diff --git a/README.md b/README.md
index 336f221..37d988d 100644
--- a/README.md
+++ b/README.md
@@ -55,7 +55,8 @@ Proxy Pool 用 Controller 协调这些变化，并让 Gateway 数据面只消费
   View 匹配请求，并在内存中按 Sequential、Random、Round Robin、Weighted 或 Least
   Connections 选择上游。无候选时支持 reject、受 `waitTimeout` 限制的本地容量等待，
   以及仍经过目标地址策略的 direct；`proxy-gateway` 通过独立控制面拨号地址维护
-  Session，并仅在持有未过期 Snapshot 时 Ready；凭据材料只保留在当前节点内存 View。
+  Session，并在每份 Snapshot 有效期的一半前接收版本递增的完整刷新；仅在持有未过期
+  Snapshot 时 Ready，凭据材料只保留在当前节点内存 View。
 - **安全边界**：Gateway、Distribution 与 Admin 使用各自的认证语义，并支持
   CIDR、可信代理、严格请求解析和敏感信息最小化。
 
diff --git a/docs/api/control-plane.md b/docs/api/control-plane.md
index b06d2f4..0fb1b51 100644
--- a/docs/api/control-plane.md
+++ b/docs/api/control-plane.md
@@ -25,7 +25,9 @@ Controller 已实现并验证 `RegisterWorker`、`AcknowledgeSnapshot` 和
 监听允许明文 fixture 模式。单消息大小、并发流数和 gRPC keepalive 策略由
 `controlPlane` 配置限定。
 
-`WatchSnapshots` 已在 Register 后发送与当前 ownership epoch 对应的基础完整快照，
+`WatchSnapshots` 已在 Register 后发送与当前 ownership epoch 对应的基础完整快照，并在
+每份快照有效期的一半前重新构建、下发版本递增的完整快照；这样 ownership、Routing 和
+凭据变化会在同一长连接内收敛，而无需等待有效期到达后重新注册。
 Gateway 校验后 ACK 并开始 Runtime 心跳。Controller 会从 Redis 的有界 Worker ownership
 索引构建已归属 Proxy 内容，并将租约到期收紧到 Proxy 的 `usable_until`。Proxy 引用的
 凭据材料按 `secret_ref + credential_version` 去重，随完整 Snapshot 经 mTLS 下发，仅保留在
@@ -108,11 +110,10 @@ Gateway 接收完整快照时必须拒绝缺失、格式错误或已到期的 `v
 保存在本地不可变视图。该整体期限到达后，调度直接按无候选处理，不再使用旧视图
 发起新的上游连接，也不查询 Redis 或 PostgreSQL 补偿。
 
-Controller 只会下发尚未到期的完整快照，并在最近一次成功下发快照的
-`valid_until` 到达时结束 `WatchSnapshots` 流。Gateway 的 `SessionSupervisor` 会在流
-结束或可恢复控制面错误后按带 jitter 的有界退避重建 Register/Watch 会话；参数、认证
-和协议不兼容错误直接返回。新快照通过校验并原子替换前，旧视图仍按其整体有效期
-fail-closed。
+Controller 只会下发尚未到期的完整快照，并在每份快照有效期的一半前发送下一版完整
+快照。刷新构建失败、流结束或可恢复控制面错误时，Gateway 的 `SessionSupervisor` 会按
+带 jitter 的有界退避重建 Register/Watch 会话；参数、认证和协议不兼容错误直接返回。
+新快照通过校验并原子替换前，旧视图仍按其整体有效期 fail-closed。
 
 Delta 声明 `base_version`。Worker 只有在本地版本恰好等于 base 且 checksum
 验证成功时才能应用；否则丢弃 Delta 并请求完整 Snapshot。构建在后台完成，
diff --git a/docs/configuration/reference.md b/docs/configuration/reference.md
index 168cca5..a2629c3 100644
--- a/docs/configuration/reference.md
+++ b/docs/configuration/reference.md
@@ -22,7 +22,7 @@ go run ./cmd/proxy-controller -config CONFIG_FILE
 PostgreSQL 管理面迁移、Redis 活动池、Distribution/Admin 独立监听与优雅停机；
 Controller Metrics 独立监听、`/livez`、`/readyz` 和基础 Prometheus 运行时指标；
 Provider 自动补池、分布式配额、动态重载和 Admin 低基数统计已装配。Gateway 进程、
-Worker 控制面会话、Snapshot 凭据分发和 Snapshot 就绪探针已装配；Checker 的任务流与
+Worker 控制面会话、持续完整 Snapshot 刷新、凭据分发和 Snapshot 就绪探针已装配；Checker 的任务流与
 租约契约已接入控制面，但 Redis 共享调度、独立执行器与代表性负载验证仍在后续实施范围。
 
 所有时间值使用 Go duration，例如 `500ms`、`30s`、`5m`。示例中的
diff --git a/docs/design/architecture.md b/docs/design/architecture.md
index 62f4e3d..0e6e377 100644
--- a/docs/design/architecture.md
+++ b/docs/design/architecture.md
@@ -57,7 +57,8 @@ Provider、Pool、Routing、Distribution 在首版需要共享事务和一致性
 - 首条命中 Routing、Upstream 选择和 Proxy least-connections 选择。
 - 本地原子容量预留、建连、Active 计数和结果上报。
 - 本地不可变 Snapshot；热路径无数据库/Redis/Provider API。
-- 独立的控制面 SessionSupervisor；快照流到期或中断后以有界退避重建会话。
+- 独立的控制面 SessionSupervisor；Controller 在有效期一半前连续下发版本递增的完整
+  快照，流中断后才以有界退避重建会话。
 - SSRF 与 DNS Rebinding 防护。
 
 ### 3.2 proxy-controller
diff --git a/findings.md b/findings.md
index d7f5865..f5c45b3 100644
--- a/findings.md
+++ b/findings.md
@@ -114,6 +114,16 @@ Routing 自上而下匹配，首条命中停止；支持 Gateway 与 Extract 两
   维护该索引。Memory 单测、Redis 集成契约和配置/Reaper 单测覆盖恢复、阈值、延后与
   Drain/ACK 后清理。
 
+## Worker Snapshot 刷新（2026-08-02）
+
+- `OwnedSnapshotSource` 每次调用只构建一份权威完整快照；新增
+  `RefreshingSnapshotSource` 以有效期一半为上限持续调用该窄接口，并传递最新版本和
+  checksum。短于全局 `maxStaleAge` 的 Proxy/ownership 租约会进一步缩短下一次刷新间隔。
+- gRPC Handler 每次成功发送完整 Snapshot 都重置服务端到期计时器；流中刷新失败或中断时
+  Gateway 使用既有 `SessionSupervisor` 退避重连。Gateway 本地 Store 对从完整 Snapshot
+  消失且仍有 Active/Reserved 的 Proxy 已按 draining 继续上报，但 Controller 尚未具备
+  自动 `BeginDrain`/`AcknowledgeDrain` 编排或 Worker 分配器，不能视为自动 Drain 闭环。
+
 ## Git 同步事实（2026-07-29）
 
 - PostgreSQL 管理面基础文档已提交为 `7951c29`。
diff --git a/internal/controller/worker/server.go b/internal/controller/worker/server.go
index 19b7793..0eb6375 100644
--- a/internal/controller/worker/server.go
+++ b/internal/controller/worker/server.go
@@ -60,9 +60,13 @@ func NewServer(controlPlane config.ControlPlane, service Service, options Server
 		}
 	}
 	if snapshots == nil {
-		snapshots, err = NewInitialSnapshotSource(service, controlPlane.MaxStaleAge.Value(), time.Now)
+		initial, initialErr := NewInitialSnapshotSource(service, controlPlane.MaxStaleAge.Value(), time.Now)
+		if initialErr != nil {
+			return nil, fmt.Errorf("%w: build initial snapshot source: %v", ErrInvalidServer, initialErr)
+		}
+		snapshots, err = NewRefreshingSnapshotSource(initial, snapshotRefreshEvery(controlPlane.MaxStaleAge.Value()))
 		if err != nil {
-			return nil, fmt.Errorf("%w: build initial snapshot source: %v", ErrInvalidServer, err)
+			return nil, fmt.Errorf("%w: build refreshing snapshot source: %v", ErrInvalidServer, err)
 		}
 	}
 	serverOptions = append(serverOptions,
diff --git a/internal/controller/worker/service.go b/internal/controller/worker/service.go
index 586ada6..aed3818 100644
--- a/internal/controller/worker/service.go
+++ b/internal/controller/worker/service.go
@@ -156,7 +156,11 @@ func NewService(store workerruntime.ControlStore, options Options) (Service, err
 		if err != nil {
 			return nil, errors.Join(ErrInvalidCommand, err)
 		}
-		result.snapshots = source
+		refreshed, refreshErr := NewRefreshingSnapshotSource(source, snapshotRefreshEvery(options.MaxStaleAge))
+		if refreshErr != nil {
+			return nil, errors.Join(ErrInvalidCommand, refreshErr)
+		}
+		result.snapshots = refreshed
 	}
 	return result, nil
 }
diff --git a/internal/controller/worker/snapshot_source.go b/internal/controller/worker/snapshot_source.go
index 20712b9..f820afe 100644
--- a/internal/controller/worker/snapshot_source.go
+++ b/internal/controller/worker/snapshot_source.go
@@ -25,6 +25,113 @@ type SnapshotSource interface {
 	Watch(context.Context, SnapshotWatchRequest) (<-chan *controlplanev1.WorkerSnapshot, error)
 }
 
+// RefreshingSnapshotSource keeps a Worker snapshot stream current without
+// requiring the Gateway to reconnect after every validity window. It only
+// forwards complete snapshots, so Gateway delta handling remains fail-closed.
+type RefreshingSnapshotSource struct {
+	source       SnapshotSource
+	refreshEvery time.Duration
+}
+
+func NewRefreshingSnapshotSource(source SnapshotSource, refreshEvery time.Duration) (*RefreshingSnapshotSource, error) {
+	if source == nil || refreshEvery <= 0 {
+		return nil, ErrSnapshotsUnavailable
+	}
+	return &RefreshingSnapshotSource{source: source, refreshEvery: refreshEvery}, nil
+}
+
+func snapshotRefreshEvery(validFor time.Duration) time.Duration {
+	if validFor <= 0 {
+		return 0
+	}
+	if half := validFor / 2; half > 0 {
+		return half
+	}
+	return validFor
+}
+
+func (source *RefreshingSnapshotSource) Watch(
+	ctx context.Context,
+	request SnapshotWatchRequest,
+) (<-chan *controlplanev1.WorkerSnapshot, error) {
+	if source == nil || source.source == nil || ctx == nil {
+		return nil, ErrSnapshotsUnavailable
+	}
+	first, err := source.next(ctx, request)
+	if err != nil {
+		return nil, err
+	}
+	updates := make(chan *controlplanev1.WorkerSnapshot, 1)
+	updates <- first
+	go source.refresh(ctx, updates, request, first)
+	return updates, nil
+}
+
+func (source *RefreshingSnapshotSource) refresh(
+	ctx context.Context,
+	updates chan<- *controlplanev1.WorkerSnapshot,
+	request SnapshotWatchRequest,
+	current *controlplanev1.WorkerSnapshot,
+) {
+	defer close(updates)
+	version := current.GetVersion()
+	checksum := append([]byte(nil), current.GetChecksum()...)
+	timer := time.NewTimer(source.delayFor(current))
+	defer timer.Stop()
+	for {
+		select {
+		case <-ctx.Done():
+			return
+		case <-timer.C:
+			request.LastAppliedVersion = version
+			request.LastChecksum = append(request.LastChecksum[:0], checksum...)
+			next, err := source.next(ctx, request)
+			if err != nil {
+				return
+			}
+			select {
+			case <-ctx.Done():
+				return
+			case updates <- next:
+				version = next.GetVersion()
+				checksum = append(checksum[:0], next.GetChecksum()...)
+				timer.Reset(source.delayFor(next))
+			}
+		}
+	}
+}
+
+func (source *RefreshingSnapshotSource) next(
+	ctx context.Context,
+	request SnapshotWatchRequest,
+) (*controlplanev1.WorkerSnapshot, error) {
+	stream, err := source.source.Watch(ctx, request)
+	if err != nil {
+		return nil, err
+	}
+	select {
+	case <-ctx.Done():
+		return nil, ctx.Err()
+	case snapshot, ok := <-stream:
+		if !ok || snapshot == nil || snapshot.GetVersion() <= request.LastAppliedVersion {
+			return nil, ErrSnapshotsUnavailable
+		}
+		return snapshot, nil
+	}
+}
+
+func (source *RefreshingSnapshotSource) delayFor(snapshot *controlplanev1.WorkerSnapshot) time.Duration {
+	delay := source.refreshEvery
+	if snapshot == nil || snapshot.GetGeneratedAt() == nil || snapshot.GetValidUntil() == nil {
+		return delay
+	}
+	lifetime := snapshot.GetValidUntil().AsTime().Sub(snapshot.GetGeneratedAt().AsTime())
+	if half := lifetime / 2; half > 0 && half < delay {
+		return half
+	}
+	return delay
+}
+
 type OwnershipEpochReader interface {
 	CurrentOwnershipEpoch(context.Context) (uint64, error)
 }
@@ -73,3 +180,4 @@ func (source *InitialSnapshotSource) Watch(ctx context.Context, request Snapshot
 }
 
 var _ SnapshotSource = (*InitialSnapshotSource)(nil)
+var _ SnapshotSource = (*RefreshingSnapshotSource)(nil)
diff --git a/internal/controller/worker/snapshot_source_test.go b/internal/controller/worker/snapshot_source_test.go
index 1b92840..fccd79c 100644
--- a/internal/controller/worker/snapshot_source_test.go
+++ b/internal/controller/worker/snapshot_source_test.go
@@ -3,9 +3,11 @@ package worker
 import (
 	"context"
 	"errors"
+	"sync"
 	"testing"
 	"time"
 
+	"google.golang.org/protobuf/types/known/timestamppb"
 	controlplanev1 "proxy-pool/gen/controlplane/v1"
 	"proxy-pool/internal/controlplane/snapshotwire"
 	ownershipDomain "proxy-pool/internal/domain/ownership"
@@ -35,6 +37,47 @@ func TestInitialSnapshotSourceIssuesNextFullSnapshot(t *testing.T) {
 	}
 }
 
+func TestRefreshingSnapshotSourceIssuesSequentialFullSnapshots(t *testing.T) {
+	base := &refreshingSnapshotSourceStub{}
+	source, err := NewRefreshingSnapshotSource(base, 10*time.Millisecond)
+	if err != nil {
+		t.Fatalf("NewRefreshingSnapshotSource() = %v", err)
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	updates, err := source.Watch(ctx, SnapshotWatchRequest{
+		WorkerID: "worker-a", SessionID: "session-a", LastAppliedVersion: 4, LastChecksum: []byte{4},
+	})
+	if err != nil {
+		t.Fatalf("Watch() = %v", err)
+	}
+	first := receiveSnapshot(t, updates)
+	second := receiveSnapshot(t, updates)
+	if first.GetVersion() != 5 || second.GetVersion() != 6 || second.GetVersion() <= first.GetVersion() {
+		t.Fatalf("refreshed versions = (%d, %d)", first.GetVersion(), second.GetVersion())
+	}
+	if calls := base.Calls(); len(calls) < 2 || calls[0] != 4 || calls[1] != 5 {
+		t.Fatalf("base requested versions = %v", calls)
+	}
+	if checksums := base.Checksums(); len(checksums) < 2 || string(checksums[0]) != string([]byte{4}) ||
+		string(checksums[1]) != string([]byte{5}) {
+		t.Fatalf("base requested checksums = %v", checksums)
+	}
+}
+
+func TestRefreshingSnapshotSourceRefreshesBeforeShortSnapshotExpiry(t *testing.T) {
+	source, err := NewRefreshingSnapshotSource(&refreshingSnapshotSourceStub{}, 2*time.Second)
+	if err != nil {
+		t.Fatalf("NewRefreshingSnapshotSource() = %v", err)
+	}
+	now := time.Now().UTC()
+	if delay := source.delayFor(&controlplanev1.WorkerSnapshot{
+		GeneratedAt: timestamppb.New(now), ValidUntil: timestamppb.New(now.Add(100 * time.Millisecond)),
+	}); delay != 50*time.Millisecond {
+		t.Fatalf("delayFor(short-lived snapshot) = %s, want 50ms", delay)
+	}
+}
+
 func TestOwnedSnapshotSourceBuildsBoundedProxySnapshot(t *testing.T) {
 	now := time.Date(2026, 7, 31, 12, 0, 0, 0, time.UTC)
 	expiresAt := now.Add(10 * time.Minute)
@@ -167,6 +210,57 @@ type epochReaderStub struct {
 	err   error
 }
 
+type refreshingSnapshotSourceStub struct {
+	mu        sync.Mutex
+	calls     []uint64
+	checksums [][]byte
+}
+
+func (source *refreshingSnapshotSourceStub) Watch(
+	_ context.Context,
+	request SnapshotWatchRequest,
+) (<-chan *controlplanev1.WorkerSnapshot, error) {
+	source.mu.Lock()
+	source.calls = append(source.calls, request.LastAppliedVersion)
+	source.checksums = append(source.checksums, append([]byte(nil), request.LastChecksum...))
+	source.mu.Unlock()
+	updates := make(chan *controlplanev1.WorkerSnapshot, 1)
+	updates <- &controlplanev1.WorkerSnapshot{
+		Version: request.LastAppliedVersion + 1, Checksum: []byte{byte(request.LastAppliedVersion + 1)},
+	}
+	return updates, nil
+}
+
+func (source *refreshingSnapshotSourceStub) Calls() []uint64 {
+	source.mu.Lock()
+	defer source.mu.Unlock()
+	return append([]uint64(nil), source.calls...)
+}
+
+func (source *refreshingSnapshotSourceStub) Checksums() [][]byte {
+	source.mu.Lock()
+	defer source.mu.Unlock()
+	checksums := make([][]byte, len(source.checksums))
+	for index := range source.checksums {
+		checksums[index] = append([]byte(nil), source.checksums[index]...)
+	}
+	return checksums
+}
+
+func receiveSnapshot(t *testing.T, updates <-chan *controlplanev1.WorkerSnapshot) *controlplanev1.WorkerSnapshot {
+	t.Helper()
+	select {
+	case snapshot, ok := <-updates:
+		if !ok || snapshot == nil {
+			t.Fatal("snapshot updates closed before next full snapshot")
+		}
+		return snapshot
+	case <-time.After(time.Second):
+		t.Fatal("timed out waiting for snapshot")
+		return nil
+	}
+}
+
 func (reader epochReaderStub) CurrentOwnershipEpoch(context.Context) (uint64, error) {
 	return reader.epoch, reader.err
 }
diff --git a/progress.md b/progress.md
index 6f320dc..4b9a920 100644
--- a/progress.md
+++ b/progress.md
@@ -15,6 +15,10 @@
 - Redis 集成标签测试在未设置 `PROXY_POOL_TEST_REDIS_URL` 时现能正确跳过所有契约
   子测试；`go test ./...`、`go test -tags=integration ./internal/adapters/redisactivity`
   与 `scripts/verify.ps1` 已通过，race 仍由 Linux CI 执行。
+- Worker 控制面改为持续完整 Snapshot 发布：Controller 在每份 Snapshot 有效期的一半
+  前重建并下发递增版本，Gateway 在同一长连接内原子替换视图；构建失败或流中断时才由
+  SessionSupervisor 按原有退避重连。该机制复用 `RefreshingSnapshotSource`，不降低
+  Delta 的 fail-closed 校验，也不把自动 Drain/ACK 记为已完成。
 
 ## 2026-07-30
 
diff --git a/task_plan.md b/task_plan.md
index 09304c9..8121e36 100644
--- a/task_plan.md
+++ b/task_plan.md
@@ -40,8 +40,8 @@
 14. [已完成] 收敛健康故障生命周期：将 `UNHEALTHY` 的持续时长纳入权威活动池状态，
     通过 Controller 后台的有界回收清理无 Worker ownership 的持续异常 Proxy；拥有
     Worker 的候选延后到既有 Drain/ACK 清除所有权后再处理。
-15. [进行中] 收敛 Worker 发布生命周期：补齐权威 Proxy/Routing 的增量发布和
-    自动 Drain 编排，使配置停用、健康淘汰与 Snapshot 撤销形成可观测闭环。
+15. [进行中] 收敛 Worker 发布生命周期：已完成权威 Proxy/Routing 的持续完整快照
+    刷新；后续补齐自动 Drain 编排，使配置停用、健康淘汰与 Snapshot 撤销形成可观测闭环。
 
 ## 串并行关系