fix(metrics): correct chart padding order — right=80 not top=80

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

audit/go.mod

@@ -1,6 +1,6 @@
 module bee/audit
 
-go 1.24.0
+go 1.25.0
 
 replace reanimator/chart => ../internal/chart
 
@@ -13,5 +13,14 @@ require (
 	github.com/dustin/go-humanize v1.0.1 // indirect
 	github.com/go-analyze/bulk v0.1.3 // indirect
 	github.com/golang/freetype v0.0.0-20170609003504-e2365dfdc4a0 // indirect
+	github.com/google/uuid v1.6.0 // indirect
+	github.com/mattn/go-isatty v0.0.20 // indirect
+	github.com/ncruces/go-strftime v1.0.0 // indirect
+	github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec // indirect
 	golang.org/x/image v0.24.0 // indirect
+	golang.org/x/sys v0.42.0 // indirect
+	modernc.org/libc v1.70.0 // indirect
+	modernc.org/mathutil v1.7.1 // indirect
+	modernc.org/memory v1.11.0 // indirect
+	modernc.org/sqlite v1.48.0 // indirect
 )

audit/go.sum

@@ -8,11 +8,30 @@ github.com/go-analyze/charts v0.5.26 h1:rSwZikLQuFX6cJzwI8OAgaWZneG1kDYxD857ms00
 github.com/go-analyze/charts v0.5.26/go.mod h1:s1YvQhjiSwtLx1f2dOKfiV9x2TT49nVSL6v2rlRpTbY=
 github.com/golang/freetype v0.0.0-20170609003504-e2365dfdc4a0 h1:DACJavvAHhabrF08vX0COfcOBJRhZ8lUbR+ZWIs0Y5g=
 github.com/golang/freetype v0.0.0-20170609003504-e2365dfdc4a0/go.mod h1:E/TSTwGwJL78qG/PmXZO1EjYhfJinVAhrmmHX6Z8B9k=
+github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
+github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
+github.com/mattn/go-isatty v0.0.20 h1:xfD0iDuEKnDkl03q4limB+vH+GxLEtL/jb4xVJSWWEY=
+github.com/mattn/go-isatty v0.0.20/go.mod h1:W+V8PltTTMOvKvAeJH7IuucS94S2C6jfK/D7dTCTo3Y=
+github.com/ncruces/go-strftime v1.0.0 h1:HMFp8mLCTPp341M/ZnA4qaf7ZlsbTc+miZjCLOFAw7w=
+github.com/ncruces/go-strftime v1.0.0/go.mod h1:Fwc5htZGVVkseilnfgOVb9mKy6w1naJmn9CehxcKcls=
 github.com/pmezard/go-difflib v1.0.0 h1:4DBwDE0NGyQoBHbLQYPwSUPoCMWR5BEzIk/f1lZbAQM=
 github.com/pmezard/go-difflib v1.0.0/go.mod h1:iKH77koFhYxTK1pcRnkKkqfTogsbg7gZNVY4sRDYZ/4=
+github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec h1:W09IVJc94icq4NjY3clb7Lk8O1qJ8BdBEF8z0ibU0rE=
+github.com/remyoudompheng/bigfft v0.0.0-20230129092748-24d4a6f8daec/go.mod h1:qqbHyh8v60DhA7CoWK5oRCqLrMHRGoxYCSS9EjAz6Eo=
 github.com/stretchr/testify v1.11.1 h1:7s2iGBzp5EwR7/aIZr8ao5+dra3wiQyKjjFuvgVKu7U=
 github.com/stretchr/testify v1.11.1/go.mod h1:wZwfW3scLgRK+23gO65QZefKpKQRnfz6sD981Nm4B6U=
 golang.org/x/image v0.24.0 h1:AN7zRgVsbvmTfNyqIbbOraYL8mSwcKncEj8ofjgzcMQ=
 golang.org/x/image v0.24.0/go.mod h1:4b/ITuLfqYq1hqZcjofwctIhi7sZh2WaCjvsBNjjya8=
+golang.org/x/sys v0.6.0/go.mod h1:oPkhp1MJrh7nUepCBck5+mAzfO9JrbApNNgaTdGDITg=
+golang.org/x/sys v0.42.0 h1:omrd2nAlyT5ESRdCLYdm3+fMfNFE/+Rf4bDIQImRJeo=
+golang.org/x/sys v0.42.0/go.mod h1:4GL1E5IUh+htKOUEOaiffhrAeqysfVGipDYzABqnCmw=
 gopkg.in/yaml.v3 v3.0.1 h1:fxVm/GzAzEWqLHuvctI91KS9hhNmmWOoWu0XTYJS7CA=
 gopkg.in/yaml.v3 v3.0.1/go.mod h1:K4uyk7z7BCEPqu6E+C64Yfv1cQ7kz7rIZviUmN+EgEM=
+modernc.org/libc v1.70.0 h1:U58NawXqXbgpZ/dcdS9kMshu08aiA6b7gusEusqzNkw=
+modernc.org/libc v1.70.0/go.mod h1:OVmxFGP1CI/Z4L3E0Q3Mf1PDE0BucwMkcXjjLntvHJo=
+modernc.org/mathutil v1.7.1 h1:GCZVGXdaN8gTqB1Mf/usp1Y/hSqgI2vAGGP4jZMCxOU=
+modernc.org/mathutil v1.7.1/go.mod h1:4p5IwJITfppl0G4sUEDtCr4DthTaT47/N3aT6MhfgJg=
+modernc.org/memory v1.11.0 h1:o4QC8aMQzmcwCK3t3Ux/ZHmwFPzE6hf2Y5LbkRs+hbI=
+modernc.org/memory v1.11.0/go.mod h1:/JP4VbVC+K5sU2wZi9bHoq2MAkCnrt2r98UGeSK7Mjw=
+modernc.org/sqlite v1.48.0 h1:ElZyLop3Q2mHYk5IFPPXADejZrlHu7APbpB0sF78bq4=
+modernc.org/sqlite v1.48.0/go.mod h1:hWjRO6Tj/5Ik8ieqxQybiEOUXy0NJFNp2tpvVpKlvig=

audit/internal/platform/gpu_metrics.go

@@ -78,48 +78,56 @@ func SampleGPUMetrics(gpuIndices []int) ([]GPUMetricRow, error) {
 
 // sampleAMDGPUMetrics queries rocm-smi for live GPU metrics.
 func sampleAMDGPUMetrics() ([]GPUMetricRow, error) {
-	// --showtemp --showuse --showpower --csv — one row per GPU
 	out, err := runROCmSMI("--showtemp", "--showuse", "--showpower", "--showmemuse", "--csv")
 	if err != nil {
 		return nil, err
 	}
-	var rows []GPUMetricRow
-	for _, line := range strings.Split(strings.TrimSpace(string(out)), "\n") {
-		line = strings.TrimSpace(line)
-		if line == "" || strings.HasPrefix(strings.ToLower(line), "device") {
-			continue
-		}
-		// CSV format: device,temp_c,gpu_use%,mem_use%,power_w (order may vary by rocm-smi version)
-		// We parse by column header from the first line.
-		parts := strings.Split(line, ",")
-		if len(parts) < 2 {
-			continue
-		}
-		idx := len(rows)
-		row := GPUMetricRow{GPUIndex: idx}
-		// rocm-smi CSV columns vary; extract what we can
-		for i, p := range parts {
-			p = strings.TrimSpace(p)
-			switch {
-			case i == 0:
-				// device index like "card0" or "0"
-			case strings.Contains(strings.ToLower(p), "n/a"):
-				// skip N/A
-			default:
-				// Try to match by position heuristic: temp, use%, memuse%, power
-				v := parseGPUFloat(p)
-				switch {
-				case i == 1 && row.TempC == 0:
-					row.TempC = v
-				case i == 2 && row.UsagePct == 0:
-					row.UsagePct = v
-				case i == 3 && row.MemUsagePct == 0:
-					row.MemUsagePct = v
-				case i == 4 && row.PowerW == 0:
-					row.PowerW = v
+	lines := strings.Split(strings.TrimSpace(string(out)), "\n")
+	if len(lines) < 2 {
+		return nil, fmt.Errorf("rocm-smi: insufficient output")
+	}
+
+	// Parse header to find column indices by name.
+	headers := strings.Split(lines[0], ",")
+	colIdx := func(keywords ...string) int {
+		for i, h := range headers {
+			hl := strings.ToLower(strings.TrimSpace(h))
+			for _, kw := range keywords {
+				if strings.Contains(hl, kw) {
+					return i
 				}
 			}
 		}
+		return -1
+	}
+	idxTemp := colIdx("sensor edge", "temperature (c)", "temp")
+	idxUse := colIdx("gpu use (%)")
+	idxMem := colIdx("vram%", "memory allocated")
+	idxPow := colIdx("average graphics package power", "power (w)")
+
+	var rows []GPUMetricRow
+	for _, line := range lines[1:] {
+		line = strings.TrimSpace(line)
+		if line == "" {
+			continue
+		}
+		parts := strings.Split(line, ",")
+		idx := len(rows)
+		row := GPUMetricRow{GPUIndex: idx}
+		get := func(i int) float64 {
+			if i < 0 || i >= len(parts) {
+				return 0
+			}
+			v := strings.TrimSpace(parts[i])
+			if strings.EqualFold(v, "n/a") {
+				return 0
+			}
+			return parseGPUFloat(v)
+		}
+		row.TempC = get(idxTemp)
+		row.UsagePct = get(idxUse)
+		row.MemUsagePct = get(idxMem)
+		row.PowerW = get(idxPow)
 		rows = append(rows, row)
 	}
 	if len(rows) == 0 {

audit/internal/platform/live_metrics.go

@@ -304,6 +304,7 @@ func classifyLiveTempGroup(chip, name string) string {
 		strings.Contains(text, "x86_pkg_temp"),
 		strings.Contains(text, "tctl"),
 		strings.Contains(text, "tdie"),
+		strings.Contains(text, "tccd"),
 		strings.Contains(text, "cpu"),
 		strings.Contains(text, "peci"):
 		return "cpu"

audit/internal/platform/network.go

@@ -134,7 +134,17 @@ func (s *System) RestoreNetworkSnapshot(snapshot NetworkSnapshot) error {
 		if len(fields) == 0 {
 			continue
 		}
-		args := append([]string{"route", "add"}, fields...)
+		// Strip state flags that ip-route(8) does not accept as add arguments.
+		filtered := fields[:0]
+		for _, f := range fields {
+			switch f {
+			case "linkdown", "dead", "onlink", "pervasive":
+				// skip
+			default:
+				filtered = append(filtered, f)
+			}
+		}
+		args := append([]string{"route", "add"}, filtered...)
 		if raw, err := exec.Command("ip", args...).CombinedOutput(); err != nil {
 			detail := strings.TrimSpace(string(raw))
 			if detail != "" {

audit/internal/platform/sat.go

@@ -33,6 +33,10 @@ var (
 		"/opt/rocm/libexec/rocm_smi/rocm_smi.py",
 		"/opt/rocm-*/libexec/rocm_smi/rocm_smi.py",
 	}
+	rvsExecutableGlobs = []string{
+		"/opt/rocm/bin/rvs",
+		"/opt/rocm-*/bin/rvs",
+	}
 )
 
 // streamExecOutput runs cmd and streams each output line to logFunc (if non-nil).
@@ -142,13 +146,27 @@ func (s *System) RunAMDStressPack(ctx context.Context, baseDir string, durationS
 	if err := ensureAMDRuntimeReady(); err != nil {
 		return "", err
 	}
+	// Write RVS GST config to a temp file
+	rvsCfg := fmt.Sprintf(`actions:
+- name: gst_stress
+  device: all
+  module: gst
+  parallel: true
+  duration: %d
+  copy_matrix: false
+  target_stress: 90
+  matrix_size_a: 8640
+  matrix_size_b: 8640
+  matrix_size_c: 8640
+`, seconds*1000)
+	cfgFile := "/tmp/bee-amd-gst.conf"
+	_ = os.WriteFile(cfgFile, []byte(rvsCfg), 0644)
+
 	return runAcceptancePackCtx(ctx, baseDir, "gpu-amd-stress", []satJob{
 		{name: "01-rocm-smi.log", cmd: []string{"rocm-smi"}},
 		{name: "02-rocm-bandwidth-test.log", cmd: []string{"rocm-bandwidth-test"}},
-		{name: fmt.Sprintf("03-rocm-smi-monitor-%ds.log", seconds), cmd: []string{
-			"bash", "-lc",
-			fmt.Sprintf("end=$((SECONDS+%d)); while [ \"$SECONDS\" -lt \"$end\" ]; do rocm-smi --showtemp --showpower --csv; sleep 1; done", seconds),
-		}},
+		{name: fmt.Sprintf("03-rvs-gst-%ds.log", seconds), cmd: []string{"rvs", "-c", cfgFile}},
+		{name: fmt.Sprintf("04-rocm-smi-after.log"), cmd: []string{"rocm-smi", "--showtemp", "--showpower", "--csv"}},
 	}, logFunc)
 }
 
@@ -606,10 +624,23 @@ func resolveSATCommand(cmd []string) ([]string, error) {
 	if len(cmd) == 0 {
 		return nil, errors.New("empty SAT command")
 	}
-	if cmd[0] != "rocm-smi" {
-		return cmd, nil
+	switch cmd[0] {
+	case "rocm-smi":
+		return resolveROCmSMICommand(cmd[1:]...)
+	case "rvs":
+		return resolveRVSCommand(cmd[1:]...)
 	}
-	return resolveROCmSMICommand(cmd[1:]...)
+	return cmd, nil
+}
+
+func resolveRVSCommand(args ...string) ([]string, error) {
+	if path, err := satLookPath("rvs"); err == nil {
+		return append([]string{path}, args...), nil
+	}
+	for _, path := range expandExistingPaths(rvsExecutableGlobs) {
+		return append([]string{path}, args...), nil
+	}
+	return nil, errors.New("rvs not found in PATH or under /opt/rocm")
 }
 
 func resolveROCmSMICommand(args ...string) ([]string, error) {

audit/internal/platform/sat_fan_stress.go

@@ -322,7 +322,9 @@ func sampleFanSpeeds() ([]FanReading, error) {
 }
 
 // parseFanSpeeds parses "ipmitool sdr type Fan" output.
-// Line format: "FAN1             | 2400.000   | RPM        | ok"
+// Handles two formats:
+//   Old: "FAN1 | 2400.000 | RPM | ok"           (value in col[1], unit in col[2])
+//   New: "FAN1 | 41h | ok | 29.1 | 4340 RPM"   (value+unit combined in last col)
 func parseFanSpeeds(raw string) []FanReading {
 	var fans []FanReading
 	for _, line := range strings.Split(strings.TrimSpace(raw), "\n") {
@@ -330,25 +332,39 @@ func parseFanSpeeds(raw string) []FanReading {
 		if len(parts) < 2 {
 			continue
 		}
-		unit := ""
-		if len(parts) >= 3 {
-			unit = strings.TrimSpace(parts[2])
+		name := strings.TrimSpace(parts[0])
+		// Find the first field that contains "RPM" (either as a standalone unit or inline)
+		rpmVal := 0.0
+		found := false
+		for _, p := range parts[1:] {
+			p = strings.TrimSpace(p)
+			if !strings.Contains(strings.ToUpper(p), "RPM") {
+				continue
+			}
+			if strings.EqualFold(p, "RPM") {
+				continue // unit-only column in old format; value is in previous field
+			}
+			val, err := parseFanRPMValue(p)
+			if err == nil {
+				rpmVal = val
+				found = true
+				break
+			}
 		}
-		valStr := strings.TrimSpace(parts[1])
-		if !strings.EqualFold(unit, "RPM") && !strings.Contains(strings.ToUpper(valStr), "RPM") {
+		// Old format: unit "RPM" is in col[2], value is in col[1]
+		if !found && len(parts) >= 3 && strings.EqualFold(strings.TrimSpace(parts[2]), "RPM") {
+			valStr := strings.TrimSpace(parts[1])
+			if !strings.EqualFold(valStr, "na") && !strings.EqualFold(valStr, "disabled") && valStr != "" {
+				if val, err := parseFanRPMValue(valStr); err == nil {
+					rpmVal = val
+					found = true
+				}
+			}
+		}
+		if !found {
 			continue
 		}
-		if strings.EqualFold(valStr, "na") || strings.EqualFold(valStr, "disabled") || valStr == "" {
-			continue
-		}
-		val, err := parseFanRPMValue(valStr)
-		if err != nil {
-			continue
-		}
-		fans = append(fans, FanReading{
-			Name: strings.TrimSpace(parts[0]),
-			RPM:  val,
-		})
+		fans = append(fans, FanReading{Name: name, RPM: rpmVal})
 	}
 	return fans
 }

audit/internal/webui/api.go

@@ -592,7 +592,7 @@ func (h *handler) handleAPIMetricsStream(w http.ResponseWriter, r *http.Request)
 	if !sseStart(w) {
 		return
 	}
-	ticker := time.NewTicker(time.Second)
+	ticker := time.NewTicker(1 * time.Second)
 	defer ticker.Stop()
 	for {
 		select {
@@ -600,47 +600,10 @@ func (h *handler) handleAPIMetricsStream(w http.ResponseWriter, r *http.Request)
 			return
 		case <-ticker.C:
 			sample := platform.SampleLiveMetrics()
-
-			// Feed server ring buffers
-			for _, t := range sample.Temps {
-				switch t.Group {
-				case "cpu":
-					h.pushNamedMetricRing(&h.cpuTempRings, t.Name, t.Celsius)
-				case "ambient":
-					h.pushNamedMetricRing(&h.ambientTempRings, t.Name, t.Celsius)
-				}
+			h.feedRings(sample)
+			if h.metricsDB != nil {
+				_ = h.metricsDB.Write(sample)
 			}
-			h.ringPower.push(sample.PowerW)
-			h.ringCPULoad.push(sample.CPULoadPct)
-			h.ringMemLoad.push(sample.MemLoadPct)
-
-			// Feed fan ring buffers (grow on first sight)
-			h.ringsMu.Lock()
-			for i, fan := range sample.Fans {
-				for len(h.ringFans) <= i {
-					h.ringFans = append(h.ringFans, newMetricsRing(120))
-					h.fanNames = append(h.fanNames, fan.Name)
-				}
-				h.ringFans[i].push(float64(fan.RPM))
-			}
-			// Feed per-GPU ring buffers (grow on first sight)
-			for _, gpu := range sample.GPUs {
-				idx := gpu.GPUIndex
-				for len(h.gpuRings) <= idx {
-					h.gpuRings = append(h.gpuRings, &gpuRings{
-						Temp:    newMetricsRing(120),
-						Util:    newMetricsRing(120),
-						MemUtil: newMetricsRing(120),
-						Power:   newMetricsRing(120),
-					})
-				}
-				h.gpuRings[idx].Temp.push(gpu.TempC)
-				h.gpuRings[idx].Util.push(gpu.UsagePct)
-				h.gpuRings[idx].MemUtil.push(gpu.MemUsagePct)
-				h.gpuRings[idx].Power.push(gpu.PowerW)
-			}
-			h.ringsMu.Unlock()
-
 			b, err := json.Marshal(sample)
 			if err != nil {
 				continue
@@ -652,6 +615,46 @@ func (h *handler) handleAPIMetricsStream(w http.ResponseWriter, r *http.Request)
 	}
 }
 
+// feedRings pushes one sample into all in-memory ring buffers.
+func (h *handler) feedRings(sample platform.LiveMetricSample) {
+	for _, t := range sample.Temps {
+		switch t.Group {
+		case "cpu":
+			h.pushNamedMetricRing(&h.cpuTempRings, t.Name, t.Celsius)
+		case "ambient":
+			h.pushNamedMetricRing(&h.ambientTempRings, t.Name, t.Celsius)
+		}
+	}
+	h.ringPower.push(sample.PowerW)
+	h.ringCPULoad.push(sample.CPULoadPct)
+	h.ringMemLoad.push(sample.MemLoadPct)
+
+	h.ringsMu.Lock()
+	for i, fan := range sample.Fans {
+		for len(h.ringFans) <= i {
+			h.ringFans = append(h.ringFans, newMetricsRing(120))
+			h.fanNames = append(h.fanNames, fan.Name)
+		}
+		h.ringFans[i].push(float64(fan.RPM))
+	}
+	for _, gpu := range sample.GPUs {
+		idx := gpu.GPUIndex
+		for len(h.gpuRings) <= idx {
+			h.gpuRings = append(h.gpuRings, &gpuRings{
+				Temp:    newMetricsRing(120),
+				Util:    newMetricsRing(120),
+				MemUtil: newMetricsRing(120),
+				Power:   newMetricsRing(120),
+			})
+		}
+		h.gpuRings[idx].Temp.push(gpu.TempC)
+		h.gpuRings[idx].Util.push(gpu.UsagePct)
+		h.gpuRings[idx].MemUtil.push(gpu.MemUsagePct)
+		h.gpuRings[idx].Power.push(gpu.PowerW)
+	}
+	h.ringsMu.Unlock()
+}
+
 func (h *handler) pushNamedMetricRing(dst *[]*namedMetricsRing, name string, value float64) {
 	if name == "" {
 		return

audit/internal/webui/metricsdb.go

@@ -0,0 +1,320 @@
+package webui
+
+import (
+	"database/sql"
+	"encoding/csv"
+	"fmt"
+	"io"
+	"strconv"
+	"time"
+
+	"bee/audit/internal/platform"
+	_ "modernc.org/sqlite"
+)
+
+const metricsDBPath = "/appdata/bee/metrics.db"
+const metricsKeepDuration = 24 * time.Hour
+
+// MetricsDB persists live metric samples to SQLite.
+type MetricsDB struct {
+	db *sql.DB
+}
+
+// openMetricsDB opens (or creates) the metrics database at the given path.
+func openMetricsDB(path string) (*MetricsDB, error) {
+	db, err := sql.Open("sqlite", path+"?_journal=WAL&_busy_timeout=5000")
+	if err != nil {
+		return nil, err
+	}
+	db.SetMaxOpenConns(1)
+	if err := initMetricsSchema(db); err != nil {
+		_ = db.Close()
+		return nil, err
+	}
+	return &MetricsDB{db: db}, nil
+}
+
+func initMetricsSchema(db *sql.DB) error {
+	_, err := db.Exec(`
+CREATE TABLE IF NOT EXISTS sys_metrics (
+  ts           INTEGER NOT NULL,
+  cpu_load_pct REAL,
+  mem_load_pct REAL,
+  power_w      REAL,
+  PRIMARY KEY (ts)
+);
+CREATE TABLE IF NOT EXISTS gpu_metrics (
+  ts            INTEGER NOT NULL,
+  gpu_index     INTEGER NOT NULL,
+  temp_c        REAL,
+  usage_pct     REAL,
+  mem_usage_pct REAL,
+  power_w       REAL,
+  PRIMARY KEY (ts, gpu_index)
+);
+CREATE TABLE IF NOT EXISTS fan_metrics (
+  ts   INTEGER NOT NULL,
+  name TEXT NOT NULL,
+  rpm  REAL,
+  PRIMARY KEY (ts, name)
+);
+CREATE TABLE IF NOT EXISTS temp_metrics (
+  ts      INTEGER NOT NULL,
+  name    TEXT NOT NULL,
+  grp     TEXT NOT NULL,
+  celsius REAL,
+  PRIMARY KEY (ts, name)
+);
+`)
+	return err
+}
+
+// Write inserts one sample into all relevant tables.
+func (m *MetricsDB) Write(s platform.LiveMetricSample) error {
+	ts := s.Timestamp.Unix()
+	tx, err := m.db.Begin()
+	if err != nil {
+		return err
+	}
+	defer func() { _ = tx.Rollback() }()
+
+	_, err = tx.Exec(
+		`INSERT OR REPLACE INTO sys_metrics(ts,cpu_load_pct,mem_load_pct,power_w) VALUES(?,?,?,?)`,
+		ts, s.CPULoadPct, s.MemLoadPct, s.PowerW,
+	)
+	if err != nil {
+		return err
+	}
+	for _, g := range s.GPUs {
+		_, err = tx.Exec(
+			`INSERT OR REPLACE INTO gpu_metrics(ts,gpu_index,temp_c,usage_pct,mem_usage_pct,power_w) VALUES(?,?,?,?,?,?)`,
+			ts, g.GPUIndex, g.TempC, g.UsagePct, g.MemUsagePct, g.PowerW,
+		)
+		if err != nil {
+			return err
+		}
+	}
+	for _, f := range s.Fans {
+		_, err = tx.Exec(
+			`INSERT OR REPLACE INTO fan_metrics(ts,name,rpm) VALUES(?,?,?)`,
+			ts, f.Name, f.RPM,
+		)
+		if err != nil {
+			return err
+		}
+	}
+	for _, t := range s.Temps {
+		_, err = tx.Exec(
+			`INSERT OR REPLACE INTO temp_metrics(ts,name,grp,celsius) VALUES(?,?,?,?)`,
+			ts, t.Name, t.Group, t.Celsius,
+		)
+		if err != nil {
+			return err
+		}
+	}
+	return tx.Commit()
+}
+
+// LoadRecent returns up to n samples in chronological order (oldest first).
+// It reconstructs LiveMetricSample from the normalized tables.
+func (m *MetricsDB) LoadRecent(n int) ([]platform.LiveMetricSample, error) {
+	rows, err := m.db.Query(
+		`SELECT ts,cpu_load_pct,mem_load_pct,power_w FROM sys_metrics ORDER BY ts DESC LIMIT ?`, n,
+	)
+	if err != nil {
+		return nil, err
+	}
+	defer rows.Close()
+
+	type sysRow struct {
+		ts          int64
+		cpu, mem, pwr float64
+	}
+	var sysRows []sysRow
+	for rows.Next() {
+		var r sysRow
+		if err := rows.Scan(&r.ts, &r.cpu, &r.mem, &r.pwr); err != nil {
+			continue
+		}
+		sysRows = append(sysRows, r)
+	}
+	if len(sysRows) == 0 {
+		return nil, nil
+	}
+	// Reverse to chronological order
+	for i, j := 0, len(sysRows)-1; i < j; i, j = i+1, j-1 {
+		sysRows[i], sysRows[j] = sysRows[j], sysRows[i]
+	}
+
+	// Collect min/max ts for range query
+	minTS := sysRows[0].ts
+	maxTS := sysRows[len(sysRows)-1].ts
+
+	// Load GPU rows in range
+	type gpuKey struct{ ts int64; idx int }
+	gpuData := map[gpuKey]platform.GPUMetricRow{}
+	gRows, err := m.db.Query(
+		`SELECT ts,gpu_index,temp_c,usage_pct,mem_usage_pct,power_w FROM gpu_metrics WHERE ts>=? AND ts<=? ORDER BY ts,gpu_index`,
+		minTS, maxTS,
+	)
+	if err == nil {
+		defer gRows.Close()
+		for gRows.Next() {
+			var ts int64
+			var g platform.GPUMetricRow
+			if err := gRows.Scan(&ts, &g.GPUIndex, &g.TempC, &g.UsagePct, &g.MemUsagePct, &g.PowerW); err == nil {
+				gpuData[gpuKey{ts, g.GPUIndex}] = g
+			}
+		}
+	}
+
+	// Load fan rows in range
+	type fanKey struct{ ts int64; name string }
+	fanData := map[fanKey]float64{}
+	fRows, err := m.db.Query(
+		`SELECT ts,name,rpm FROM fan_metrics WHERE ts>=? AND ts<=?`, minTS, maxTS,
+	)
+	if err == nil {
+		defer fRows.Close()
+		for fRows.Next() {
+			var ts int64
+			var name string
+			var rpm float64
+			if err := fRows.Scan(&ts, &name, &rpm); err == nil {
+				fanData[fanKey{ts, name}] = rpm
+			}
+		}
+	}
+
+	// Load temp rows in range
+	type tempKey struct{ ts int64; name string }
+	tempData := map[tempKey]platform.TempReading{}
+	tRows, err := m.db.Query(
+		`SELECT ts,name,grp,celsius FROM temp_metrics WHERE ts>=? AND ts<=?`, minTS, maxTS,
+	)
+	if err == nil {
+		defer tRows.Close()
+		for tRows.Next() {
+			var ts int64
+			var t platform.TempReading
+			if err := tRows.Scan(&ts, &t.Name, &t.Group, &t.Celsius); err == nil {
+				tempData[tempKey{ts, t.Name}] = t
+			}
+		}
+	}
+
+	// Collect unique GPU indices and fan names from loaded data (preserve order)
+	seenGPU := map[int]bool{}
+	var gpuIndices []int
+	for k := range gpuData {
+		if !seenGPU[k.idx] {
+			seenGPU[k.idx] = true
+			gpuIndices = append(gpuIndices, k.idx)
+		}
+	}
+	seenFan := map[string]bool{}
+	var fanNames []string
+	for k := range fanData {
+		if !seenFan[k.name] {
+			seenFan[k.name] = true
+			fanNames = append(fanNames, k.name)
+		}
+	}
+	seenTemp := map[string]bool{}
+	var tempNames []string
+	for k := range tempData {
+		if !seenTemp[k.name] {
+			seenTemp[k.name] = true
+			tempNames = append(tempNames, k.name)
+		}
+	}
+
+	samples := make([]platform.LiveMetricSample, len(sysRows))
+	for i, r := range sysRows {
+		s := platform.LiveMetricSample{
+			Timestamp:  time.Unix(r.ts, 0).UTC(),
+			CPULoadPct: r.cpu,
+			MemLoadPct: r.mem,
+			PowerW:     r.pwr,
+		}
+		for _, idx := range gpuIndices {
+			if g, ok := gpuData[gpuKey{r.ts, idx}]; ok {
+				s.GPUs = append(s.GPUs, g)
+			}
+		}
+		for _, name := range fanNames {
+			if rpm, ok := fanData[fanKey{r.ts, name}]; ok {
+				s.Fans = append(s.Fans, platform.FanReading{Name: name, RPM: rpm})
+			}
+		}
+		for _, name := range tempNames {
+			if t, ok := tempData[tempKey{r.ts, name}]; ok {
+				s.Temps = append(s.Temps, t)
+			}
+		}
+		samples[i] = s
+	}
+	return samples, nil
+}
+
+// Prune deletes samples older than keepDuration.
+func (m *MetricsDB) Prune(keepDuration time.Duration) {
+	cutoff := time.Now().Add(-keepDuration).Unix()
+	for _, table := range []string{"sys_metrics", "gpu_metrics", "fan_metrics", "temp_metrics"} {
+		_, _ = m.db.Exec(fmt.Sprintf("DELETE FROM %s WHERE ts < ?", table), cutoff)
+	}
+}
+
+// ExportCSV writes all sys+gpu data as CSV to w.
+func (m *MetricsDB) ExportCSV(w io.Writer) error {
+	rows, err := m.db.Query(`
+		SELECT s.ts, s.cpu_load_pct, s.mem_load_pct, s.power_w,
+		       g.gpu_index, g.temp_c, g.usage_pct, g.mem_usage_pct, g.power_w
+		FROM sys_metrics s
+		LEFT JOIN gpu_metrics g ON g.ts = s.ts
+		ORDER BY s.ts, g.gpu_index
+	`)
+	if err != nil {
+		return err
+	}
+	defer rows.Close()
+
+	cw := csv.NewWriter(w)
+	_ = cw.Write([]string{"ts", "cpu_load_pct", "mem_load_pct", "sys_power_w", "gpu_index", "gpu_temp_c", "gpu_usage_pct", "gpu_mem_pct", "gpu_power_w"})
+	for rows.Next() {
+		var ts int64
+		var cpu, mem, pwr float64
+		var gpuIdx sql.NullInt64
+		var gpuTemp, gpuUse, gpuMem, gpuPow sql.NullFloat64
+		if err := rows.Scan(&ts, &cpu, &mem, &pwr, &gpuIdx, &gpuTemp, &gpuUse, &gpuMem, &gpuPow); err != nil {
+			continue
+		}
+		row := []string{
+			strconv.FormatInt(ts, 10),
+			strconv.FormatFloat(cpu, 'f', 2, 64),
+			strconv.FormatFloat(mem, 'f', 2, 64),
+			strconv.FormatFloat(pwr, 'f', 1, 64),
+		}
+		if gpuIdx.Valid {
+			row = append(row,
+				strconv.FormatInt(gpuIdx.Int64, 10),
+				strconv.FormatFloat(gpuTemp.Float64, 'f', 1, 64),
+				strconv.FormatFloat(gpuUse.Float64, 'f', 1, 64),
+				strconv.FormatFloat(gpuMem.Float64, 'f', 1, 64),
+				strconv.FormatFloat(gpuPow.Float64, 'f', 1, 64),
+			)
+		} else {
+			row = append(row, "", "", "", "", "")
+		}
+		_ = cw.Write(row)
+	}
+	cw.Flush()
+	return cw.Error()
+}
+
+// Close closes the database.
+func (m *MetricsDB) Close() { _ = m.db.Close() }
+
+func nullFloat(v float64) sql.NullFloat64 {
+	return sql.NullFloat64{Float64: v, Valid: true}
+}

audit/internal/webui/pages.go

@@ -61,7 +61,8 @@ tbody tr:hover td{background:rgba(0,0,0,.03)}
 .badge-err{background:var(--crit-bg);color:var(--crit-fg);border:1px solid var(--crit-border)}
 .badge-unknown{background:var(--surface-2);color:var(--muted);border:1px solid var(--border)}
 /* Output terminal */
-.terminal{background:#1b1c1d;border:1px solid rgba(0,0,0,.2);border-radius:4px;padding:14px;font-family:monospace;font-size:12px;color:#b5cea8;max-height:400px;overflow-y:auto;white-space:pre-wrap;word-break:break-all}
+.terminal{background:#1b1c1d;border:1px solid rgba(0,0,0,.2);border-radius:4px;padding:14px;font-family:monospace;font-size:12px;color:#b5cea8;max-height:400px;overflow-y:auto;white-space:pre-wrap;word-break:break-all;user-select:text;-webkit-user-select:text}
+.terminal-wrap{position:relative}.terminal-copy{position:absolute;top:6px;right:6px;background:#2d2f30;border:1px solid #444;color:#aaa;font-size:11px;padding:2px 8px;border-radius:3px;cursor:pointer;opacity:.7}.terminal-copy:hover{opacity:1}
 /* Forms */
 .form-row{margin-bottom:14px}
 .form-row label{display:block;font-size:12px;color:var(--muted);margin-bottom:5px;font-weight:700}
@@ -187,6 +188,16 @@ func renderPage(page string, opts HandlerOptions) string {
 		body +
 		`</div></div>` +
 		renderAuditModal() +
+		`<script>
+// Add copy button to every .terminal on the page
+document.querySelectorAll('.terminal').forEach(function(t){
+  var w=document.createElement('div');w.className='terminal-wrap';
+  t.parentNode.insertBefore(w,t);w.appendChild(t);
+  var btn=document.createElement('button');btn.className='terminal-copy';btn.textContent='Copy';
+  btn.onclick=function(){navigator.clipboard.writeText(t.textContent).then(function(){btn.textContent='Copied!';setTimeout(function(){btn.textContent='Copy';},1500);});};
+  w.appendChild(btn);
+});
+</script>` +
 		`</body></html>`
 }
 
@@ -392,12 +403,6 @@ func renderMetrics() string {
   </div>
 </div>
 
-<div class="card" style="margin-bottom:16px">
-  <div class="card-head">Temperature — GPUs</div>
-  <div class="card-body" style="padding:8px">
-    <img id="chart-server-temp-gpu" src="/api/metrics/chart/server-temp-gpu.svg" style="width:100%;display:block;border-radius:6px" alt="GPU temperature">
-  </div>
-</div>
 
 <div class="card" style="margin-bottom:16px">
   <div class="card-head">Temperature — Ambient Sensors</div>
@@ -413,79 +418,57 @@ func renderMetrics() string {
   </div>
 </div>
 
-<div class="card" style="margin-bottom:16px">
+<div id="card-server-fans" class="card" style="margin-bottom:16px;display:none">
   <div class="card-head">Server — Fan RPM</div>
   <div class="card-body" style="padding:8px">
     <img id="chart-server-fans" src="/api/metrics/chart/server-fans.svg" style="width:100%;display:block;border-radius:6px" alt="Fan RPM">
-    <div id="sys-table" style="margin-top:8px;font-size:12px"></div>
   </div>
 </div>
 
-<div id="gpu-charts"></div>
+<div class="card" style="margin-bottom:16px">
+  <div class="card-head">GPU — Compute Load</div>
+  <div class="card-body" style="padding:8px">
+    <img id="chart-gpu-all-load" src="/api/metrics/chart/gpu-all-load.svg" style="width:100%;display:block;border-radius:6px" alt="GPU compute load">
+  </div>
+</div>
+<div class="card" style="margin-bottom:16px">
+  <div class="card-head">GPU — Memory Load</div>
+  <div class="card-body" style="padding:8px">
+    <img id="chart-gpu-all-memload" src="/api/metrics/chart/gpu-all-memload.svg" style="width:100%;display:block;border-radius:6px" alt="GPU memory load">
+  </div>
+</div>
+<div class="card" style="margin-bottom:16px">
+  <div class="card-head">GPU — Power</div>
+  <div class="card-body" style="padding:8px">
+    <img id="chart-gpu-all-power" src="/api/metrics/chart/gpu-all-power.svg" style="width:100%;display:block;border-radius:6px" alt="GPU power">
+  </div>
+</div>
+<div class="card" style="margin-bottom:16px">
+  <div class="card-head">GPU — Temperature</div>
+  <div class="card-body" style="padding:8px">
+    <img id="chart-gpu-all-temp" src="/api/metrics/chart/gpu-all-temp.svg" style="width:100%;display:block;border-radius:6px" alt="GPU temperature">
+  </div>
+</div>
 
 <script>
-let knownGPUs = [];
-
 function refreshCharts() {
   const t = '?t=' + Date.now();
-  ['chart-server-load','chart-server-temp-cpu','chart-server-temp-gpu','chart-server-temp-ambient','chart-server-power','chart-server-fans'].forEach(id => {
+  ['chart-server-load','chart-server-temp-cpu','chart-server-temp-gpu','chart-server-temp-ambient','chart-server-power','chart-server-fans',
+   'chart-gpu-all-load','chart-gpu-all-memload','chart-gpu-all-power','chart-gpu-all-temp'].forEach(id => {
     const el = document.getElementById(id);
     if (el) el.src = el.src.split('?')[0] + t;
   });
-  knownGPUs.forEach(idx => {
-    ['load','power'].forEach(kind => {
-      const el = document.getElementById('chart-gpu-' + idx + '-' + kind);
-      if (el) el.src = el.src.split('?')[0] + t;
-    });
-  });
 }
-setInterval(refreshCharts, 2000);
+setInterval(refreshCharts, 3000);
 
 const es = new EventSource('/api/metrics/stream');
 es.addEventListener('metrics', e => {
   const d = JSON.parse(e.data);
 
-  // Add GPU chart cards as GPUs appear
-  (d.gpus||[]).forEach(g => {
-    if (knownGPUs.includes(g.index)) return;
-    knownGPUs.push(g.index);
-    const div = document.createElement('div');
-    div.className = 'card';
-    div.style.marginBottom = '16px';
-    div.innerHTML =
-      '<div class="card-head">GPU ' + g.index + ' — Load</div>' +
-      '<div class="card-body" style="padding:8px">' +
-        '<img id="chart-gpu-' + g.index + '-load" src="/api/metrics/chart/gpu/' + g.index + '-load.svg" style="width:100%;display:block;border-radius:6px" alt="GPU ' + g.index + ' load">' +
-      '</div>' +
-      '<div class="card-head">GPU ' + g.index + ' — Power</div>' +
-      '<div class="card-body" style="padding:8px">' +
-        '<img id="chart-gpu-' + g.index + '-power" src="/api/metrics/chart/gpu/' + g.index + '-power.svg" style="width:100%;display:block;border-radius:6px" alt="GPU ' + g.index + ' power">' +
-        '<div id="gpu-table-' + g.index + '" style="margin-top:8px;font-size:12px"></div>' +
-      '</div>';
-    document.getElementById('gpu-charts').appendChild(div);
-  });
+  // Show/hide Fan RPM card based on data availability
+  const fanCard = document.getElementById('card-server-fans');
+  if (fanCard) fanCard.style.display = (d.fans && d.fans.length > 0) ? '' : 'none';
 
-  // Update numeric tables
-  let sysHTML = '';
-  (d.temps||[]).filter(t => t.group === 'cpu').forEach(t => {
-    sysHTML += '<tr><td>'+t.name+'</td><td>'+t.celsius.toFixed(1)+'°C</td></tr>';
-  });
-  if (d.cpu_load_pct) sysHTML += '<tr><td>CPU Load</td><td>'+d.cpu_load_pct.toFixed(1)+'%</td></tr>';
-  if (d.mem_load_pct) sysHTML += '<tr><td>Mem Load</td><td>'+d.mem_load_pct.toFixed(1)+'%</td></tr>';
-  (d.fans||[]).forEach(f => sysHTML += '<tr><td>'+f.name+'</td><td>'+f.rpm+' RPM</td></tr>');
-  if (d.power_w) sysHTML += '<tr><td>Power</td><td>'+d.power_w.toFixed(0)+' W</td></tr>';
-  const st = document.getElementById('sys-table');
-  if (st) st.innerHTML = sysHTML ? '<table>'+sysHTML+'</table>' : '<p style="color:var(--muted)">No sensor data (ipmitool/sensors required)</p>';
-
-  (d.gpus||[]).forEach(g => {
-    const t = document.getElementById('gpu-table-' + g.index);
-    if (!t) return;
-    t.innerHTML = '<table>' +
-      '<tr><td>Temp</td><td>'+g.temp_c+'°C</td>' +
-      '<td>Load</td><td>'+g.usage_pct+'%</td>' +
-      '<td>Mem</td><td>'+g.mem_usage_pct+'%</td>' +
-      '<td>Power</td><td>'+g.power_w+' W</td></tr></table>';
-  });
 });
 es.onerror = () => {};
 </script>`

audit/internal/webui/server.go

@@ -132,6 +132,8 @@ type handler struct {
 	// per-GPU rings (index = GPU index)
 	gpuRings []*gpuRings
 	ringsMu  sync.Mutex
+	// metrics persistence (nil if DB unavailable)
+	metricsDB *MetricsDB
 	// install job (at most one at a time)
 	installJob *jobState
 	installMu  sync.Mutex
@@ -158,11 +160,24 @@ func NewHandler(opts HandlerOptions) http.Handler {
 		ringMemLoad: newMetricsRing(120),
 		ringPower:   newMetricsRing(120),
 	}
+
+	// Open metrics DB and pre-fill ring buffers from history.
+	if db, err := openMetricsDB(metricsDBPath); err == nil {
+		h.metricsDB = db
+		db.Prune(metricsKeepDuration)
+		if samples, err := db.LoadRecent(120); err == nil {
+			for _, s := range samples {
+				h.feedRings(s)
+			}
+		}
+	}
+
 	globalQueue.startWorker(&opts)
 	mux := http.NewServeMux()
 
 	// ── Infrastructure ──────────────────────────────────────────────────────
 	mux.HandleFunc("GET /healthz", h.handleHealthz)
+	mux.HandleFunc("GET /api/ready", h.handleReady)
 
 	// ── Existing read-only endpoints (preserved for compatibility) ──────────
 	mux.HandleFunc("GET /audit.json", h.handleAuditJSON)
@@ -230,9 +245,10 @@ func NewHandler(opts HandlerOptions) http.Handler {
 	mux.HandleFunc("POST /api/install/run", h.handleAPIInstallRun)
 	mux.HandleFunc("GET /api/install/stream", h.handleAPIInstallStream)
 
-	// Metrics — SSE stream of live sensor data + server-side SVG charts
+	// Metrics — SSE stream of live sensor data + server-side SVG charts + CSV export
 	mux.HandleFunc("GET /api/metrics/stream", h.handleAPIMetricsStream)
 	mux.HandleFunc("GET /api/metrics/chart/", h.handleMetricsChartSVG)
+	mux.HandleFunc("GET /api/metrics/export.csv", h.handleAPIMetricsExportCSV)
 
 	// Reanimator chart static assets (viewer template expects /static/*)
 	mux.Handle("GET /static/", http.StripPrefix("/static/", web.Static()))
@@ -448,7 +464,80 @@ func (h *handler) handleMetricsChartSVG(w http.ResponseWriter, r *http.Request)
 		yMin = floatPtr(0)
 		yMax = autoMax120(datasets...)
 
-	// ── GPU sub-charts ────────────────────────────────────────────────────
+	// ── Combined GPU charts (all GPUs on one chart) ───────────────────────
+	case path == "gpu-all-load":
+		title = "GPU Compute Load"
+		h.ringsMu.Lock()
+		for idx, gr := range h.gpuRings {
+			if gr == nil {
+				continue
+			}
+			vUtil, l := gr.Util.snapshot()
+			datasets = append(datasets, vUtil)
+			names = append(names, fmt.Sprintf("GPU %d", idx))
+			if len(labels) == 0 {
+				labels = l
+			}
+		}
+		h.ringsMu.Unlock()
+		yMin = floatPtr(0)
+		yMax = floatPtr(100)
+
+	case path == "gpu-all-memload":
+		title = "GPU Memory Load"
+		h.ringsMu.Lock()
+		for idx, gr := range h.gpuRings {
+			if gr == nil {
+				continue
+			}
+			vMem, l := gr.MemUtil.snapshot()
+			datasets = append(datasets, vMem)
+			names = append(names, fmt.Sprintf("GPU %d", idx))
+			if len(labels) == 0 {
+				labels = l
+			}
+		}
+		h.ringsMu.Unlock()
+		yMin = floatPtr(0)
+		yMax = floatPtr(100)
+
+	case path == "gpu-all-power":
+		title = "GPU Power"
+		h.ringsMu.Lock()
+		for idx, gr := range h.gpuRings {
+			if gr == nil {
+				continue
+			}
+			vPow, l := gr.Power.snapshot()
+			datasets = append(datasets, vPow)
+			names = append(names, fmt.Sprintf("GPU %d", idx))
+			if len(labels) == 0 {
+				labels = l
+			}
+		}
+		h.ringsMu.Unlock()
+		yMin = floatPtr(0)
+		yMax = autoMax120(datasets...)
+
+	case path == "gpu-all-temp":
+		title = "GPU Temperature"
+		h.ringsMu.Lock()
+		for idx, gr := range h.gpuRings {
+			if gr == nil {
+				continue
+			}
+			vTemp, l := gr.Temp.snapshot()
+			datasets = append(datasets, vTemp)
+			names = append(names, fmt.Sprintf("GPU %d", idx))
+			if len(labels) == 0 {
+				labels = l
+			}
+		}
+		h.ringsMu.Unlock()
+		yMin = floatPtr(0)
+		yMax = autoMax120(datasets...)
+
+	// ── Per-GPU sub-charts ────────────────────────────────────────────────
 	case strings.HasPrefix(path, "gpu/"):
 		rest := strings.TrimPrefix(path, "gpu/")
 		// rest is either "{idx}-load", "{idx}-temp", "{idx}-power", or legacy "{idx}"
@@ -544,6 +633,16 @@ func renderChartSVG(title string, datasets [][]float64, names []string, labels [
 			datasets[i] = make([]float64, n)
 		}
 	}
+	// Append global min/avg/max to title.
+	mn, avg, mx := globalStats(datasets)
+	if mx > 0 {
+		title = fmt.Sprintf("%s    ↓%s  ~%s  ↑%s",
+			title,
+			chartLegendNumber(mn),
+			chartLegendNumber(avg),
+			chartLegendNumber(mx),
+		)
+	}
 	title = sanitizeChartText(title)
 	names = sanitizeChartTexts(names)
 	sparse := sanitizeChartTexts(sparseLabels(labels, 6))
@@ -552,6 +651,9 @@ func renderChartSVG(title string, datasets [][]float64, names []string, labels [
 	opt.Title = gocharts.TitleOption{Text: title}
 	opt.XAxis.Labels = sparse
 	opt.Legend = gocharts.LegendOption{SeriesNames: names}
+	opt.Symbol = gocharts.SymbolNone
+	// Right padding: reserve space for the MarkLine label (library recommendation).
+	opt.Padding = gocharts.NewBox(20, 20, 80, 20)
 	if yMin != nil || yMax != nil {
 		opt.YAxis = []gocharts.YAxisOption{{
 			Min:            yMin,
@@ -560,6 +662,12 @@ func renderChartSVG(title string, datasets [][]float64, names []string, labels [
 		}}
 	}
 
+	// Add a single peak mark line on the series that holds the global maximum.
+	peakIdx, _ := globalPeakSeries(datasets)
+	if peakIdx >= 0 && peakIdx < len(opt.SeriesList) {
+		opt.SeriesList[peakIdx].MarkLine = gocharts.NewMarkLine(gocharts.SeriesMarkTypeMax)
+	}
+
 	p := gocharts.NewPainter(gocharts.PainterOptions{
 		OutputFormat: gocharts.ChartOutputSVG,
 		Width:        1400,
@@ -571,6 +679,48 @@ func renderChartSVG(title string, datasets [][]float64, names []string, labels [
 	return p.Bytes()
 }
 
+// globalPeakSeries returns the index of the series containing the global maximum
+// value across all datasets, and that maximum value.
+func globalPeakSeries(datasets [][]float64) (idx int, peak float64) {
+	idx = -1
+	for i, ds := range datasets {
+		for _, v := range ds {
+			if v > peak {
+				peak = v
+				idx = i
+			}
+		}
+	}
+	return idx, peak
+}
+
+// globalStats returns min, average, and max across all values in all datasets.
+func globalStats(datasets [][]float64) (mn, avg, mx float64) {
+	var sum float64
+	var count int
+	first := true
+	for _, ds := range datasets {
+		for _, v := range ds {
+			if first {
+				mn, mx = v, v
+				first = false
+			}
+			if v < mn {
+				mn = v
+			}
+			if v > mx {
+				mx = v
+			}
+			sum += v
+			count++
+		}
+	}
+	if count > 0 {
+		avg = sum / float64(count)
+	}
+	return mn, avg, mx
+}
+
 func sanitizeChartText(s string) string {
 	if s == "" {
 		return ""
@@ -652,11 +802,79 @@ func sparseLabels(labels []string, n int) []string {
 	return out
 }
 
+func (h *handler) handleAPIMetricsExportCSV(w http.ResponseWriter, r *http.Request) {
+	if h.metricsDB == nil {
+		http.Error(w, "metrics database not available", http.StatusServiceUnavailable)
+		return
+	}
+	w.Header().Set("Content-Type", "text/csv; charset=utf-8")
+	w.Header().Set("Content-Disposition", `attachment; filename="bee-metrics.csv"`)
+	w.Header().Set("Cache-Control", "no-store")
+	_ = h.metricsDB.ExportCSV(w)
+}
+
 // ── Page handler ─────────────────────────────────────────────────────────────
 
+func (h *handler) handleReady(w http.ResponseWriter, r *http.Request) {
+	w.Header().Set("Cache-Control", "no-store")
+	if _, err := os.Stat(h.opts.AuditPath); err != nil {
+		w.WriteHeader(http.StatusServiceUnavailable)
+		_, _ = w.Write([]byte("starting"))
+		return
+	}
+	w.WriteHeader(http.StatusOK)
+	_, _ = w.Write([]byte("ready"))
+}
+
+const loadingPageHTML = `<!DOCTYPE html>
+<html lang="en">
+<head>
+<meta charset="UTF-8">
+<title>EASY-BEE</title>
+<style>
+*{margin:0;padding:0;box-sizing:border-box}
+html,body{height:100%;background:#0f1117;display:flex;align-items:center;justify-content:center;font-family:'Courier New',monospace;color:#e2e8f0}
+.logo{font-size:13px;line-height:1.4;color:#f6c90e;margin-bottom:48px;white-space:pre}
+.spinner{width:48px;height:48px;border:4px solid #2d3748;border-top-color:#f6c90e;border-radius:50%;animation:spin .8s linear infinite;margin:0 auto 24px}
+@keyframes spin{to{transform:rotate(360deg)}}
+.status{font-size:14px;color:#a0aec0;letter-spacing:.05em}
+</style>
+</head>
+<body>
+<div style="text-align:center">
+  <div class="logo">  ███████╗ █████╗ ███████╗██╗   ██╗      ██████╗ ███████╗███████╗
+  ██╔════╝██╔══██╗██╔════╝╚██╗ ██╔╝      ██╔══██╗██╔════╝██╔════╝
+  █████╗  ███████║███████╗ ╚████╔╝ █████╗██████╔╝█████╗  █████╗
+  ██╔══╝  ██╔══██║╚════██║  ╚██╔╝  ╚════╝██╔══██╗██╔══╝  ██╔══╝
+  ███████╗██║  ██║███████║   ██║         ██████╔╝███████╗███████╗
+  ╚══════╝╚═╝  ╚═╝╚══════╝   ╚═╝         ╚═════╝ ╚══════╝╚══════╝</div>
+  <div class="spinner"></div>
+  <div class="status" id="s">Starting up...</div>
+</div>
+<script>
+function probe(){
+  fetch('/api/ready',{cache:'no-store'})
+    .then(function(r){
+      if(r.ok){window.location.replace('/');}
+      else{setTimeout(probe,1000);}
+    })
+    .catch(function(){setTimeout(probe,1000);});
+}
+probe();
+</script>
+</body>
+</html>`
+
 func (h *handler) handlePage(w http.ResponseWriter, r *http.Request) {
 	page := strings.TrimPrefix(r.URL.Path, "/")
 	if page == "" {
+		// Serve loading page until audit snapshot exists
+		if _, err := os.Stat(h.opts.AuditPath); err != nil {
+			w.Header().Set("Cache-Control", "no-store")
+			w.Header().Set("Content-Type", "text/html; charset=utf-8")
+			_, _ = w.Write([]byte(loadingPageHTML))
+			return
+		}
 		page = "dashboard"
 	}
 	// Redirect old routes to new names

iso/builder/VERSIONS

@@ -11,5 +11,12 @@ CUDA_USERSPACE_VERSION=13.0.96-1
 DCGM_VERSION=3.3.9
 ROCM_VERSION=6.3.4
 ROCM_SMI_VERSION=7.4.0.60304-76~22.04
+ROCM_BANDWIDTH_TEST_VERSION=1.4.0.60304-76~22.04
+ROCM_VALIDATION_SUITE_VERSION=1.1.0.60304-76~22.04
+ROCBLAS_VERSION=4.3.0.60304-76~22.04
+ROCRAND_VERSION=3.2.0.60304-76~22.04
+HIP_RUNTIME_AMD_VERSION=6.3.42134.60304-76~22.04
+HIPBLASLT_VERSION=0.10.0.60304-76~22.04
+COMGR_VERSION=2.8.0.60304-76~22.04
 GO_VERSION=1.24.0
 AUDIT_VERSION=1.0.0

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

@@ -7,15 +7,16 @@ echo "=== bee chroot setup ==="
 
 ensure_bee_console_user() {
     if id bee >/dev/null 2>&1; then
-        usermod -d /home/bee -s /bin/sh bee 2>/dev/null || true
+        usermod -d /home/bee -s /bin/bash bee 2>/dev/null || true
     else
-        useradd -d /home/bee -m -s /bin/sh -U bee
+        useradd -d /home/bee -m -s /bin/bash -U bee
     fi
 
     mkdir -p /home/bee
     chown -R bee:bee /home/bee
     echo "bee:eeb" | chpasswd
-    usermod -aG sudo,video,input bee 2>/dev/null || true
+    groupadd -f ipmi 2>/dev/null || true
+    usermod -aG sudo,video,input,render,ipmi bee 2>/dev/null || true
 }
 
 ensure_bee_console_user
@@ -46,11 +47,13 @@ chmod +x /usr/local/bin/bee-log-run    2>/dev/null || true
 # Reload udev rules
 udevadm control --reload-rules 2>/dev/null || true
 
-# rocm-smi symlink (package installs to /opt/rocm-*/bin/rocm-smi)
-if [ ! -e /usr/local/bin/rocm-smi ]; then
-    smi_path="$(find /opt -path '*/bin/rocm-smi' -type f 2>/dev/null | sort | tail -1)"
-    [ -n "${smi_path}" ] && ln -sf "${smi_path}" /usr/local/bin/rocm-smi
-fi
+# rocm symlinks (packages install to /opt/rocm-*/bin/)
+for tool in rocm-smi rocm-bandwidth-test rvs; do
+    if [ ! -e /usr/local/bin/${tool} ]; then
+        bin_path="$(find /opt -path "*/bin/${tool}" -type f 2>/dev/null | sort | tail -1)"
+        [ -n "${bin_path}" ] && ln -sf "${bin_path}" /usr/local/bin/${tool}
+    fi
+done
 
 # Create export directory
 mkdir -p /appdata/bee/export

iso/builder/config/hooks/normal/9100-memtest.hook.binary

@@ -0,0 +1,13 @@
+#!/bin/sh
+# Copy memtest86+ binaries from chroot /boot into the ISO boot directory
+# so GRUB can chainload them directly (they must be on the ISO filesystem,
+# not inside the squashfs).
+set -e
+
+for f in memtest86+x64.bin memtest86+x64.efi memtest86+ia32.bin memtest86+ia32.efi; do
+    src="chroot/boot/${f}"
+    if [ -f "${src}" ]; then
+        cp "${src}" "binary/boot/${f}"
+        echo "memtest: copied ${f} to binary/boot/"
+    fi
+done

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

@@ -75,8 +75,15 @@ firmware-qlogic
 # NVIDIA DCGM (Data Center GPU Manager) — dcgmi diag for acceptance testing
 datacenter-gpu-manager=1:%%DCGM_VERSION%%
 
-# AMD ROCm SMI — GPU monitoring for Instinct cards (repo: rocm/apt/6.3.4 jammy)
+# AMD ROCm — GPU monitoring, bandwidth test, and compute stress (RVS GST)
 rocm-smi-lib=%%ROCM_SMI_VERSION%%
+rocm-bandwidth-test=%%ROCM_BANDWIDTH_TEST_VERSION%%
+rocm-validation-suite=%%ROCM_VALIDATION_SUITE_VERSION%%
+rocblas=%%ROCBLAS_VERSION%%
+rocrand=%%ROCRAND_VERSION%%
+hip-runtime-amd=%%HIP_RUNTIME_AMD_VERSION%%
+hipblaslt=%%HIPBLASLT_VERSION%%
+comgr=%%COMGR_VERSION%%
 
 # glibc compat helpers (for any external binaries that need it)
 libc6

iso/overlay/etc/modules-load.d/bee-ipmi.conf

@@ -0,0 +1,3 @@
+# Load IPMI modules for fan/sensor/power monitoring via ipmitool
+ipmi_si
+ipmi_devintf

iso/overlay/etc/systemd/system/bee-web.service

@@ -1,6 +1,6 @@
 [Unit]
 Description=Bee: hardware audit web viewer
-After=bee-network.service bee-audit.service
+After=bee-network.service
 Wants=bee-audit.service
 
 [Service]

iso/overlay/etc/udev/rules.d/99-ipmi.rules

@@ -0,0 +1,2 @@
+# Allow ipmi group to access IPMI device without root
+KERNEL=="ipmi[0-9]*", GROUP="ipmi", MODE="0660"

iso/overlay/usr/local/bin/bee-openbox-session

@@ -2,22 +2,19 @@
 # openbox session: launch tint2 taskbar + chromium, then openbox as WM.
 # This file is used as an xinitrc by bee-desktop.
 
-# Wait for bee-web to be accepting connections (up to 15 seconds)
-i=0
-while [ $i -lt 15 ]; do
-    if curl -sf http://localhost/healthz >/dev/null 2>&1; then
-        break
-    fi
-    sleep 1
-    i=$((i+1))
-done
-
 # Disable screensaver and DPMS
 xset s off
 xset -dpms
 xset s noblank
 
 tint2 &
+# Wait for bee-web to bind (Go starts fast, usually <2s)
+i=0
+while [ $i -lt 30 ]; do
+    if curl -sf http://localhost/healthz >/dev/null 2>&1; then break; fi
+    sleep 1
+    i=$((i+1))
+done
 chromium \
     --disable-infobars \
     --disable-translate \