diff --git a/docs/cluster.md b/docs/cluster.md
index db0a56a..0e9926e 100644
--- a/docs/cluster.md
+++ b/docs/cluster.md
@@ -123,10 +123,65 @@ handle `Sandbox.Promote` returns is still owner-bound — `template.New` and
 `template.Delete` dial the owner directly, no gossip involved — and is the
 race-free choice immediately after a promote.
 
+## State ownership
+
+Each kind of state has exactly one source of truth, and a restart rebuilds
+fully from it:
+
+| state | source of truth | survives restart |
+|---|---|---|
+| operator config (`tenants`, `secrets`, egress policies, `bridge`/`network`, `mesh`, `preview_secret`, `egress_ca`) | `config.json` (human/deploy-tool owned) | re-read at boot |
+| API-applied pool targets (`PUT /v1/pools`) | `<data_dir>/pools.json` (machine owned) | yes |
+| claims | the claims journal + `Reconcile` | yes |
+| placement hints (warm counts, template sets) | gossip | rebuilt |
+
+Pools are managed API-first. The first time a node takes `PUT /v1/pools`, it
+writes the applied set to `<data_dir>/pools.json` and from then on **that file
+seeds the pools at boot**, overriding the `pools` section of `config.json` (a
+loud log line notes this; if the two disagree because someone edited
+`config.json` afterward, a second warning names it). The file's presence is the
+ownership marker — delete `pools.json` to return the node to config-owned pools.
+Egress stays config-owned regardless: the API rejects egress specs, so
+`pools.json` never carries them, and egress policies re-merge from `config.json`
+by pool key at boot.
+
+Cluster-wide pool changes are a client-side fan-out, not a gossiped desired
+state (pools are legitimately heterogeneous per node): `Client.SetPoolsCluster`
+PUTs the set to the entry node and every peer and returns a per-node result;
+retrying the failed nodes is the whole protocol, because the apply is an
+idempotent declarative replace. A non-nil error means peer discovery itself
+failed and the fan-out reached only the entry node — an incomplete apply to
+retry, kept distinct from a genuine single-node cluster (nil error). It fits a
+homogeneous cluster — a spec set that names an egress-lane pool is refused on a
+node without an attachment, so nodes that differ in capacity or attachment take
+a per-node `Client.SetPools`.
+
+### Cluster-invariant config
+
+Some config must match on every node or the cluster fails in confusing ways:
+
+| config | what breaks on mismatch |
+|---|---|
+| `api_token`, `tenants` | the SDK replays the authorizing token across a redirect; a peer missing that tenant/token answers 401 |
+| `preview_secret` | a preview URL signed on one node fails verification on another |
+| `mesh.cluster_key` | nodes cannot join / decrypt gossip at all |
+| `egress_ca` cluster root | a guest checkpointed/redirected across nodes trusts the root; a divergent root fails interception |
+
+Each node gossips a digest of these (HMAC-keyed by `cluster_key` when set;
+otherwise a token-free digest of tenant names + the CA root, so nothing
+brute-forceable rides cleartext gossip). A mismatch logs a warning at the moment
+the divergent node appears — not at the first unlucky redirect — and raises the
+`sandboxd_config_digest_mismatch` gauge. It is warn-only: a rolling credential
+rotation is a legitimate transient mismatch, so a divergence never partitions
+the mesh.
+
 ## Cluster checklist
 
 - memberlist port (e.g. 7946) open node-to-node, TCP **and** UDP
 - `advertise_addr` set to a routable address on every node (never loopback,
   never a wildcard)
-- same `api_token` and `tenants` set everywhere
+- same `api_token`, `tenants`, `preview_secret`, and `egress_ca` root everywhere
+  (a mismatch warns and shows in `sandboxd_config_digest_mismatch`)
 - `cluster_key` set if the gossip network is not otherwise trusted
+- pool changes via `Client.SetPoolsCluster` (or per-node `SetPools`); the applied
+  set persists to `pools.json` and survives restart
diff --git a/docs/deploy.md b/docs/deploy.md
index 54a6b93..05ac088 100644
--- a/docs/deploy.md
+++ b/docs/deploy.md
@@ -66,7 +66,7 @@ sandboxd reads one JSON file (`-config`, default
 | `archive_after_seconds` | 0 (off) | tier below hibernation: a hibernated claim idle this long is checkpointed to the store and its local VM dropped, freeing the node entirely; the next call restores it transparently (a checkpoint restore's latency). Requires `idle_hibernate_seconds > 0` and must exceed it. Node-wide for unpooled keys; per-pool overrides for that pool |
 | `archive_delete_after_seconds` | 0 (keep) | purge an archived claim's store checkpoint this long after it was archived, reclaiming storage; the claim is then gone for good. Same node-wide/per-pool split |
 | `mesh` | unset | join a cluster ([Clusters](cluster.md)); unset = single node |
-| `pools[]` | — | warm pools. `warm` defaults to 4; `net` is `none` or `egress`; `size` is a tier, below. Retune online without a restart via [`PUT /v1/pools`](sandboxd-api.md#put-v1pools) — omitted pools drain |
+| `pools[]` | — | warm pools. `warm` defaults to 4; `net` is `none` or `egress`; `size` is a tier, below. Retune online without a restart via [`PUT /v1/pools`](sandboxd-api.md#put-v1pools) — omitted pools drain. This is the **first-boot seed**: once a node takes a `PUT /v1/pools`, the applied set persists to `<data_dir>/pools.json` and overrides this section on every later boot (a startup log notes it); delete `pools.json` to return to config-owned pools. Egress stays config-owned either way. See [state ownership](cluster.md#state-ownership) |
 
 Size tiers (free-form CPU/memory is deliberately not accepted — it would
 fragment the warm pools):
diff --git a/sandboxd/config/config.go b/sandboxd/config/config.go
index 507765f..911791c 100644
--- a/sandboxd/config/config.go
+++ b/sandboxd/config/config.go
@@ -2,9 +2,15 @@
 package config
 
 import (
+	"crypto/hmac"
+	"crypto/sha256"
+	"encoding/base64"
+	"encoding/hex"
+	"encoding/json"
 	"fmt"
 	"os"
 	"slices"
+	"strings"
 
 	"github.com/cocoonstack/sandbox/sandboxd/egress"
 	"github.com/cocoonstack/sandbox/sandboxd/store/s3"
@@ -204,6 +210,34 @@ func (c *Config) HasEgress() bool {
 	return c.Bridge != "" || c.Network != ""
 }
 
+// ClusterDigest fingerprints the must-match config so a divergent node is
+// caught early, not at a later 401. With cluster_key it HMACs token material
+// too; without one only tenant names + CA root ride the (maybe cleartext) wire.
+func (c *Config) ClusterDigest(caFingerprint string) string {
+	names := make([]string, len(c.Tenants))
+	for i, t := range c.Tenants {
+		names[i] = t.Name
+	}
+	slices.Sort(names)
+	if c.Mesh != nil && c.Mesh.ClusterKey != "" {
+		if key, err := base64.StdEncoding.DecodeString(c.Mesh.ClusterKey); err == nil {
+			type auth struct{ Name, Token string }
+			tenants := make([]auth, len(c.Tenants))
+			for i, t := range c.Tenants {
+				tenants[i] = auth{t.Name, t.Token}
+			}
+			slices.SortFunc(tenants, func(a, b auth) int { return strings.Compare(a.Name, b.Name) })
+			raw, _ := json.Marshal([]any{c.APIToken, c.PreviewSecret, caFingerprint, tenants})
+			mac := hmac.New(sha256.New, key)
+			mac.Write(raw)
+			return hex.EncodeToString(mac.Sum(nil))
+		}
+	}
+	raw, _ := json.Marshal([]any{caFingerprint, names})
+	sum := sha256.Sum256(raw)
+	return hex.EncodeToString(sum[:])
+}
+
 // guardsEgressLane: any tenant policy counts (claims mint egress keys without a
 // pool), a none-lane pool policy does not (it locks no tap).
 func (c *Config) guardsEgressLane() bool {
diff --git a/sandboxd/config/config_test.go b/sandboxd/config/config_test.go
index c6833b0..613ff11 100644
--- a/sandboxd/config/config_test.go
+++ b/sandboxd/config/config_test.go
@@ -1,6 +1,7 @@
 package config
 
 import (
+	"encoding/base64"
 	"os"
 	"path/filepath"
 	"strings"
@@ -9,6 +10,30 @@ import (
 	"github.com/cocoonstack/sandbox/sandboxd/types"
 )
 
+func TestClusterDigest(t *testing.T) {
+	base := &Config{APIToken: "tok", PreviewSecret: "ps", Tenants: []TenantSpec{{Name: "acme", Token: "t1"}}}
+	d := base.ClusterDigest("ca-fp")
+	if base.ClusterDigest("ca-fp") != d {
+		t.Fatal("digest is not stable for identical config")
+	}
+	if (&Config{APIToken: "tok", PreviewSecret: "ps", Tenants: []TenantSpec{{Name: "beta", Token: "t1"}}}).ClusterDigest("ca-fp") == d {
+		t.Error("a tenant-name change is not reflected")
+	}
+	if base.ClusterDigest("other-fp") == d {
+		t.Error("an egress CA root change is not reflected")
+	}
+	// Keyless: token material must stay off the (maybe cleartext) wire.
+	if (&Config{APIToken: "other", PreviewSecret: "ps", Tenants: base.Tenants}).ClusterDigest("ca-fp") != d {
+		t.Error("api_token leaked into the keyless digest")
+	}
+	key := base64.StdEncoding.EncodeToString(make([]byte, 32))
+	keyed := &Config{APIToken: "tok", PreviewSecret: "ps", Tenants: base.Tenants, Mesh: &MeshConfig{ClusterKey: key}}
+	keyedDiff := &Config{APIToken: "other", PreviewSecret: "ps", Tenants: base.Tenants, Mesh: &MeshConfig{ClusterKey: key}}
+	if keyed.ClusterDigest("ca-fp") == keyedDiff.ClusterDigest("ca-fp") {
+		t.Error("with a cluster_key the api_token must be covered by the digest")
+	}
+}
+
 func TestLoadAppliesDefaults(t *testing.T) {
 	path := writeConfig(t, `{"pools":[{"template":"rt:24.04","net":"none","size":"small"}]}`)
 	cfg, err := Load(path)
diff --git a/sandboxd/main.go b/sandboxd/main.go
index 91614e3..3904b36 100644
--- a/sandboxd/main.go
+++ b/sandboxd/main.go
@@ -82,7 +82,7 @@ func main() {
 
 	var placer server.Placer
 	if cfg.Mesh != nil {
-		msh, err := startMesh(cfg)
+		msh, err := startMesh(cfg, mgr)
 		if err != nil {
 			logger.Fatalf(ctx, err, "start mesh")
 		}
@@ -141,7 +141,7 @@ func main() {
 	logger.Info(ctx, "sandboxd stopped; VMs stay alive for the next reconcile")
 }
 
-func startMesh(cfg *config.Config) (*mesh.Mesh, error) {
+func startMesh(cfg *config.Config, mgr *pool.Manager) (*mesh.Mesh, error) {
 	mc := cfg.Mesh
 	mlCfg := memberlist.DefaultLANConfig()
 	host, portStr, err := net.SplitHostPort(mc.Bind)
@@ -167,10 +167,12 @@ func startMesh(cfg *config.Config) (*mesh.Mesh, error) {
 			return nil, fmt.Errorf("mesh cluster key: %w", err)
 		}
 	}
-	msh, err := mesh.New(mlCfg, nodeID, cfg.AdvertiseAddr, key)
+	msh, err := mesh.New(mlCfg, nodeID, cfg.AdvertiseAddr, key, cfg.DataDir)
 	if err != nil {
 		return nil, err
 	}
+	// Publish the config digest before Join, so the first gossip carries it.
+	msh.SetSelfDigest(cfg.ClusterDigest(mgr.EgressCAFingerprint()))
 	if err := msh.Join(mc.Join); err != nil {
 		_ = msh.Shutdown()
 		return nil, err
diff --git a/sandboxd/mesh/epoch.go b/sandboxd/mesh/epoch.go
new file mode 100644
index 0000000..f943950
--- /dev/null
+++ b/sandboxd/mesh/epoch.go
@@ -0,0 +1,29 @@
+package mesh
+
+import (
+	"math"
+	"os"
+	"strconv"
+	"strings"
+
+	"github.com/cocoonstack/sandbox/sandboxd/utils"
+)
+
+// loadEpoch reads the persisted gossip epoch, or 0 when the file is absent,
+// unreadable, or above MaxInt64 — a UnixNano-derived counter cannot legitimately
+// exceed it, so a corrupt value falls back to the wall-clock seed.
+func loadEpoch(path string) uint64 {
+	raw, err := os.ReadFile(path) //nolint:gosec // node-local data-dir path
+	if err != nil {
+		return 0
+	}
+	n, err := strconv.ParseUint(strings.TrimSpace(string(raw)), 10, 64)
+	if err != nil || n > math.MaxInt64 {
+		return 0
+	}
+	return n
+}
+
+func storeEpoch(path string, epoch uint64) error {
+	return utils.WriteFileSync(path, []byte(strconv.FormatUint(epoch, 10)), 0o600)
+}
diff --git a/sandboxd/mesh/mesh.go b/sandboxd/mesh/mesh.go
index 473cacd..d2e6aea 100644
--- a/sandboxd/mesh/mesh.go
+++ b/sandboxd/mesh/mesh.go
@@ -7,15 +7,18 @@
 package mesh
 
 import (
+	"context"
 	"encoding/json"
 	"fmt"
 	"maps"
 	"math/rand/v2"
+	"path/filepath"
 	"slices"
 	"sync"
 	"time"
 
 	"github.com/hashicorp/memberlist"
+	"github.com/projecteru2/core/log"
 )
 
 // leaveTimeout bounds the graceful-leave broadcast on shutdown so a wedged
@@ -30,33 +33,44 @@ type NodeState struct {
 	Epoch     uint64         `json:"epoch"`
 	Pools     map[string]int `json:"pools"`               // PoolKey hash → warm count
 	Templates []string       `json:"templates,omitempty"` // promoted-template key hashes on disk
+	Digest    string         `json:"digest,omitempty"`    // cluster-invariant config digest
 }
 
 // Mesh is the node's view of the cluster and its own gossiped state.
 type Mesh struct {
-	ml *memberlist.Memberlist
+	ml        *memberlist.Memberlist
+	epochPath string
 
 	mu   sync.Mutex
 	self NodeState
 	view map[string]NodeState // node_id → latest known state (includes self)
+
+	epochMu      sync.Mutex
+	epochWritten uint64 // highest epoch on disk; guarded by epochMu
 }
 
 // New starts a mesh member listening per cfg. selfAddr is the data-plane
 // address peers should dial for a redirect; secretKey (16/24/32 bytes) enables
-// gossip encryption when non-empty.
-func New(cfg *memberlist.Config, nodeID, selfAddr string, secretKey []byte) (*Mesh, error) {
+// gossip encryption when non-empty; dataDir holds the persisted epoch.
+func New(cfg *memberlist.Config, nodeID, selfAddr string, secretKey []byte, dataDir string) (*Mesh, error) {
+	epochPath := filepath.Join(dataDir, "mesh-epoch")
+	// Seed strictly above the persisted floor (the last epoch peers saw):
+	// seeding at it ties their stale copy and merge's `>` rejects the restart's
+	// fresh state. loadEpoch caps the floor so the +1 cannot wrap.
+	epoch := max(uint64(time.Now().UnixNano()), loadEpoch(epochPath)+1) //nolint:gosec // UnixNano is positive for current times
 	m := &Mesh{
+		epochPath: epochPath,
 		self: NodeState{
 			NodeID: nodeID,
 			Addr:   selfAddr,
-			// Seed the epoch from wall-clock so a restarted node's fresh counts
-			// aren't rejected by peers still holding its pre-restart (higher)
-			// epoch; Epoch++ keeps intra-process monotonicity above that base.
-			Epoch: uint64(time.Now().UnixNano()), //nolint:gosec // UnixNano is positive for current times
-			Pools: map[string]int{},
+			Epoch:  epoch,
+			Pools:  map[string]int{},
 		},
 		view: map[string]NodeState{},
 	}
+	if err := m.persistEpoch(epoch); err != nil {
+		return nil, fmt.Errorf("persist mesh epoch: %w", err)
+	}
 	m.view[nodeID] = m.self
 
 	cfg.Name = nodeID
@@ -90,16 +104,55 @@ func (m *Mesh) Join(seeds []string) error {
 // every second for nothing.
 func (m *Mesh) UpdateSelf(pools map[string]int, templates []string) {
 	m.mu.Lock()
-	defer m.mu.Unlock()
 	if maps.Equal(m.self.Pools, pools) && slices.Equal(m.self.Templates, templates) {
+		m.mu.Unlock()
+		return
+	}
+	epoch := m.self.Epoch + 1
+	m.mu.Unlock()
+	// Persist the candidate before publishing it: memberlist gossips self the
+	// instant it enters the view, so a crash before the write would strand peers
+	// on an epoch a backwards-clock restart can't beat. Hold old state on failure.
+	if err := m.persistEpoch(epoch); err != nil {
+		log.WithFunc("mesh.UpdateSelf").Warnf(context.Background(), "persist epoch: %v", err)
 		return
 	}
-	m.self.Epoch++
-	m.self.Pools = pools
-	m.self.Templates = templates
+	m.mu.Lock()
+	if epoch > m.self.Epoch {
+		m.self.Epoch = epoch
+		m.self.Pools = pools
+		m.self.Templates = templates
+		m.view[m.self.NodeID] = m.self
+	}
+	m.mu.Unlock()
+}
+
+// SetSelfDigest records this node's cluster-invariant config digest so peers can
+// detect divergence; call it once before Join, before any gossip ships.
+func (m *Mesh) SetSelfDigest(digest string) {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	m.self.Digest = digest
 	m.view[m.self.NodeID] = m.self
 }
 
+// ConfigMismatches counts peers whose config digest differs from this node's —
+// the gauge for alerting on a divergent cluster, recomputed per read.
+func (m *Mesh) ConfigMismatches() int {
+	m.mu.Lock()
+	defer m.mu.Unlock()
+	if m.self.Digest == "" {
+		return 0
+	}
+	n := 0
+	for id, st := range m.view {
+		if id != m.self.NodeID && st.Digest != "" && st.Digest != m.self.Digest {
+			n++
+		}
+	}
+	return n
+}
+
 // Candidates returns up to two peer addresses that report warm(keyHash) > 0,
 // chosen power-of-two-choices to avoid herding every waiter onto one node.
 // Self is never a candidate — the caller has already missed locally.
@@ -188,6 +241,21 @@ func (m *Mesh) Shutdown() error {
 	return m.ml.Shutdown()
 }
 
+// persistEpoch durably records the epoch, serialized so a slower concurrent
+// UpdateSelf's lower value cannot overwrite a higher one already on disk.
+func (m *Mesh) persistEpoch(epoch uint64) error {
+	m.epochMu.Lock()
+	defer m.epochMu.Unlock()
+	if epoch <= m.epochWritten {
+		return nil
+	}
+	if err := storeEpoch(m.epochPath, epoch); err != nil {
+		return err
+	}
+	m.epochWritten = epoch
+	return nil
+}
+
 // forget drops a departed node from the placement view so redirects stop
 // targeting a dead peer; SWIM detected the death, the view must follow.
 func (m *Mesh) forget(nodeID string) {
@@ -207,10 +275,28 @@ func (m *Mesh) merge(states []NodeState) {
 		if st.NodeID == m.self.NodeID {
 			continue
 		}
-		if cur, ok := m.view[st.NodeID]; !ok || st.Epoch > cur.Epoch {
-			m.view[st.NodeID] = st
+		cur, ok := m.view[st.NodeID]
+		if ok && st.Epoch <= cur.Epoch {
+			continue
+		}
+		// Warn on each distinct divergent digest (not once per lifetime): a
+		// mismatched api_token/tenants/preview_secret/CA root 401s cross-node
+		// redirects and fails interception. Warn-only — refusing would partition
+		// a rolling credential rotation.
+		if m.self.Digest != "" && st.Digest != "" && st.Digest != m.self.Digest && (!ok || cur.Digest != st.Digest) {
+			log.WithFunc("mesh.merge").Warnf(context.Background(),
+				"peer %s config digest %s differs from this node's %s: cluster-invariant config diverges (redirects may 401, interception may fail)",
+				st.NodeID, short(st.Digest), short(m.self.Digest))
 		}
+		m.view[st.NodeID] = st
+	}
+}
+
+func short(digest string) string {
+	if len(digest) > 12 {
+		return digest[:12]
 	}
+	return digest
 }
 
 var _ memberlist.Delegate = (*delegate)(nil)
diff --git a/sandboxd/mesh/mesh_test.go b/sandboxd/mesh/mesh_test.go
index 3f18de6..60e0b7a 100644
--- a/sandboxd/mesh/mesh_test.go
+++ b/sandboxd/mesh/mesh_test.go
@@ -4,6 +4,7 @@ import (
 	"fmt"
 	"io"
 	"log"
+	"path/filepath"
 	"slices"
 	"testing"
 	"time"
@@ -140,8 +141,9 @@ func TestTwoNodeClusterGossipsPools(t *testing.T) {
 func newTestMesh(t *testing.T, id string) *Mesh {
 	t.Helper()
 	return &Mesh{
-		self: NodeState{NodeID: id, Addr: id + ":7777", Pools: map[string]int{}},
-		view: map[string]NodeState{id: {NodeID: id, Addr: id + ":7777"}},
+		epochPath: filepath.Join(t.TempDir(), "mesh-epoch"),
+		self:      NodeState{NodeID: id, Addr: id + ":7777", Pools: map[string]int{}},
+		view:      map[string]NodeState{id: {NodeID: id, Addr: id + ":7777"}},
 	}
 }
 
@@ -160,7 +162,7 @@ func startNode(t *testing.T, host string, port int, id string) *node {
 	cfg.AdvertiseAddr = host
 	cfg.PushPullInterval = 200 * time.Millisecond
 	cfg.Logger = discardLogger()
-	m, err := New(cfg, id, id+":7777", nil)
+	m, err := New(cfg, id, id+":7777", nil, t.TempDir())
 	if err != nil {
 		t.Fatalf("new mesh %s: %v", id, err)
 	}
diff --git a/sandboxd/mesh/state_test.go b/sandboxd/mesh/state_test.go
new file mode 100644
index 0000000..309694e
--- /dev/null
+++ b/sandboxd/mesh/state_test.go
@@ -0,0 +1,118 @@
+package mesh
+
+import (
+	"os"
+	"path/filepath"
+	"sync"
+	"testing"
+
+	"github.com/hashicorp/memberlist"
+)
+
+func newBoundMesh(t *testing.T, dataDir string) *Mesh {
+	t.Helper()
+	cfg := memberlist.DefaultLocalConfig()
+	cfg.BindPort = 0
+	cfg.Logger = discardLogger()
+	m, err := New(cfg, "self", "self:7777", nil, dataDir)
+	if err != nil {
+		t.Fatalf("new mesh: %v", err)
+	}
+	t.Cleanup(func() { _ = m.Shutdown() })
+	return m
+}
+
+func TestEpochRoundTrip(t *testing.T) {
+	p := filepath.Join(t.TempDir(), "e")
+	if loadEpoch(p) != 0 {
+		t.Error("missing epoch file must read 0")
+	}
+	if err := storeEpoch(p, 42); err != nil {
+		t.Fatalf("store: %v", err)
+	}
+	if got := loadEpoch(p); got != 42 {
+		t.Errorf("loadEpoch = %d, want 42", got)
+	}
+}
+
+func TestLoadEpochRejectsImplausibleValue(t *testing.T) {
+	p := filepath.Join(t.TempDir(), "e")
+	// A saturated value must not seed a counter that +1 can't climb past.
+	if err := os.WriteFile(p, []byte("18446744073709551615"), 0o600); err != nil {
+		t.Fatalf("write: %v", err)
+	}
+	if got := loadEpoch(p); got != 0 {
+		t.Errorf("loadEpoch of a MaxUint64 file = %d, want 0 (treated as corrupt)", got)
+	}
+}
+
+func TestEpochSeededAbovePersistedFloor(t *testing.T) {
+	dir := t.TempDir()
+	huge := uint64(1) << 62 // above any plausible wall-clock nanos
+	if err := storeEpoch(filepath.Join(dir, "mesh-epoch"), huge); err != nil {
+		t.Fatalf("store: %v", err)
+	}
+	m := newBoundMesh(t, dir)
+	// Strictly above: seeding at the persisted value would tie peers' stale copy.
+	if m.self.Epoch != huge+1 {
+		t.Errorf("epoch seeded %d, want persisted floor + 1 = %d (a backwards clock must land above peers' last-seen epoch)", m.self.Epoch, huge+1)
+	}
+}
+
+func TestUpdateSelfPersistFailKeepsOldState(t *testing.T) {
+	dir := t.TempDir()
+	m := newBoundMesh(t, dir)
+	before := m.self.Epoch
+	// A non-existent dir makes the durable write fail: the epoch must not publish.
+	m.epochPath = filepath.Join(dir, "gone", "mesh-epoch")
+	m.UpdateSelf(map[string]int{"k": 1}, nil)
+	if m.self.Epoch != before {
+		t.Errorf("self epoch advanced to %d on a failed persist, want %d held", m.self.Epoch, before)
+	}
+	if len(m.self.Pools) != 0 {
+		t.Errorf("self pools published %v despite a failed persist", m.self.Pools)
+	}
+}
+
+func TestUpdateSelfPersistsEpoch(t *testing.T) {
+	dir := t.TempDir()
+	m := newBoundMesh(t, dir)
+	before := m.self.Epoch
+	m.UpdateSelf(map[string]int{"k": 1}, nil)
+	if got := loadEpoch(filepath.Join(dir, "mesh-epoch")); got <= before {
+		t.Errorf("persisted epoch %d did not advance past the seed %d", got, before)
+	}
+}
+
+func TestPersistEpochMonotonic(t *testing.T) {
+	dir := t.TempDir()
+	m := newBoundMesh(t, dir)
+	base := m.self.Epoch // New already seeded a wall-clock floor
+	high := base + 1000
+	if err := m.persistEpoch(high); err != nil {
+		t.Fatalf("persist high: %v", err)
+	}
+	// Concurrent lower values must never regress the on-disk floor below high.
+	var wg sync.WaitGroup
+	for e := base + 1; e < high; e++ {
+		wg.Go(func() {
+			if err := m.persistEpoch(e); err != nil {
+				t.Errorf("persist %d: %v", e, err)
+			}
+		})
+	}
+	wg.Wait()
+	if got := loadEpoch(filepath.Join(dir, "mesh-epoch")); got != high {
+		t.Errorf("persisted epoch %d, want the %d floor held against lower concurrent writes", got, high)
+	}
+}
+
+func TestConfigDigestMismatch(t *testing.T) {
+	m := newTestMesh(t, "self")
+	m.SetSelfDigest("self-digest")
+	m.merge([]NodeState{{NodeID: "peerA", Addr: "a:1", Epoch: 1, Digest: "self-digest"}})
+	m.merge([]NodeState{{NodeID: "peerB", Addr: "b:1", Epoch: 1, Digest: "other-digest"}})
+	if n := m.ConfigMismatches(); n != 1 {
+		t.Errorf("ConfigMismatches = %d, want 1 (only peerB diverges)", n)
+	}
+}
diff --git a/sandboxd/pool/pool.go b/sandboxd/pool/pool.go
index 0596fb7..aa8e67f 100644
--- a/sandboxd/pool/pool.go
+++ b/sandboxd/pool/pool.go
@@ -155,6 +155,9 @@ type Manager struct {
 	maxFork int
 	store   *claimStore
 
+	poolStore      *poolStore
+	configSeedHash string // config pools' hash, to warn when a file edit is overridden
+
 	// idleDefault is the idle-hibernate threshold for unpooled keys; pooled
 	// keys carry theirs on the pool struct. Zero means disabled.
 	idleDefault time.Duration
@@ -251,6 +254,7 @@ func NewManager(ctx context.Context, cfg *config.Config, eng Engine, secrets *eg
 		lockEgress:      cfg.Bridge != "",
 		maxFork:         maxFork,
 		store:           newClaimStore(cfg.DataDir),
+		poolStore:       newPoolStore(cfg.DataDir),
 		pools:           make(map[types.PoolKey]*pool, len(cfg.Pools)),
 		claimed:         map[string]*types.Sandbox{},
 		tenantLive:      map[string]int{},
@@ -336,12 +340,25 @@ func NewManager(ctx context.Context, cfg *config.Config, eng Engine, secrets *eg
 		}
 		m.egressCA = ca
 	}
+	m.configSeedHash = poolSeedHash(cfg.Pools)
+	if err := m.adoptPersistedPools(ctx); err != nil {
+		return nil, err
+	}
 	if m.archiveEnabled && m.ckptTTL == 0 {
 		log.WithFunc("pool.NewManager").Warn(ctx, "archive enabled with checkpoint_ttl_hours=0: a checkpoint whose delete fails is not reclaimed")
 	}
 	return m, nil
 }
 
+// EgressCAFingerprint is the egress cluster root's fingerprint, or "" when the
+// node does no interception.
+func (m *Manager) EgressCAFingerprint() string {
+	if m.egressCA == nil {
+		return ""
+	}
+	return m.egressCA.Fingerprint()
+}
+
 func loadEgressCA(cfg *config.EgressCAConfig) (*egress.CA, error) {
 	root, err := os.ReadFile(cfg.RootCert) //nolint:gosec // operator-configured ca path
 	if err != nil {
diff --git a/sandboxd/pool/poolstore.go b/sandboxd/pool/poolstore.go
new file mode 100644
index 0000000..490af3f
--- /dev/null
+++ b/sandboxd/pool/poolstore.go
@@ -0,0 +1,135 @@
+package pool
+
+import (
+	"context"
+	"crypto/sha256"
+	"encoding/hex"
+	"encoding/json"
+	"errors"
+	"fmt"
+	"io/fs"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+	"sync"
+	"sync/atomic"
+
+	"github.com/projecteru2/core/log"
+
+	"github.com/cocoonstack/sandbox/sandboxd/config"
+	"github.com/cocoonstack/sandbox/sandboxd/utils"
+)
+
+// poolsFile records the last API-applied pool set so a restart rebuilds from
+// operator intent, not the config.json seed. ConfigSeed is the config pools'
+// hash at write time; a boot finding it changed warns the edit is overridden.
+type poolsFile struct {
+	ConfigSeed string            `json:"config_seed"`
+	Pools      []config.PoolSpec `json:"pools"`
+}
+
+type poolStore struct {
+	path string
+
+	seq     atomic.Uint64 // sequence source; bumped under the manager mutex (apply order)
+	mu      sync.Mutex
+	written uint64 // highest sequence on disk; guarded by mu
+}
+
+func newPoolStore(dataDir string) *poolStore {
+	return &poolStore{path: filepath.Join(dataDir, "pools.json")}
+}
+
+// load returns the persisted pool file, or nil when the node has never taken a
+// PUT /v1/pools (config.json seeds the pools, unchanged behavior).
+func (s *poolStore) load() (*poolsFile, error) {
+	raw, err := os.ReadFile(s.path)
+	if errors.Is(err, fs.ErrNotExist) {
+		return nil, nil
+	}
+	if err != nil {
+		return nil, fmt.Errorf("read pools file: %w", err)
+	}
+	var pf poolsFile
+	if err := json.Unmarshal(raw, &pf); err != nil {
+		return nil, fmt.Errorf("parse pools file: %w", err)
+	}
+	return &pf, nil
+}
+
+// commit durably writes the applied set, serialized by s.mu; a seq no newer
+// than the last written is a no-op, so a superseded concurrent write can't win.
+func (s *poolStore) commit(seq uint64, pf poolsFile) error {
+	s.mu.Lock()
+	defer s.mu.Unlock()
+	if seq <= s.written {
+		return nil
+	}
+	raw, err := json.Marshal(pf)
+	if err != nil {
+		return fmt.Errorf("encode pools file: %w", err)
+	}
+	if err := utils.WriteFileSync(s.path, raw, 0o600); err != nil {
+		return err
+	}
+	s.written = seq
+	return nil
+}
+
+// adoptPersistedPools rebuilds the pools from pools.json (the last API-applied
+// set) over the config seed when present. Egress stays config-owned; a restored
+// pool that no longer validates fails the boot, matching config-load discipline.
+func (m *Manager) adoptPersistedPools(ctx context.Context) error {
+	pf, err := m.poolStore.load()
+	if err != nil {
+		return err
+	}
+	if pf == nil {
+		return nil
+	}
+	logger := log.WithFunc("pool.adoptPersistedPools")
+	if pf.ConfigSeed != m.configSeedHash {
+		logger.Warnf(ctx, "config.json pools differ from the API-applied set and are overridden; delete %s to return to config-owned pools", m.poolStore.path)
+	}
+	clear(m.pools)
+	m.idleEnabled = m.idleDefault > 0
+	m.archiveEnabled = m.archiveAfterDefault > 0
+	for _, spec := range pf.Pools {
+		spec = normalizePoolSpec(spec)
+		if err := m.validate(spec.PoolKey); err != nil {
+			return fmt.Errorf("restore pool %q: %w", spec.Template, err)
+		}
+		if err := spec.ValidateLimits(); err != nil {
+			return fmt.Errorf("restore pool %q: %w", spec.Template, err)
+		}
+		p := &pool{key: spec.PoolKey}
+		p.applySpec(spec)
+		if g := filepath.Join(m.goldensDir(), spec.Hash()); dirExists(g) && m.goldenCAMatches(g, m.poolEgress[spec.PoolKey].Intercepts()) {
+			p.goldenDir = g
+		}
+		m.pools[spec.PoolKey] = p
+		if spec.IdleHibernateSeconds > 0 {
+			m.idleEnabled = true
+		}
+		if spec.ArchiveAfterSeconds > 0 {
+			m.archiveEnabled = true
+		}
+	}
+	logger.Infof(ctx, "restored %d API-applied pools from pools.json", len(pf.Pools))
+	return nil
+}
+
+// poolSeedHash digests a pool set's warm-target shape, order-independent and
+// egress-excluded, so it tracks exactly the targets pools.json would override.
+func poolSeedHash(specs []config.PoolSpec) string {
+	shaped := make([]config.PoolSpec, len(specs))
+	copy(shaped, specs)
+	for i := range shaped {
+		shaped[i].Egress = nil
+	}
+	slices.SortFunc(shaped, func(a, b config.PoolSpec) int { return strings.Compare(a.Hash(), b.Hash()) })
+	raw, _ := json.Marshal(shaped)
+	sum := sha256.Sum256(raw)
+	return hex.EncodeToString(sum[:])
+}
diff --git a/sandboxd/pool/poolstore_test.go b/sandboxd/pool/poolstore_test.go
new file mode 100644
index 0000000..c0e6a10
--- /dev/null
+++ b/sandboxd/pool/poolstore_test.go
@@ -0,0 +1,99 @@
+package pool
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"sync"
+	"testing"
+
+	"github.com/cocoonstack/sandbox/sandboxd/config"
+	"github.com/cocoonstack/sandbox/sandboxd/types"
+)
+
+var (
+	seedKey = types.PoolKey{Template: "seed:24.04", Net: types.NetNone, Size: types.SizeSmall}
+	apiKey  = types.PoolKey{Template: "api:24.04", Net: types.NetNone, Size: types.SizeSmall}
+)
+
+func TestPersistedPoolsSurviveRestart(t *testing.T) {
+	dir := t.TempDir()
+	m := newTestManagerAt(t, newFakeEngine(), dir, config.PoolSpec{PoolKey: seedKey, Warm: 1})
+	if err := m.SetPools(t.Context(), []config.PoolSpec{{PoolKey: apiKey, Warm: 3}}); err != nil {
+		t.Fatalf("SetPools: %v", err)
+	}
+	// Restart on the same data dir, still config-seeded: must rebuild from pools.json.
+	m2 := newTestManagerAt(t, newFakeEngine(), dir, config.PoolSpec{PoolKey: seedKey, Warm: 1})
+	m2.mu.Lock()
+	_, hasSeed := m2.pools[seedKey]
+	restored, hasAPI := m2.pools[apiKey]
+	m2.mu.Unlock()
+	if hasSeed || !hasAPI {
+		t.Fatalf("restart did not restore API pools: seed=%v api=%v", hasSeed, hasAPI)
+	}
+	if restored.floor != 3 {
+		t.Errorf("restored warm floor = %d, want 3", restored.floor)
+	}
+}
+
+func TestDeletingPoolsFileRestoresConfigSeed(t *testing.T) {
+	dir := t.TempDir()
+	m := newTestManagerAt(t, newFakeEngine(), dir, config.PoolSpec{PoolKey: seedKey, Warm: 1})
+	if err := m.SetPools(t.Context(), []config.PoolSpec{{PoolKey: apiKey, Warm: 3}}); err != nil {
+		t.Fatalf("SetPools: %v", err)
+	}
+	if err := os.Remove(filepath.Join(dir, "pools.json")); err != nil {
+		t.Fatalf("remove pools.json: %v", err)
+	}
+	m2 := newTestManagerAt(t, newFakeEngine(), dir, config.PoolSpec{PoolKey: seedKey, Warm: 1})
+	m2.mu.Lock()
+	_, hasSeed := m2.pools[seedKey]
+	_, hasAPI := m2.pools[apiKey]
+	m2.mu.Unlock()
+	if !hasSeed || hasAPI {
+		t.Fatalf("deleting pools.json did not restore the config seed: seed=%v api=%v", hasSeed, hasAPI)
+	}
+}
+
+func TestConcurrentSetPoolsPersistLatest(t *testing.T) {
+	dir := t.TempDir()
+	m := newTestManagerAt(t, newFakeEngine(), dir, config.PoolSpec{PoolKey: seedKey, Warm: 1})
+	var wg sync.WaitGroup
+	for warm := 1; warm <= 8; warm++ {
+		wg.Go(func() {
+			if err := m.SetPools(t.Context(), []config.PoolSpec{{PoolKey: apiKey, Warm: warm}}); err != nil {
+				t.Errorf("SetPools(warm=%d): %v", warm, err)
+			}
+		})
+	}
+	wg.Wait()
+	// The last-applied set owns memory; the persisted file must match it.
+	m.mu.Lock()
+	want := m.pools[apiKey].floor
+	m.mu.Unlock()
+	pf, err := m.poolStore.load()
+	if err != nil || pf == nil {
+		t.Fatalf("load pools.json: pf=%v err=%v", pf, err)
+	}
+	if len(pf.Pools) != 1 || pf.Pools[0].Warm != want {
+		t.Fatalf("persisted %+v, want the last-applied warm %d", pf.Pools, want)
+	}
+}
+
+func TestRestoredEgressPoolNeedsAttachment(t *testing.T) {
+	dir := t.TempDir()
+	pf := poolsFile{Pools: []config.PoolSpec{
+		{PoolKey: types.PoolKey{Template: "eg:24.04", Net: types.NetEgress, Size: types.SizeSmall}, Warm: 1},
+	}}
+	raw, err := json.Marshal(pf)
+	if err != nil {
+		t.Fatalf("marshal: %v", err)
+	}
+	if err := os.WriteFile(filepath.Join(dir, "pools.json"), raw, 0o600); err != nil {
+		t.Fatalf("write pools.json: %v", err)
+	}
+	// A node with no bridge/network must refuse to start a restored egress pool.
+	if _, err := NewManager(t.Context(), &config.Config{DataDir: dir}, newFakeEngine(), testSecrets(t)); err == nil {
+		t.Error("NewManager accepted a restored egress-lane pool without an attachment")
+	}
+}
diff --git a/sandboxd/pool/setpools.go b/sandboxd/pool/setpools.go
index f1236e2..5822c34 100644
--- a/sandboxd/pool/setpools.go
+++ b/sandboxd/pool/setpools.go
@@ -40,6 +40,8 @@ func (m *Manager) SetPools(ctx context.Context, specs []config.PoolSpec) error {
 
 	var trim []string
 	m.mu.Lock()
+	// Sequenced under the mutex (apply order) so commit drops a write a later apply superseded.
+	seq := m.poolStore.seq.Add(1)
 	now := time.Now()
 	for key, p := range m.pools {
 		spec, ok := desired[key]
@@ -92,7 +94,12 @@ func (m *Manager) SetPools(ctx context.Context, specs []config.PoolSpec) error {
 		m.destroy(ctx, trim[i])
 	}).Wait()
 	m.refillOnce(runCtx)
-	return nil
+	// Persist the applied set so a restart rebuilds from it, not the config seed.
+	persisted := make([]config.PoolSpec, 0, len(desired))
+	for _, spec := range desired {
+		persisted = append(persisted, spec)
+	}
+	return m.poolStore.commit(seq, poolsFile{ConfigSeed: m.configSeedHash, Pools: persisted})
 }
 
 // normalizePoolSpec fills the wire defaults, mirroring ClaimRequest.Key():
diff --git a/sandboxd/server/metrics.go b/sandboxd/server/metrics.go
index 5ff8cfa..148331d 100644
--- a/sandboxd/server/metrics.go
+++ b/sandboxd/server/metrics.go
@@ -48,6 +48,11 @@ func (s *Server) handleMetrics(w http.ResponseWriter, _ *http.Request) {
 		_, _ = fmt.Fprintf(w, "sandboxd_pool_target{template=%q,net=%q,size=%q} %d\n", p.Key.Template, p.Key.Net, p.Key.Size, p.Target)
 	}
 
+	if s.placer != nil {
+		metric("config_digest_mismatch", "gauge", "peers whose cluster-invariant config digest differs from this node")
+		_, _ = fmt.Fprintf(w, "sandboxd_config_digest_mismatch %d\n", s.placer.ConfigMismatches())
+	}
+
 	metric("claims_total", "counter", "claims served, by provisioning tier")
 	_, _ = fmt.Fprintf(w, "sandboxd_claims_total{tier=\"warm\"} %d\n", c.ClaimsWarm)
 	_, _ = fmt.Fprintf(w, "sandboxd_claims_total{tier=\"clone\"} %d\n", c.ClaimsClone)
diff --git a/sandboxd/server/server.go b/sandboxd/server/server.go
index 9079b01..8aa48bf 100644
--- a/sandboxd/server/server.go
+++ b/sandboxd/server/server.go
@@ -90,6 +90,7 @@ type Placer interface {
 	Candidates(keyHash string) []string
 	TemplateOwners(keyHash string) []string
 	PeerAddrs() []string
+	ConfigMismatches() int
 }
 
 // InfoResponse is the wire reply of GET /v1/info. Peers lists the other nodes'
diff --git a/sandboxd/server/server_test.go b/sandboxd/server/server_test.go
index 9006b71..328fab5 100644
--- a/sandboxd/server/server_test.go
+++ b/sandboxd/server/server_test.go
@@ -990,3 +990,4 @@ type fakePlacer struct {
 func (f *fakePlacer) Candidates(string) []string     { return f.addrs }
 func (f *fakePlacer) TemplateOwners(string) []string { return f.owners }
 func (f *fakePlacer) PeerAddrs() []string            { return f.addrs }
+func (f *fakePlacer) ConfigMismatches() int          { return 0 }
diff --git a/sandboxd/utils/utils.go b/sandboxd/utils/utils.go
index 04cebf4..df1a0a6 100644
--- a/sandboxd/utils/utils.go
+++ b/sandboxd/utils/utils.go
@@ -7,9 +7,57 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"os"
+	"path/filepath"
 	"strings"
 )
 
+// WriteFileSync durably replaces path with data (temp + fsync + rename + dir
+// fsync) so the rename survives a crash. The temp name is per-call unique, so
+// concurrent writers to the same path can't race a shared temp into a lost rename.
+func WriteFileSync(path string, data []byte, perm os.FileMode) error {
+	dir := filepath.Dir(path)
+	f, err := os.CreateTemp(dir, filepath.Base(path)+".*.tmp")
+	if err != nil {
+		return fmt.Errorf("create temp: %w", err)
+	}
+	tmp := f.Name()
+	renamed := false
+	defer func() {
+		if !renamed {
+			_ = os.Remove(tmp) //nolint:gosec // our own just-created temp under a caller-owned data-dir path
+		}
+	}()
+	if err = f.Chmod(perm); err != nil {
+		_ = f.Close()
+		return fmt.Errorf("chmod %s: %w", tmp, err)
+	}
+	if _, err = f.Write(data); err != nil {
+		_ = f.Close()
+		return fmt.Errorf("write %s: %w", tmp, err)
+	}
+	if err = f.Sync(); err != nil {
+		_ = f.Close()
+		return fmt.Errorf("sync %s: %w", tmp, err)
+	}
+	if err = f.Close(); err != nil {
+		return fmt.Errorf("close %s: %w", tmp, err)
+	}
+	if err = os.Rename(tmp, path); err != nil { //nolint:gosec // our own temp, caller-owned data-dir path
+		return fmt.Errorf("rename %s: %w", path, err)
+	}
+	renamed = true
+	d, err := os.Open(dir) //nolint:gosec // caller-owned data-dir path
+	if err != nil {
+		return fmt.Errorf("open dir: %w", err)
+	}
+	defer func() { _ = d.Close() }()
+	if err = d.Sync(); err != nil {
+		return fmt.Errorf("sync dir: %w", err)
+	}
+	return nil
+}
+
 // DecodeStrictJSON decodes one JSON value into v, refusing unknown fields,
 // duplicated keys, and trailing data — for hand-edited operator input, where
 // a typo must fail instead of silently changing what was configured (an
diff --git a/sdk/go/info.go b/sdk/go/info.go
index 4d2b6c7..149a16e 100644
--- a/sdk/go/info.go
+++ b/sdk/go/info.go
@@ -44,17 +44,22 @@ func (c *Client) Info(ctx context.Context) (*NodeInfo, error) {
 	return &info, nil
 }
 
-// peers fetches the cluster's node addresses, best-effort. It reads the
-// tenant-accessible /v1/peers, not /v1/info (operator-only), so Lookup
-// works under a tenant token.
+// peers fetches the cluster's node addresses, best-effort (nil on failure).
 func (c *Client) peers(ctx context.Context) []string {
+	addrs, _ := c.peersOrErr(ctx)
+	return addrs
+}
+
+// peersOrErr fetches the node addresses, surfacing a discovery failure. It reads
+// /v1/peers (tenant-accessible), so it works under a tenant token.
+func (c *Client) peersOrErr(ctx context.Context) ([]string, error) {
 	ctx, cancel := context.WithTimeout(ctx, peersTimeout)
 	defer cancel()
 	reply, err := doJSON[struct {
 		Peers []string `json:"peers"`
 	}](ctx, c, http.MethodGet, c.addr, "/v1/peers", nil, c.apiToken, "peers")
 	if err != nil {
-		return nil
+		return nil, err
 	}
-	return reply.Peers
+	return reply.Peers, nil
 }
diff --git a/sdk/go/pools.go b/sdk/go/pools.go
new file mode 100644
index 0000000..9395ca5
--- /dev/null
+++ b/sdk/go/pools.go
@@ -0,0 +1,81 @@
+package sandbox
+
+import (
+	"bytes"
+	"context"
+	"fmt"
+	"net/http"
+	"sync"
+)
+
+// PoolSpec is a desired warm pool for SetPools. Egress is config-owned on the
+// node and rejected by the API, so it has no field here.
+type PoolSpec struct {
+	Template                  string   `json:"template"`
+	Net                       NetShape `json:"net,omitempty"`
+	Size                      Size     `json:"size,omitempty"`
+	Warm                      int      `json:"warm"`
+	WarmMax                   int      `json:"warm_max,omitempty"`
+	IdleHibernateSeconds      int      `json:"idle_hibernate_seconds,omitempty"`
+	ArchiveAfterSeconds       int      `json:"archive_after_seconds,omitempty"`
+	ArchiveDeleteAfterSeconds int      `json:"archive_delete_after_seconds,omitempty"`
+}
+
+// PoolResult is one node's outcome from SetPoolsCluster.
+type PoolResult struct {
+	Addr string
+	Info *NodeInfo
+	Err  error
+}
+
+type poolUpdate struct {
+	Pools []PoolSpec `json:"pools"`
+}
+
+// SetPools replaces the entry node's desired warm pools (PUT /v1/pools): a
+// declarative full replace, so omitted pools drain. Requires the operator token.
+func (c *Client) SetPools(ctx context.Context, pools []PoolSpec) (*NodeInfo, error) {
+	return c.setPoolsAt(ctx, c.addr, pools)
+}
+
+// SetPoolsCluster applies pools to the entry node and every peer, returning a
+// per-node result; retrying failed nodes is the whole protocol (idempotent
+// replace). A non-nil error means peer discovery failed and only the entry node
+// was reached — an incomplete apply to retry, not a single-node cluster (nil).
+func (c *Client) SetPoolsCluster(ctx context.Context, pools []PoolSpec) ([]PoolResult, error) {
+	peers, peersErr := c.peersOrErr(ctx)
+	seen := map[string]struct{}{}
+	var addrs []string
+	for _, a := range append([]string{c.addr}, peers...) {
+		if _, dup := seen[a]; dup {
+			continue
+		}
+		seen[a] = struct{}{}
+		addrs = append(addrs, a)
+	}
+	results := make([]PoolResult, len(addrs))
+	var wg sync.WaitGroup
+	for i, addr := range addrs {
+		wg.Go(func() {
+			info, err := c.setPoolsAt(ctx, addr, pools)
+			results[i] = PoolResult{Addr: addr, Info: info, Err: err}
+		})
+	}
+	wg.Wait()
+	if peersErr != nil {
+		return results, fmt.Errorf("discover peers: %w", peersErr)
+	}
+	return results, nil
+}
+
+func (c *Client) setPoolsAt(ctx context.Context, addr string, pools []PoolSpec) (*NodeInfo, error) {
+	body, err := encodeBody("pools", poolUpdate{Pools: pools})
+	if err != nil {
+		return nil, err
+	}
+	info, err := doJSON[NodeInfo](ctx, c, http.MethodPut, addr, "/v1/pools", bytes.NewReader(body), c.apiToken, "pools")
+	if err != nil {
+		return nil, err
+	}
+	return &info, nil
+}
diff --git a/sdk/go/pools_test.go b/sdk/go/pools_test.go
new file mode 100644
index 0000000..b76e0c6
--- /dev/null
+++ b/sdk/go/pools_test.go
@@ -0,0 +1,111 @@
+package sandbox
+
+import (
+	"encoding/json"
+	"net/http"
+	"net/http/httptest"
+	"strings"
+	"sync"
+	"testing"
+)
+
+func TestSetPools(t *testing.T) {
+	var got poolUpdate
+	ts := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.Method != http.MethodPut || r.URL.Path != "/v1/pools" {
+			t.Errorf("got %s %s, want PUT /v1/pools", r.Method, r.URL.Path)
+		}
+		if auth := r.Header.Get("Authorization"); auth != "Bearer root" {
+			t.Errorf("auth = %q, want Bearer root", auth)
+		}
+		_ = json.NewDecoder(r.Body).Decode(&got)
+		_ = json.NewEncoder(w).Encode(NodeInfo{Claimed: 2})
+	}))
+	defer ts.Close()
+
+	c := testClient(t, ts, WithAPIToken("root"))
+	info, err := c.SetPools(t.Context(), []PoolSpec{{Template: "rt:24.04", Net: NetNone, Size: Small, Warm: 3}})
+	if err != nil {
+		t.Fatalf("SetPools: %v", err)
+	}
+	if info.Claimed != 2 {
+		t.Errorf("info = %+v, want Claimed 2", info)
+	}
+	if len(got.Pools) != 1 || got.Pools[0].Warm != 3 || got.Pools[0].Template != "rt:24.04" {
+		t.Errorf("server received %+v", got.Pools)
+	}
+}
+
+func TestSetPoolsClusterFansOut(t *testing.T) {
+	var mu sync.Mutex
+	applied := map[string]int{}
+	pools := func(w http.ResponseWriter, r *http.Request) {
+		var body poolUpdate
+		_ = json.NewDecoder(r.Body).Decode(&body)
+		mu.Lock()
+		applied[r.Host] = body.Pools[0].Warm
+		mu.Unlock()
+		_ = json.NewEncoder(w).Encode(NodeInfo{})
+	}
+	peer := httptest.NewServer(http.HandlerFunc(pools))
+	defer peer.Close()
+	peerAddr := strings.TrimPrefix(peer.URL, "http://")
+
+	entry := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/v1/peers" {
+			_ = json.NewEncoder(w).Encode(map[string][]string{"peers": {peerAddr}})
+			return
+		}
+		pools(w, r)
+	}))
+	defer entry.Close()
+
+	c := testClient(t, entry, WithAPIToken("root"))
+	results, err := c.SetPoolsCluster(t.Context(), []PoolSpec{{Template: "rt:24.04", Net: NetNone, Size: Small, Warm: 5}})
+	if err != nil {
+		t.Fatalf("SetPoolsCluster: %v", err)
+	}
+	if len(results) != 2 {
+		t.Fatalf("results = %d, want 2 (entry + one peer)", len(results))
+	}
+	for _, res := range results {
+		if res.Err != nil {
+			t.Errorf("node %s: %v", res.Addr, res.Err)
+		}
+	}
+	mu.Lock()
+	defer mu.Unlock()
+	if len(applied) != 2 {
+		t.Fatalf("applied to %d nodes, want 2: %v", len(applied), applied)
+	}
+	for addr, warm := range applied {
+		if warm != 5 {
+			t.Errorf("node %s saw warm %d, want 5", addr, warm)
+		}
+	}
+}
+
+func TestSetPoolsClusterPeerDiscoveryFails(t *testing.T) {
+	var applied int
+	entry := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		if r.URL.Path == "/v1/peers" {
+			w.WriteHeader(http.StatusInternalServerError)
+			return
+		}
+		applied++
+		_ = json.NewEncoder(w).Encode(NodeInfo{})
+	}))
+	defer entry.Close()
+
+	c := testClient(t, entry, WithAPIToken("root"))
+	results, err := c.SetPoolsCluster(t.Context(), []PoolSpec{{Template: "rt:24.04", Net: NetNone, Size: Small, Warm: 5}})
+	if err == nil {
+		t.Fatal("SetPoolsCluster returned nil error despite peer-discovery failure")
+	}
+	if len(results) != 1 || results[0].Err != nil {
+		t.Fatalf("results = %+v, want the entry node applied cleanly", results)
+	}
+	if applied != 1 {
+		t.Errorf("applied to %d nodes, want 1 (entry only)", applied)
+	}
+}