platform/cmd/webui: add a scriptable test-scenario engine, load scenarios from blackbox USB, GPU status detail inventory fallback

Investigating the CG480-S6053 reboot needed a way to run an ad-hoc load
(nvbandwidth across a specific GPU set) while sampling IPMI/nvidia-smi
telemetry in the background — without hardcoding a one-off test into the
SAT pack code for a single investigation.

- audit/internal/platform/scenario.go: ScenarioSpec/ScenarioJob (JSON,
  no new dependency) + System.RunScenario. "command" jobs run sequential
  or parallel (per-job "parallel" flag); "sampler" jobs run concurrently
  in the background on their own interval until every command job
  finishes or the scenario's timeout elapses. "{{gpus}}" in a command's
  cmd is substituted from that job's gpu_indices. Command jobs are wired
  through the same satJobBoundaryHook/satSyncBracketHook seams the SAT
  job runner uses, so a scenario run gets the same durability treatment
  (evidence that a risky command started/finished reaches blackbox before
  a possible crash, not just whatever streamed to the RAM-backed export
  dir).
- export.go: ReadScenarioFromRemovableMedia mounts each removable target
  looking for scenarios/<name>.json — an air-gapped engineer can author a
  scenario elsewhere, drop it under scenarios/ on the same USB stick
  already plugged in for blackbox, and run it with no network path onto
  the host.
- cmd/bee: new `bee run <file.json|name>` (bare name = looked up on
  removable media); `bee scenario run <arg>` kept as a longer alias.
- scenarios/nvbandwidth-all-gpu-power-watch.json: the scenario that
  reproduced the actual reboot (full nvbandwidth across all GPUs, which
  crashed, vs. clean per-socket passes), with IPMI sensor + GPU power/temp
  sampling for a power-delivery correlation check.

Also: webui/page_topo.go — the /topo page's component-status-detail modal
(GET /api/component-detail/{type}) showed "No status data recorded yet"
for any component type ComponentStatusDB has no history for yet (e.g. GPU
before a SAT run this boot), even though the topology card for the same
component already showed "N OK" from the audit inventory snapshot.
inventoryFallbackRecords now synthesizes records from that same inventory
snapshot when StatusDB is empty, using the same device classifiers
(isGPUDeviceClass etc.) and severity mapping (classifyTopoSeverity) the
topology card itself uses, so the two views never disagree.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>

audit/internal/platform/export.go

@@ -151,3 +151,55 @@ func (s *System) ExportFileToTarget(src string, target RemovableTarget) (dst str
 
 	return dst, nil
 }
+
+// mountRemovableTargetReadOnly mounts target for reading if it isn't
+// already mounted (reusing its existing mountpoint otherwise), returning
+// whether this call did the mounting so the caller knows whether to
+// unmount afterward.
+func mountRemovableTargetReadOnly(target RemovableTarget) (mountpoint string, mountedHere bool, err error) {
+	if mp := strings.TrimSpace(target.Mountpoint); mp != "" {
+		return mp, false, nil
+	}
+	mountpoint = filepath.Join("/tmp", "bee-scenario-"+filepath.Base(target.Device))
+	if err := os.MkdirAll(mountpoint, 0755); err != nil {
+		return "", false, err
+	}
+	if raw, err := exportExecCommand("mount", target.Device, mountpoint).CombinedOutput(); err != nil {
+		_ = os.Remove(mountpoint)
+		return "", false, formatMountTargetError(target, string(raw), err)
+	}
+	return mountpoint, true, nil
+}
+
+// ReadScenarioFromRemovableMedia mounts each removable target in turn
+// (unmounting again afterward if it mounted it itself), looking for
+// scenarios/<name>.json, and returns the contents of the first one found.
+// Lets an air-gapped engineer author a scenario JSON file on another
+// machine, drop it under scenarios/ on the same flash drive already
+// plugged in for blackbox, and run it on the host with no network path
+// required — see RunScenario/ParseScenarioJSON.
+func (s *System) ReadScenarioFromRemovableMedia(name string) ([]byte, error) {
+	name = strings.TrimSpace(name)
+	if name == "" {
+		return nil, fmt.Errorf("scenario name is required")
+	}
+	targets, err := s.ListRemovableTargets()
+	if err != nil {
+		return nil, err
+	}
+	for _, target := range targets {
+		mountpoint, mountedHere, mountErr := mountRemovableTargetReadOnly(target)
+		if mountErr != nil {
+			continue
+		}
+		data, readErr := os.ReadFile(filepath.Join(mountpoint, "scenarios", name+".json"))
+		if mountedHere {
+			_, _ = exportExecCommand("umount", mountpoint).CombinedOutput()
+			_ = os.Remove(mountpoint)
+		}
+		if readErr == nil {
+			return data, nil
+		}
+	}
+	return nil, fmt.Errorf("scenarios/%s.json not found on any removable media", name)
+}

audit/internal/platform/export_test.go

@@ -110,3 +110,50 @@ NAME="sdb1" TYPE="part" PKNAME="sdb" RM="1" RO="0" FSTYPE="vfat" MOUNTPOINT="/me
 		t.Fatalf("device=%q want /dev/sdb1", got)
 	}
 }
+
+func TestReadScenarioFromRemovableMediaFindsFileOnAlreadyMountedTarget(t *testing.T) {
+	mountpoint := t.TempDir()
+	if err := os.MkdirAll(filepath.Join(mountpoint, "scenarios"), 0755); err != nil {
+		t.Fatalf("mkdir scenarios: %v", err)
+	}
+	want := []byte(`{"name":"power-watch","jobs":[{"name":"a","type":"command","cmd":["true"]}]}`)
+	if err := os.WriteFile(filepath.Join(mountpoint, "scenarios", "power-watch.json"), want, 0644); err != nil {
+		t.Fatalf("write scenario file: %v", err)
+	}
+
+	oldExec := exportExecCommand
+	lsblkOut := `NAME="sdb1" TYPE="part" PKNAME="sdb" RM="1" RO="0" FSTYPE="vfat" MOUNTPOINT="` + mountpoint + `" SIZE="29.8G" LABEL="USB" MODEL=""`
+	exportExecCommand = func(name string, args ...string) *exec.Cmd {
+		cmd := exec.Command("sh", "-c", "printf '%s\n' \"$LSBLK_OUT\"")
+		cmd.Env = append(os.Environ(), "LSBLK_OUT="+lsblkOut)
+		return cmd
+	}
+	t.Cleanup(func() { exportExecCommand = oldExec })
+
+	s := &System{}
+	got, err := s.ReadScenarioFromRemovableMedia("power-watch")
+	if err != nil {
+		t.Fatalf("ReadScenarioFromRemovableMedia error: %v", err)
+	}
+	if string(got) != string(want) {
+		t.Fatalf("got=%q want=%q", got, want)
+	}
+}
+
+func TestReadScenarioFromRemovableMediaNotFound(t *testing.T) {
+	mountpoint := t.TempDir() // no scenarios/ dir at all
+
+	oldExec := exportExecCommand
+	lsblkOut := `NAME="sdb1" TYPE="part" PKNAME="sdb" RM="1" RO="0" FSTYPE="vfat" MOUNTPOINT="` + mountpoint + `" SIZE="29.8G" LABEL="USB" MODEL=""`
+	exportExecCommand = func(name string, args ...string) *exec.Cmd {
+		cmd := exec.Command("sh", "-c", "printf '%s\n' \"$LSBLK_OUT\"")
+		cmd.Env = append(os.Environ(), "LSBLK_OUT="+lsblkOut)
+		return cmd
+	}
+	t.Cleanup(func() { exportExecCommand = oldExec })
+
+	s := &System{}
+	if _, err := s.ReadScenarioFromRemovableMedia("nope"); err == nil {
+		t.Fatal("expected error for missing scenario file")
+	}
+}

audit/internal/platform/scenario.go

@@ -0,0 +1,293 @@
+package platform
+
+import (
+	"context"
+	"encoding/json"
+	"fmt"
+	"os"
+	"path/filepath"
+	"strings"
+	"sync"
+	"time"
+)
+
+// ScenarioJob is one step of a ScenarioSpec: either a one-shot/blocking
+// command ("command") or a periodic background sampler ("sampler") that
+// keeps running until every command job in the scenario has finished (or
+// the scenario's overall timeout elapses).
+type ScenarioJob struct {
+	Name string   `json:"name"`
+	Type string   `json:"type"` // "command" | "sampler"
+	Cmd  []string `json:"cmd"`
+	// GPUIndices, if non-empty, replaces the literal token "{{gpus}}" in Cmd
+	// with a comma-joined index list — lets a scenario file say
+	// ["dcgmi","diag","-r","nvbandwidth","-i","{{gpus}}"] once instead of
+	// hardcoding a specific host's GPU indices into the file.
+	GPUIndices []int `json:"gpu_indices,omitempty"`
+	// IntervalSec is the sampling period for a "sampler" job; ignored for
+	// "command" jobs.
+	IntervalSec int `json:"interval_sec,omitempty"`
+	// Parallel, for a "command" job, means it starts alongside whichever
+	// other parallel command jobs precede it instead of waiting for them to
+	// finish first. Sequential (default) command jobs run in file order,
+	// each waiting for any preceding parallel batch to finish first.
+	Parallel bool `json:"parallel,omitempty"`
+}
+
+// ScenarioSpec is a user-authored test scenario: what to run, on which
+// GPUs, and what to sample in the background while it runs. Parsed from
+// plain JSON (see ParseScenarioJSON) so authoring one needs no tooling
+// beyond a text editor and no new Go dependency for this codebase.
+//
+// Example — reproduce a hard reboot seen only when all GPUs run nvbandwidth
+// together (not on any single socket alone), while watching PSU/IPMI
+// sensors for a power-delivery correlation:
+//
+//	{
+//	  "name": "nvbandwidth-all-gpu-power-watch",
+//	  "timeout_sec": 1800,
+//	  "jobs": [
+//	    {"name": "ipmi-sensors", "type": "sampler", "interval_sec": 2,
+//	     "cmd": ["ipmitool", "sensor"]},
+//	    {"name": "gpu-power", "type": "sampler", "interval_sec": 2,
+//	     "cmd": ["nvidia-smi", "--query-gpu=index,power.draw,temperature.gpu,clocks.sm", "--format=csv"]},
+//	    {"name": "nvbandwidth-all", "type": "command",
+//	     "cmd": ["dcgmi", "diag", "-r", "nvbandwidth", "-i", "{{gpus}}"],
+//	     "gpu_indices": [0, 1, 2, 3, 4]}
+//	  ]
+//	}
+type ScenarioSpec struct {
+	Name       string        `json:"name"`
+	TimeoutSec int           `json:"timeout_sec,omitempty"`
+	Jobs       []ScenarioJob `json:"jobs"`
+}
+
+// ParseScenarioJSON parses and validates a scenario file's contents.
+func ParseScenarioJSON(data []byte) (ScenarioSpec, error) {
+	var spec ScenarioSpec
+	if err := json.Unmarshal(data, &spec); err != nil {
+		return ScenarioSpec{}, fmt.Errorf("parse scenario: %w", err)
+	}
+	if strings.TrimSpace(spec.Name) == "" {
+		return ScenarioSpec{}, fmt.Errorf(`scenario: "name" is required`)
+	}
+	if len(spec.Jobs) == 0 {
+		return ScenarioSpec{}, fmt.Errorf("scenario: at least one job is required")
+	}
+	seen := map[string]bool{}
+	for i, j := range spec.Jobs {
+		if strings.TrimSpace(j.Name) == "" {
+			return ScenarioSpec{}, fmt.Errorf("scenario: job %d: \"name\" is required", i)
+		}
+		if seen[j.Name] {
+			return ScenarioSpec{}, fmt.Errorf("scenario: duplicate job name %q", j.Name)
+		}
+		seen[j.Name] = true
+		if j.Type != "command" && j.Type != "sampler" {
+			return ScenarioSpec{}, fmt.Errorf("scenario: job %q: type must be \"command\" or \"sampler\", got %q", j.Name, j.Type)
+		}
+		if len(j.Cmd) == 0 {
+			return ScenarioSpec{}, fmt.Errorf("scenario: job %q: \"cmd\" is required", j.Name)
+		}
+		if j.Type == "sampler" && j.IntervalSec <= 0 {
+			return ScenarioSpec{}, fmt.Errorf("scenario: sampler job %q: \"interval_sec\" must be > 0", j.Name)
+		}
+	}
+	return spec, nil
+}
+
+// RunScenario executes a ScenarioSpec: runs its "command" jobs (sequential
+// unless marked Parallel, each blocking until it exits), while every
+// "sampler" job runs concurrently in the background on its own interval
+// until all command jobs have finished or TimeoutSec elapses. Every job's
+// own output streams live to its own file under the returned run
+// directory.
+//
+// Command jobs are wired through the same satJobBoundaryHook/
+// satSyncBracketHook seams the SAT job runner uses (see sat.go), so a
+// scenario run gets the same durability treatment: a crash mid-command
+// still leaves whatever printed up to that point, and — if blackbox is
+// running — that evidence is requested (and, for the sync-bracket hook,
+// waited on) to reach removable media before/after the command runs rather
+// than only at the end of an adaptive flush period.
+func (s *System) RunScenario(ctx context.Context, baseDir string, spec ScenarioSpec, logFunc func(string)) (string, error) {
+	if baseDir == "" {
+		baseDir = "/var/log/bee-sat"
+	}
+	ts := time.Now().UTC().Format("20060102-150405")
+	runDir := filepath.Join(baseDir, "scenario-"+sanitizeScenarioName(spec.Name)+"-"+ts)
+	if err := os.MkdirAll(runDir, 0755); err != nil {
+		return "", err
+	}
+	verboseLog := filepath.Join(runDir, "verbose.log")
+
+	if spec.TimeoutSec > 0 {
+		var cancel context.CancelFunc
+		ctx, cancel = context.WithTimeout(ctx, time.Duration(spec.TimeoutSec)*time.Second)
+		defer cancel()
+	}
+
+	sampCtx, stopSamplers := context.WithCancel(ctx)
+	defer stopSamplers()
+	var samplerWG sync.WaitGroup
+	for _, job := range spec.Jobs {
+		if job.Type != "sampler" {
+			continue
+		}
+		job := job
+		samplerWG.Add(1)
+		go func() {
+			defer samplerWG.Done()
+			runScenarioSampler(sampCtx, runDir, verboseLog, job, logFunc)
+		}()
+	}
+
+	var (
+		mu      sync.Mutex
+		summary strings.Builder
+		cmdWG   sync.WaitGroup
+	)
+	fmt.Fprintf(&summary, "scenario=%s\n", spec.Name)
+	fmt.Fprintf(&summary, "run_at_utc=%s\n", time.Now().UTC().Format(time.RFC3339))
+
+	runOne := func(job ScenarioJob) {
+		err := runScenarioCommand(ctx, runDir, verboseLog, job, logFunc)
+		mu.Lock()
+		if err != nil {
+			fmt.Fprintf(&summary, "%s_status=FAILED\n", job.Name)
+			fmt.Fprintf(&summary, "%s_error=%s\n", job.Name, err.Error())
+		} else {
+			fmt.Fprintf(&summary, "%s_status=OK\n", job.Name)
+		}
+		mu.Unlock()
+	}
+
+	for _, job := range spec.Jobs {
+		if job.Type != "command" {
+			continue
+		}
+		if job.Parallel {
+			cmdWG.Add(1)
+			go func(j ScenarioJob) {
+				defer cmdWG.Done()
+				runOne(j)
+			}(job)
+			continue
+		}
+		// A sequential job waits for any parallel batch launched ahead of it
+		// to finish first, so scenario-file order stays intuitive: parallel
+		// jobs run together, the next sequential job starts only once they
+		// (and any earlier sequential job) are done.
+		cmdWG.Wait()
+		runOne(job)
+	}
+	cmdWG.Wait()
+
+	stopSamplers()
+	samplerWG.Wait()
+
+	if err := os.WriteFile(filepath.Join(runDir, "summary.txt"), []byte(summary.String()), 0644); err != nil {
+		return "", err
+	}
+	return runDir, nil
+}
+
+func runScenarioCommand(ctx context.Context, runDir, verboseLog string, job ScenarioJob, logFunc func(string)) error {
+	cmd := substituteGPUIndices(job.Cmd, job.GPUIndices)
+	livePath := filepath.Join(runDir, sanitizeScenarioName(job.Name)+".log")
+
+	if satSyncBracketHook != nil {
+		if err := satSyncBracketHook(job.Name, "before"); err != nil && logFunc != nil {
+			logFunc(fmt.Sprintf("%s: blackbox sync wait (before) did not complete cleanly: %v", job.Name, err))
+		}
+	}
+	_, err := runSATCommandCtx(ctx, verboseLog, job.Name, cmd, nil, logFunc, livePath)
+	if satJobBoundaryHook != nil {
+		satJobBoundaryHook(job.Name)
+	}
+	if satSyncBracketHook != nil {
+		if syncErr := satSyncBracketHook(job.Name, "after"); syncErr != nil && logFunc != nil {
+			logFunc(fmt.Sprintf("%s: blackbox sync wait (after) did not complete cleanly: %v", job.Name, syncErr))
+		}
+	}
+	return err
+}
+
+// runScenarioSampler runs job.Cmd once immediately (so even a scenario that
+// finishes very quickly gets a baseline reading) and then every
+// job.IntervalSec until ctx is done, appending each timestamped sample to
+// its own file under runDir.
+func runScenarioSampler(ctx context.Context, runDir, verboseLog string, job ScenarioJob, logFunc func(string)) {
+	path := filepath.Join(runDir, sanitizeScenarioName(job.Name)+".log")
+	f, err := os.OpenFile(path, os.O_CREATE|os.O_APPEND|os.O_WRONLY, 0644)
+	if err != nil {
+		if logFunc != nil {
+			logFunc(fmt.Sprintf("%s: could not open sampler output %s: %v", job.Name, path, err))
+		}
+		return
+	}
+	defer f.Close()
+
+	cmd := substituteGPUIndices(job.Cmd, job.GPUIndices)
+	sample := func() {
+		out, cmdErr := satExecCommand(cmd[0], cmd[1:]...).CombinedOutput()
+		ts := time.Now().UTC().Format(time.RFC3339Nano)
+		fmt.Fprintf(f, "=== %s ===\n", ts)
+		if cmdErr != nil {
+			fmt.Fprintf(f, "error: %v\n", cmdErr)
+		}
+		_, _ = f.Write(out)
+		if len(out) == 0 || out[len(out)-1] != '\n' {
+			_, _ = f.Write([]byte("\n"))
+		}
+		_ = f.Sync()
+	}
+
+	sample()
+	ticker := time.NewTicker(time.Duration(job.IntervalSec) * time.Second)
+	defer ticker.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			appendSATVerboseLog(verboseLog, fmt.Sprintf("[%s] sampler %s stopped", time.Now().UTC().Format(time.RFC3339), job.Name))
+			return
+		case <-ticker.C:
+			sample()
+		}
+	}
+}
+
+// substituteGPUIndices replaces the literal token "{{gpus}}" in each cmd
+// argument with a comma-joined GPU index list. Returns cmd unchanged
+// (same backing values, no allocation) when gpuIndices is empty.
+func substituteGPUIndices(cmd []string, gpuIndices []int) []string {
+	if len(gpuIndices) == 0 {
+		return cmd
+	}
+	joined := joinIndexList(gpuIndices)
+	out := make([]string, len(cmd))
+	for i, arg := range cmd {
+		out[i] = strings.ReplaceAll(arg, "{{gpus}}", joined)
+	}
+	return out
+}
+
+// sanitizeScenarioName converts a scenario/job name into a safe filename
+// component: lowercase alphanumerics, '-', and '_' pass through; anything
+// else (spaces, punctuation) becomes '-'.
+func sanitizeScenarioName(name string) string {
+	var b strings.Builder
+	for _, r := range strings.ToLower(strings.TrimSpace(name)) {
+		switch {
+		case r >= 'a' && r <= 'z', r >= '0' && r <= '9', r == '-', r == '_':
+			b.WriteRune(r)
+		default:
+			b.WriteRune('-')
+		}
+	}
+	s := b.String()
+	if s == "" {
+		return "scenario"
+	}
+	return s
+}

audit/internal/platform/scenario_test.go

@@ -0,0 +1,206 @@
+package platform
+
+import (
+	"context"
+	"os"
+	"os/exec"
+	"path/filepath"
+	"strings"
+	"sync/atomic"
+	"testing"
+	"time"
+)
+
+func TestParseScenarioJSONValidatesRequiredFields(t *testing.T) {
+	cases := []struct {
+		name    string
+		json    string
+		wantErr string
+	}{
+		{"missing name", `{"jobs":[{"name":"a","type":"command","cmd":["true"]}]}`, `"name" is required`},
+		{"no jobs", `{"name":"x","jobs":[]}`, "at least one job is required"},
+		{"job missing name", `{"name":"x","jobs":[{"type":"command","cmd":["true"]}]}`, `"name" is required`},
+		{"duplicate job name", `{"name":"x","jobs":[{"name":"a","type":"command","cmd":["true"]},{"name":"a","type":"command","cmd":["true"]}]}`, "duplicate job name"},
+		{"bad type", `{"name":"x","jobs":[{"name":"a","type":"bogus","cmd":["true"]}]}`, `type must be "command" or "sampler"`},
+		{"missing cmd", `{"name":"x","jobs":[{"name":"a","type":"command"}]}`, `"cmd" is required`},
+		{"sampler no interval", `{"name":"x","jobs":[{"name":"a","type":"sampler","cmd":["true"]}]}`, `"interval_sec" must be > 0`},
+	}
+	for _, c := range cases {
+		t.Run(c.name, func(t *testing.T) {
+			_, err := ParseScenarioJSON([]byte(c.json))
+			if err == nil || !strings.Contains(err.Error(), c.wantErr) {
+				t.Fatalf("err=%v want containing %q", err, c.wantErr)
+			}
+		})
+	}
+}
+
+func TestParseScenarioJSONValid(t *testing.T) {
+	spec, err := ParseScenarioJSON([]byte(`{
+		"name": "power-watch",
+		"timeout_sec": 60,
+		"jobs": [
+			{"name": "ipmi", "type": "sampler", "interval_sec": 2, "cmd": ["ipmitool", "sensor"]},
+			{"name": "load", "type": "command", "cmd": ["dcgmi", "diag", "-r", "nvbandwidth", "-i", "{{gpus}}"], "gpu_indices": [0,1,2]}
+		]
+	}`))
+	if err != nil {
+		t.Fatalf("ParseScenarioJSON error: %v", err)
+	}
+	if spec.Name != "power-watch" || spec.TimeoutSec != 60 || len(spec.Jobs) != 2 {
+		t.Fatalf("spec=%+v", spec)
+	}
+}
+
+func TestSubstituteGPUIndices(t *testing.T) {
+	cmd := []string{"dcgmi", "diag", "-r", "nvbandwidth", "-i", "{{gpus}}"}
+	got := substituteGPUIndices(cmd, []int{0, 2, 4})
+	want := []string{"dcgmi", "diag", "-r", "nvbandwidth", "-i", "0,2,4"}
+	if len(got) != len(want) {
+		t.Fatalf("got=%v want=%v", got, want)
+	}
+	for i := range want {
+		if got[i] != want[i] {
+			t.Fatalf("got=%v want=%v", got, want)
+		}
+	}
+	// No GPUIndices: cmd passes through untouched (no {{gpus}} to replace).
+	same := substituteGPUIndices(cmd, nil)
+	if same[5] != "{{gpus}}" {
+		t.Fatalf("expected token left alone when no indices given, got %v", same)
+	}
+}
+
+func TestSanitizeScenarioName(t *testing.T) {
+	cases := map[string]string{
+		"power-watch":    "power-watch",
+		"Power Watch!!":  "power-watch--",
+		"":               "scenario",
+		"already_ok-123": "already_ok-123",
+	}
+	for in, want := range cases {
+		if got := sanitizeScenarioName(in); got != want {
+			t.Fatalf("sanitizeScenarioName(%q)=%q want %q", in, got, want)
+		}
+	}
+}
+
+func TestRunScenarioRunsCommandsAndSamplersConcurrently(t *testing.T) {
+	old := satExecCommand
+	var samplerCalls int32
+	satExecCommand = func(name string, args ...string) *exec.Cmd {
+		if name == "sample-cmd" {
+			atomic.AddInt32(&samplerCalls, 1)
+			return exec.Command("printf", "sampled\n")
+		}
+		return exec.Command(name, args...)
+	}
+	t.Cleanup(func() { satExecCommand = old })
+
+	dir := t.TempDir()
+	spec := ScenarioSpec{
+		Name:       "test-scenario",
+		TimeoutSec: 10,
+		Jobs: []ScenarioJob{
+			{Name: "watch", Type: "sampler", Cmd: []string{"sample-cmd"}, IntervalSec: 1},
+			{Name: "load", Type: "command", Cmd: []string{"printf", "loaded\n"}},
+		},
+	}
+
+	s := &System{}
+	runDir, err := s.RunScenario(context.Background(), dir, spec, nil)
+	if err != nil {
+		t.Fatalf("RunScenario error: %v", err)
+	}
+
+	if _, err := os.Stat(filepath.Join(runDir, "load.log")); err != nil {
+		t.Fatalf("load.log missing: %v", err)
+	}
+	loadOut, err := os.ReadFile(filepath.Join(runDir, "load.log"))
+	if err != nil || strings.TrimSpace(string(loadOut)) != "loaded" {
+		t.Fatalf("load.log=%q err=%v", loadOut, err)
+	}
+
+	watchOut, err := os.ReadFile(filepath.Join(runDir, "watch.log"))
+	if err != nil {
+		t.Fatalf("watch.log missing: %v", err)
+	}
+	if !strings.Contains(string(watchOut), "sampled") {
+		t.Fatalf("watch.log=%q want it to contain a sample", watchOut)
+	}
+	if atomic.LoadInt32(&samplerCalls) < 1 {
+		t.Fatalf("sampler command was never invoked")
+	}
+
+	summary, err := os.ReadFile(filepath.Join(runDir, "summary.txt"))
+	if err != nil {
+		t.Fatalf("summary.txt missing: %v", err)
+	}
+	if !strings.Contains(string(summary), "load_status=OK") {
+		t.Fatalf("summary=%q want load_status=OK", summary)
+	}
+}
+
+func TestRunScenarioStopsSamplersWhenCommandsFinish(t *testing.T) {
+	old := satExecCommand
+	var samplerCalls int32
+	satExecCommand = func(name string, args ...string) *exec.Cmd {
+		atomic.AddInt32(&samplerCalls, 1)
+		return exec.Command("true")
+	}
+	t.Cleanup(func() { satExecCommand = old })
+
+	dir := t.TempDir()
+	spec := ScenarioSpec{
+		Name: "fast-scenario",
+		Jobs: []ScenarioJob{
+			// Interval much longer than the scenario itself takes to run —
+			// if the sampler weren't stopped promptly when the command
+			// finishes, this test would need to wait out the interval.
+			{Name: "watch", Type: "sampler", Cmd: []string{"sample-cmd"}, IntervalSec: 3600},
+			{Name: "load", Type: "command", Cmd: []string{"true"}},
+		},
+	}
+
+	s := &System{}
+	start := time.Now()
+	if _, err := s.RunScenario(context.Background(), dir, spec, nil); err != nil {
+		t.Fatalf("RunScenario error: %v", err)
+	}
+	if elapsed := time.Since(start); elapsed > 5*time.Second {
+		t.Fatalf("RunScenario took %s — sampler wasn't stopped promptly", elapsed)
+	}
+	// Exactly one sample: the immediate baseline one, no second tick.
+	if got := atomic.LoadInt32(&samplerCalls); got != 1 {
+		t.Fatalf("samplerCalls=%d want 1 (baseline only)", got)
+	}
+}
+
+func TestRunScenarioRecordsFailedCommandInSummary(t *testing.T) {
+	old := satExecCommand
+	satExecCommand = func(name string, args ...string) *exec.Cmd {
+		return exec.Command(name, args...)
+	}
+	t.Cleanup(func() { satExecCommand = old })
+
+	dir := t.TempDir()
+	spec := ScenarioSpec{
+		Name: "failing-scenario",
+		Jobs: []ScenarioJob{
+			{Name: "will-fail", Type: "command", Cmd: []string{"false"}},
+		},
+	}
+
+	s := &System{}
+	runDir, err := s.RunScenario(context.Background(), dir, spec, nil)
+	if err != nil {
+		t.Fatalf("RunScenario error: %v", err)
+	}
+	summary, err := os.ReadFile(filepath.Join(runDir, "summary.txt"))
+	if err != nil {
+		t.Fatalf("summary.txt missing: %v", err)
+	}
+	if !strings.Contains(string(summary), "will-fail_status=FAILED") {
+		t.Fatalf("summary=%q want will-fail_status=FAILED", summary)
+	}
+}