diff --git a/litestream/README.md b/litestream/README.md
index f19a0e0..b8c7516 100644
--- a/litestream/README.md
+++ b/litestream/README.md
@@ -9,8 +9,7 @@ The `"litestream"` SQLite VFS implements Litestream
 
 See the [example](example_test.go) for how to use.
 
-To improve performance,
-increase `PollInterval` (and `MinLevel`) as much as you can,
+To improve performance, increase `PollInterval` as much as you can,
 and set [`PRAGMA cache_size=N`](https://www.sqlite.org/pragma.html#pragma_cache_size)
 (or use `_pragma=cache_size(N)`).
 
diff --git a/litestream/api.go b/litestream/api.go
index 6bb9554..ff54fe3 100644
--- a/litestream/api.go
+++ b/litestream/api.go
@@ -40,9 +40,6 @@ type ReplicaOptions struct {
 	// used by the replica at MinLevel+1.
 	PollInterval time.Duration
 
-	// Minimum compaction level to track.
-	MinLevel int
-
 	// CacheSize is the maximum size of the page cache in bytes.
 	// Zero means DefaultCacheSize, negative disables caching.
 	CacheSize int
@@ -61,7 +58,6 @@ func NewReplica(name string, client ReplicaClient, options ReplicaOptions) {
 	if options.CacheSize == 0 {
 		options.CacheSize = DefaultCacheSize
 	}
-	options.MinLevel = max(0, min(options.MinLevel, litestream.SnapshotLevel))
 
 	liteMtx.Lock()
 	defer liteMtx.Unlock()
diff --git a/litestream/vfs.go b/litestream/vfs.go
index f385a06..b4e218c 100644
--- a/litestream/vfs.go
+++ b/litestream/vfs.go
@@ -168,7 +168,7 @@ func (f *liteFile) Pragma(name, value string) (string, error) {
 		if txid == 0 {
 			// Outside transaction.
 			f.db.mtx.Lock()
-			txid = f.db.txids[f.db.opts.MinLevel]
+			txid = f.db.txids[0]
 			f.db.mtx.Unlock()
 		}
 		return txid.String(), nil
@@ -245,10 +245,10 @@ func (f *liteDB) pollReplica(ctx context.Context) (*pageIndex, ltx.TXID, error)
 
 	// Limit polling interval.
 	if time.Since(f.lastPoll) < f.opts.PollInterval {
-		return f.pages, f.txids[f.opts.MinLevel], nil
+		return f.pages, f.txids[0], nil
 	}
 
-	for level := range pollLevels(f.opts.MinLevel) {
+	for level := range []int{0, 1, litestream.SnapshotLevel} {
 		if err := f.updateLevel(ctx, level); err != nil {
 			f.opts.Logger.Error("cannot poll replica", "error", err)
 			return nil, 0, err
@@ -256,7 +256,7 @@ func (f *liteDB) pollReplica(ctx context.Context) (*pageIndex, ltx.TXID, error)
 	}
 
 	f.lastPoll = time.Now()
-	return f.pages, f.txids[f.opts.MinLevel], nil
+	return f.pages, f.txids[0], nil
 }
 
 // +checklocks:f.mtx
@@ -313,26 +313,10 @@ func (f *liteDB) updateInfo(ctx context.Context, info *ltx.FileInfo) error {
 	// Track the MaxTXID for each level.
 	maxTXID := &f.txids[info.Level]
 	*maxTXID = max(*maxTXID, info.MaxTXID)
+	f.txids[0] = max(f.txids[0], *maxTXID)
 	return nil
 }
 
-func pollLevels(minLevel int) (r []int) {
-	// Updating from lower to upper levels is non-racy,
-	// since LTX files are compacted into higher levels
-	// before the lower level LTX files are deleted.
-
-	// Also, only level 0 compactions and snapshots delete files,
-	// so the intermediate levels never need to be updated.
-
-	if minLevel <= 0 {
-		return append(r, 0, 1, litestream.SnapshotLevel)
-	}
-	if minLevel >= litestream.SnapshotLevel {
-		return append(r, litestream.SnapshotLevel)
-	}
-	return append(r, minLevel, litestream.SnapshotLevel)
-}
-
 // Type aliases; these are a mouthful.
 type pageIndex = wbt.Tree[uint32, ltx.PageIndexElem]
 type levelTXIDs = [litestream.SnapshotLevel + 1]ltx.TXID
diff --git a/litestream/vfs_test.go b/litestream/vfs_test.go
deleted file mode 100644
index 30237b8..0000000
--- a/litestream/vfs_test.go
+++ /dev/null
@@ -1,34 +0,0 @@
-package litestream
-
-import (
-	"slices"
-	"strconv"
-	"testing"
-
-	"github.com/benbjohnson/litestream"
-
-	_ "github.com/ncruces/go-sqlite3/embed"
-)
-
-func Test_pollLevels(t *testing.T) {
-	tests := []struct {
-		minLevel int
-		want     []int
-	}{
-		{minLevel: -1, want: []int{0, 1, litestream.SnapshotLevel}},
-		{minLevel: 0, want: []int{0, 1, litestream.SnapshotLevel}},
-		{minLevel: 1, want: []int{1, litestream.SnapshotLevel}},
-		{minLevel: 2, want: []int{2, litestream.SnapshotLevel}},
-		{minLevel: 3, want: []int{3, litestream.SnapshotLevel}},
-		{minLevel: litestream.SnapshotLevel, want: []int{litestream.SnapshotLevel}},
-		{minLevel: litestream.SnapshotLevel + 1, want: []int{litestream.SnapshotLevel}},
-	}
-	for _, tt := range tests {
-		t.Run(strconv.Itoa(tt.minLevel), func(t *testing.T) {
-			got := pollLevels(tt.minLevel)
-			if !slices.Equal(got, tt.want) {
-				t.Errorf("pollLevels() = %v, want %v", got, tt.want)
-			}
-		})
-	}
-}