block (d52ec67)
+
+**What was done:** First introduction of bee-logo.png with:
+```
++ image {
+    top = 4%
+    left = 50%-200
+    width = 400
+    height = 400
+    file = "bee-logo.png"
+}
+```
+PNG at this point was RGBA (color_type=6).
+
+**Result:** Error appeared immediately on first ISO build.
+
+---
+
+### Attempt 2 — remove `width`/`height` from image block (aa284ae)
+
+**Hypothesis:** Explicit scaling dimensions trigger the scale path; removing them avoids it.
+
+**What was done:** Removed `width = 400` and `height = 400` from the image block.
+```
++ image {
+    top = 4%
+    left = 50%-200
+    file = "bee-logo.png"
+}
+```
+
+**Result:** Error persists. GRUB calls `grub_video_bitmap_create_scaled` regardless of whether
+`width`/`height` are specified — if the bitmap is null (loading failed), the error fires either way.
+
+---
+
+### Attempt 3 — convert PNG to RGBA + strip metadata chunks (6112094)
+
+**Hypothesis:** GRUB's minimal PNG parser is confused by metadata chunks (cHRM, bKGD, tIME, tEXt).
+Also re-ordered `terminal_output gfxterm` before `insmod png` / theme load.
+
+**What was done:**
+- Converted PNG to RGBA color_type=6, stripped all ancillary chunks
+- Moved `terminal_output gfxterm` earlier in config.cfg
+- Removed echo ASCII art banner from grub.cfg
+
+**Result:** Error persists — and this change actually confirmed RGBA does not work:
+GRUB's PNG loader does not render RGBA PNGs correctly on this platform.
+
+---
+
+### Attempt 4 — convert PNG from RGBA back to RGB (333c44f, most recent)
+
+**Hypothesis:** GRUB does not support RGBA (color_type=6); RGB (color_type=2) is the correct format.
+Alpha channel composited onto black background (#000000) to match `desktop-color`.
+
+**What was done:** Converted bee-logo.png from RGBA to RGB via ImageMagick.
+
+**Current file state:**
+- 400×400 px, 8-bit/color RGB, non-interlaced
+- Only IHDR + IDAT + IEND chunks (no metadata)
+- `insmod png` is present in config.cfg
+- `terminal_output gfxterm` runs before theme is sourced
+- No explicit `width`/`height` in image block
+
+**Result:** Error still occurs on real hardware. Despite the PNG being nominally correct
+(RGB, non-interlaced, minimal chunks), the bitmap load returns null.
+
+## Confirmed root cause (verified on 172.16.41.94, 2026-04-30)
+
+The EFI partition (`sda2`, vfat, 5 MB) contains only:
+```
+/EFI/boot/bootia32.efi
+/EFI/boot/bootx64.efi
+/EFI/boot/grubx64.efi
+/boot/grub/grub.cfg
+```
+
+`config.cfg`, `theme.cfg`, and the entire `live-theme/` directory (including `bee-logo.png`)
+are **absent from the EFI image**. `live-build`'s `lb binary_grub-efi` stage is not
+copying these files. GRUB boots, sources only `grub.cfg`, then fails to load the theme
+because the file does not exist — returning a null bitmap regardless of PNG format.
+
+All four fix attempts were targeting the wrong layer (PNG format/content).
+
+## Fix (applied 2026-04-30)
+
+Switched from PNG to TGA format:
+
+1. Converted `bee-logo.png` → `bee-logo.tga` (24-bit uncompressed BGR, top-left origin,
+   480018 bytes). Conversion done via Python stdlib (no external tools needed).
+2. `config.cfg`: `insmod png` → `insmod tga`
+3. `theme.txt`: `file = "bee-logo.png"` → `file = "bee-logo.tga"`
+
+**Why TGA works:** GRUB's TGA reader (`tga.mod`) handles uncompressed 24-bit images
+trivially — no decompression, no complex chunk parsing. The module is present on-disk
+(`x86_64-efi/tga.mod`). PNG was failing despite a valid file; the exact GRUB bug is
+unknown but the PNG reader in Debian bookworm's grub2 is known to be fragile.
+
+The old `bee-logo.png` is kept in the tree (may be useful for other tools) but is no
+longer referenced by the theme.
+
+## Relevant files
+
+| File | Purpose |
+|------|---------|
+| `iso/builder/config/bootloaders/grub-efi/config.cfg` | insmod png, gfxterm init, theme source |
+| `iso/builder/config/bootloaders/grub-efi/theme.cfg` | sets `theme=` path |
+| `iso/builder/config/bootloaders/grub-efi/live-theme/theme.txt` | image component definition |
+| `iso/builder/config/bootloaders/grub-efi/live-theme/bee-logo.png` | the logo PNG |

iso/builder/config/bootloaders/grub-efi/config.cfg

@@ -27,5 +27,5 @@ insmod gfxterm
 terminal_input console serial
 terminal_output gfxterm serial
 
-insmod png
+insmod tga
 source /boot/grub/theme.cfg

iso/builder/config/bootloaders/grub-efi/live-theme/theme.txt

@@ -9,7 +9,7 @@ terminal-font: "Unifont Regular 16"
 + image {
         top = 4%
         left = 50%-200
-        file = "bee-logo.png"
+        file = "bee-logo.tga"
 }
 
 #help bar at the bottom

iso/builder/config/hooks/normal/9000-bee-setup.hook.chroot

@@ -31,6 +31,7 @@ systemctl enable bee-preflight.service
 systemctl enable bee-audit.service
 systemctl enable bee-web.service
 systemctl enable bee-sshsetup.service
+systemctl enable bee-blackbox.service
 systemctl enable bee-selfheal.timer
 systemctl enable bee-boot-status.service
 systemctl enable ssh.service

iso/builder/config/package-lists/bee.list.chroot

@@ -66,9 +66,6 @@ jq
 curl
 net-tools
 
-# QR codes (for displaying audit results)
-qrencode
-
 # Local desktop (openbox + chromium kiosk)
 gparted
 openbox