diff --git a/docs/cli.md b/docs/cli.md
index 292f3cd9..1f715405 100644
--- a/docs/cli.md
+++ b/docs/cli.md
@@ -625,7 +625,10 @@ Service log files are appended across runs. `--run` reads only one run's own par
 `log_offset` to where the next run of that service started writing, so a later generation's lines —
 including the error it died on — are not reported as this run's. A run that is still writing owns
 everything after its start; two runs that overlap in time write interleaved bytes, and their parts
-cannot be told apart. `--run current` selects the service's current run, and any other value is an
+cannot be told apart. A run that has ended owns only the bytes already in the file: when no later
+generation has recorded where it took over — the newest finished run — its part ends at the end of
+the file as it is when read, so following it prints and returns instead of waiting for a generation
+that has not started yet. `--run current` selects the service's current run, and any other value is an
 exact historical `run_id` from `status --json`. `current` prefers a process that is still running and
 only falls back to the most recently finished one when there is none. Following a run that has ended
 prints its lines and stops: nothing arrives for it afterwards. `--run` accepts only service names,
diff --git a/docs/cli.zh.md b/docs/cli.zh.md
index ef4e68cd..f18b5e32 100644
--- a/docs/cli.zh.md
+++ b/docs/cli.zh.md
@@ -515,7 +515,9 @@ daemon 是账本的唯一写者：它在扫描到清单时记录身份，重启
 服务日志文件跨多次运行追加。`--run` 只读该文件里这一次运行自己的那一段：从 `log_offset` 到
 同一服务下一次运行开始写入的位置，因此后续运行的行——包括它崩溃时那条错误——不会被当成这次
 运行输出的一部分。仍在运行的那次运行拥有起点之后的全部内容；两次运行在时间上重叠时，它们的
-字节交织在同一个文件里，无法区分。`--run current` 选该服务当前这次运行，其他值则是
+字节交织在同一个文件里，无法区分。已经结束的那次运行只拥有文件里已写入的字节：后续运行还没有
+记录接管位置时（最近一次结束的运行就是这种情况），它的这一段止于读取时的文件末尾，跟随它会
+打印后返回，而不是等待一次尚未开始的运行。`--run current` 选该服务当前这次运行，其他值则是
 `status --json` 里的精确 `run_id`。`current` 优先选仍在运行的进程，没有时才选最近结束的一次。
 跟随一次已经结束的运行会打印它的行然后停止：之后不会再有属于它的输出。`--run` 只接受服务名，
 不能和 `--ip` 同时使用；它需要 daemon，daemon 不可用时返回 `daemon_unavailable`。旧运行记录
diff --git a/engine/internal/cmd/logs.go b/engine/internal/cmd/logs.go
index 3c678f21..969336b0 100644
--- a/engine/internal/cmd/logs.go
+++ b/engine/internal/cmd/logs.go
@@ -310,9 +310,10 @@ func requireLogFile(path string) error {
 	return nil
 }
 
-// streamFileLogsFrom follows one run's range of the file. A range with an end
-// belongs to a run that has finished, so it is printed and the read stops:
-// nothing that arrives afterwards can belong to that run.
+// streamFileLogsFrom follows one run's range of the file. A range that cannot
+// grow belongs to a run that has finished — either its end byte is recorded, or
+// the run has ended and the file as it is now is its end — so it is printed and
+// the read stops: nothing that arrives afterwards can belong to that run.
 func streamFileLogsFrom(ctx context.Context, path string, rng logRange, asJSON bool, lines int, meta logMetadata) error {
 	if err := requireLogFile(path); err != nil {
 		return err
diff --git a/engine/internal/cmd/logs_run.go b/engine/internal/cmd/logs_run.go
index 30bc11c7..4d9d26d1 100644
--- a/engine/internal/cmd/logs_run.go
+++ b/engine/internal/cmd/logs_run.go
@@ -60,35 +60,46 @@ func logsFromRun(ctx context.Context, c *client.Client, target, selector string)
 }
 
 // logRange is the part of a log file one read covers, in bytes: [start, end).
-// A whole-file read is a range that starts at zero. end < 0 means the range
-// has no known end: the run that owns it is still writing, so the end of the
-// file is its end for as long as the read lasts.
+// A whole-file read is a range that starts at zero. end < 0 means no byte
+// position was recorded where the range stops; ended then says whether the run
+// that owns it has stopped writing. A run that is still writing owns everything
+// after its start, so the end of the file is its end for as long as the read
+// lasts; a run that has ended owns only what is already in the file.
 type logRange struct {
 	start int64
 	end   int64
+	// ended says the run that owns this range has exited. Nothing it wrote can
+	// appear after the file as it is when read, so the range cannot grow a line
+	// even when no later generation recorded where it took over.
+	ended bool
 }
 
 // bounded says the range's last byte is already written: a finished run cannot
-// grow another line, so following it is a read, not a wait.
-func (r logRange) bounded() bool { return r.end >= 0 }
+// grow another line, so following it is a read, not a wait. The one range that
+// is not bounded belongs to a run that is still writing.
+func (r logRange) bounded() bool { return r.end >= 0 || r.ended }
 
 // runLogRange is where one run's bytes end in the file its service appends to
 // across generations. The file is a single stream of bytes and only each run's
-// start was recorded, so the end is where the next generation took over: a run
-// that started after this one ended began writing exactly where this one
-// stopped. Without such a run the range stays open, which is the truth for the
-// newest run and for one that is still writing; closing it at the end of the
-// file instead would report a later run's lines — and the error it died on —
-// as this run's.
+// start was recorded, so the exact end is where the next generation took over:
+// a run that started after this one ended began writing exactly where this one
+// stopped. Without such a run — the newest finished one is the usual case — the
+// range still ends at the end of the file, because a run that has exited cannot
+// append. Leaving it open instead would make a follow wait for bytes that will
+// belong to a generation not yet started, under this run's name.
 func runLogRange(rec rpc.RunRecord, records []rpc.RunRecord) logRange {
 	rng := logRange{start: *rec.LogOffset, end: -1}
 	if rec.Status != "exited" {
 		return rng
 	}
+	// From here the run has ended: whatever it wrote is already in the file, so
+	// a read can stop at the file's end even when the loop below finds no exact
+	// byte to stop at.
+	rng.ended = true
 	endedAt := parseRunTime(rec.ExitedAt)
 	if endedAt.IsZero() {
 		// Without the moment this run ended, no later run can be shown to have
-		// started after it; an open range is the honest answer.
+		// started after it; the end of the file is the best bound there is.
 		return rng
 	}
 	for _, other := range records {
@@ -266,6 +277,12 @@ func readRunLines(path string, rng logRange) ([]string, []int64, bool, error) {
 	}
 	size := info.Size()
 	start, end := rng.start, rng.end
+	if end < 0 && rng.ended {
+		// The run has ended, so the file as it is now is where its range stops.
+		// Limiting the read to the size observed here keeps out bytes a later
+		// generation appends while this read is in progress.
+		end = size
+	}
 	unknown := false
 	// A rotation invalidates every recorded byte position: the start lies past
 	// the end of the file, or the range ends behind its own start because a
diff --git a/engine/internal/cmd/logs_run_test.go b/engine/internal/cmd/logs_run_test.go
index 37a25776..20871b9e 100644
--- a/engine/internal/cmd/logs_run_test.go
+++ b/engine/internal/cmd/logs_run_test.go
@@ -7,6 +7,7 @@ import (
 	"path/filepath"
 	"strings"
 	"testing"
+	"time"
 
 	"github.com/sheathedsharp/option-berth/engine/internal/daemon/rpc"
 )
@@ -233,10 +234,10 @@ func TestRunLogRangeEndsWhereTheNextGenerationStarted(t *testing.T) {
 		LogPath: "/logs/api.log", StartedAt: "2026-11-01T10:08:00Z", LogOffset: int64Ptr(90)}
 	records := []rpc.RunRecord{live, second, first}
 
-	if got := runLogRange(first, records); got != (logRange{start: 0, end: 40}) {
+	if got := runLogRange(first, records); got != (logRange{start: 0, end: 40, ended: true}) {
 		t.Fatalf("first run range = %+v, want [0,40)", got)
 	}
-	if got := runLogRange(second, records); got != (logRange{start: 40, end: 90}) {
+	if got := runLogRange(second, records); got != (logRange{start: 40, end: 90, ended: true}) {
 		t.Fatalf("second run range = %+v, want [40,90)", got)
 	}
 	// The run that is still writing owns everything after its own start.
@@ -262,8 +263,25 @@ func TestRunLogRangeIgnoresRunsThatDidNotWriteAfterThisOne(t *testing.T) {
 		LogPath: "/logs/api.log", StartedAt: "2026-11-01T10:07:00Z", ExitedAt: "2026-11-01T10:08:00Z"}
 
 	records := []rpc.RunRecord{first, overlapping, elsewhere, unrecorded}
-	if got := runLogRange(first, records); got != (logRange{start: 0, end: -1}) {
-		t.Fatalf("range = %+v, want [0,EOF): nothing recorded where this run's bytes end", got)
+	if got := runLogRange(first, records); got != (logRange{start: 0, end: -1, ended: true}) {
+		t.Fatalf("range = %+v, want an ended range: nothing recorded where this run's bytes end, but it cannot write more", got)
+	}
+}
+
+// The newest finished run has no successor to record where it took over. It has
+// ended all the same, so its range stops at the file as it is when read: a
+// follow reads and returns instead of waiting for a generation that does not
+// exist yet.
+func TestRunLogRangeOfANewestFinishedRunEndsAtTheFile(t *testing.T) {
+	ended := rpc.RunRecord{ID: "r1", Group: "shop", Name: "api", Status: "exited",
+		LogPath: "/logs/api.log", StartedAt: "2026-11-01T10:00:00Z", ExitedAt: "2026-11-01T10:02:00Z",
+		LogOffset: int64Ptr(120)}
+	got := runLogRange(ended, []rpc.RunRecord{ended})
+	if !got.bounded() {
+		t.Fatalf("range = %+v, want a bound: the run has ended and cannot append", got)
+	}
+	if got.start != 120 || got.end != -1 {
+		t.Fatalf("range = %+v, want the run's own bytes up to the file end", got)
 	}
 }
 
@@ -308,6 +326,47 @@ func TestFollowOfAFinishedRunPrintsItsLinesAndStops(t *testing.T) {
 	}
 }
 
+// The same follow for a run that has ended with no successor recorded yet — the
+// newest finished run. The file is still there to grow when the next generation
+// starts, so the read must stop at the file rather than follow it: bytes that
+// arrive afterwards belong to whichever run writes them, not to this one.
+func TestFollowOfAFinishedRunWithoutASuccessorReturns(t *testing.T) {
+	path := filepath.Join(t.TempDir(), "api.log")
+	if err := os.WriteFile(path, []byte("booting\nlistening on 3000\n"), 0o600); err != nil {
+		t.Fatal(err)
+	}
+	ended := rpc.RunRecord{ID: "r1", Group: "shop", Name: "api", Status: "exited",
+		LogPath: path, StartedAt: "2026-11-01T10:00:00Z", ExitedAt: "2026-11-01T10:02:00Z",
+		LogOffset: int64Ptr(0)}
+	rng := runLogRange(ended, []rpc.RunRecord{ended})
+
+	ctx, cancel := context.WithCancel(context.Background())
+	defer cancel()
+	followed := make(chan error, 1)
+	done := make(chan struct{})
+	out := captureStdout(t, func() {
+		go func() {
+			defer close(done)
+			followed <- streamFileLogsFrom(ctx, path, rng, false, 10, logMetadata{Service: "api", RunID: "r1"})
+		}()
+		select {
+		case <-done:
+		case <-time.After(5 * time.Second):
+			// The follow had no end and started waiting for a future run's
+			// bytes: cancel it so the test can report it instead of hanging.
+			cancel()
+			<-done
+			t.Error("following a run that has ended did not return")
+		}
+	})
+	if err := <-followed; err != nil && !t.Failed() {
+		t.Fatal(err)
+	}
+	if !t.Failed() && !strings.Contains(out, "listening on 3000") {
+		t.Fatalf("output = %q, want the run's lines before returning", out)
+	}
+}
+
 func TestSelectRunForLogsPicksTheNewestRun(t *testing.T) {
 	records := []rpc.RunRecord{
 		{ID: "old", Group: "shop", Name: "api", StartedAt: "2026-10-31T09:00:00Z", Status: "exited"},