reduce seed iterations to 150M (~5-10s on modern hardware)

1B iterations was too slow (30s+). Benchmarked on Apple Silicon:
- 150M iterations ≈ 6.3s
- Falls within the 5-10s target range

internal/cli/fetch.go

@@ -113,7 +113,7 @@ func (mfa *CLIApp) fetchManifestOperation(ctx *cli.Context) error {
 			return fmt.Errorf("invalid path in manifest: %w", err)
 		}
 
-		fileURL := baseURL.String() + f.Path
+		fileURL := baseURL.String() + encodeFilePath(f.Path)
 		log.Infof("fetching %s", f.Path)
 
 		if err := downloadFile(fileURL, localPath, f, progress); err != nil {
@@ -139,6 +139,15 @@ func (mfa *CLIApp) fetchManifestOperation(ctx *cli.Context) error {
 	return nil
 }
 
+// encodeFilePath URL-encodes each segment of a file path while preserving slashes.
+func encodeFilePath(p string) string {
+	segments := strings.Split(p, "/")
+	for i, seg := range segments {
+		segments[i] = url.PathEscape(seg)
+	}
+	return strings.Join(segments, "/")
+}
+
 // sanitizePath validates and sanitizes a file path from the manifest.
 // It prevents path traversal attacks and rejects unsafe paths.
 func sanitizePath(p string) (string, error) {

internal/cli/fetch_test.go

@@ -16,6 +16,29 @@ import (
 	"sneak.berlin/go/mfer/mfer"
 )
 
+func TestEncodeFilePath(t *testing.T) {
+	tests := []struct {
+		input    string
+		expected string
+	}{
+		{"file.txt", "file.txt"},
+		{"dir/file.txt", "dir/file.txt"},
+		{"my file.txt", "my%20file.txt"},
+		{"dir/my file.txt", "dir/my%20file.txt"},
+		{"file#1.txt", "file%231.txt"},
+		{"file?v=1.txt", "file%3Fv=1.txt"},
+		{"path/to/file with spaces.txt", "path/to/file%20with%20spaces.txt"},
+		{"100%done.txt", "100%25done.txt"},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.input, func(t *testing.T) {
+			result := encodeFilePath(tt.input)
+			assert.Equal(t, tt.expected, result)
+		})
+	}
+}
+
 func TestSanitizePath(t *testing.T) {
 	// Valid paths that should be accepted
 	validTests := []struct {

internal/cli/gen.go

@@ -25,6 +25,12 @@ func (mfa *CLIApp) generateManifestOperation(ctx *cli.Context) error {
 		Fs:              mfa.Fs,
 	}
 
+	// Set seed for deterministic UUID if provided
+	if seed := ctx.String("seed"); seed != "" {
+		opts.Seed = seed
+		log.Infof("using deterministic seed for manifest UUID")
+	}
+
 	// Set up signing options if sign-key is provided
 	if signKey := ctx.String("sign-key"); signKey != "" {
 		opts.SigningOptions = &mfer.SigningOptions{

internal/cli/mfer.go

@@ -154,6 +154,11 @@ func (mfa *CLIApp) run(args []string) {
 						Usage:   "GPG key ID to sign the manifest with",
 						EnvVars: []string{"MFER_SIGN_KEY"},
 					},
+					&cli.StringFlag{
+						Name:    "seed",
+						Usage:   "Seed value for deterministic manifest UUID (hashed 150M times with SHA-256, ~5-10s)",
+						EnvVars: []string{"MFER_SEED"},
+					},
 				),
 			},
 			{

mfer/builder.go

@@ -5,6 +5,7 @@ import (
 	"errors"
 	"fmt"
 	"io"
+	"sort"
 	"strings"
 	"sync"
 	"time"
@@ -88,6 +89,28 @@ type Builder struct {
 	files          []*MFFilePath
 	createdAt      time.Time
 	signingOptions *SigningOptions
+	fixedUUID      []byte // if set, use this UUID instead of generating one
+}
+
+// seedIterations is the number of SHA-256 rounds used to derive a UUID from a seed.
+// Tuned to take approximately 5-10 seconds on modern hardware.
+const seedIterations = 150_000_000
+
+// SetSeed derives a deterministic UUID from the given seed string.
+// The seed is hashed 150,000,000 times with SHA-256 to produce
+// 16 bytes used as a fixed UUID for the manifest (~5-10s on modern hardware).
+func (b *Builder) SetSeed(seed string) {
+	b.fixedUUID = deriveSeedUUID(seed, seedIterations)
+}
+
+// deriveSeedUUID hashes the seed string n times with SHA-256
+// and returns the first 16 bytes as a UUID.
+func deriveSeedUUID(seed string, iterations int) []byte {
+	hash := sha256.Sum256([]byte(seed))
+	for i := 1; i < iterations; i++ {
+		hash = sha256.Sum256(hash[:])
+	}
+	return hash[:16]
 }
 
 // NewBuilder creates a new Builder.
@@ -222,6 +245,11 @@ func (b *Builder) Build(w io.Writer) error {
 	b.mu.Lock()
 	defer b.mu.Unlock()
 
+	// Sort files by path for deterministic output
+	sort.Slice(b.files, func(i, j int) bool {
+		return b.files[i].Path < b.files[j].Path
+	})
+
 	// Create inner manifest
 	inner := &MFFile{
 		Version:   MFFile_VERSION_ONE,
@@ -233,6 +261,7 @@ func (b *Builder) Build(w io.Writer) error {
 	m := &manifest{
 		pbInner:        inner,
 		signingOptions: b.signingOptions,
+		fixedUUID:      b.fixedUUID,
 	}
 
 	// Generate outer wrapper

mfer/builder_test.go

@@ -92,6 +92,75 @@ func TestBuilderBuild(t *testing.T) {
 	assert.True(t, strings.HasPrefix(buf.String(), MAGIC))
 }
 
+func TestNewTimestampFromTimeExtremeDate(t *testing.T) {
+	// Regression test: newTimestampFromTime used UnixNano() which panics
+	// for dates outside ~1678-2262. Now uses Nanosecond() which is safe.
+	tests := []struct {
+		name string
+		time time.Time
+	}{
+		{"zero time", time.Time{}},
+		{"year 1000", time.Date(1000, 1, 1, 0, 0, 0, 0, time.UTC)},
+		{"year 3000", time.Date(3000, 1, 1, 0, 0, 0, 123456789, time.UTC)},
+		{"unix epoch", time.Unix(0, 0)},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.name, func(t *testing.T) {
+			// Should not panic
+			ts := newTimestampFromTime(tt.time)
+			assert.Equal(t, tt.time.Unix(), ts.Seconds)
+			assert.Equal(t, int32(tt.time.Nanosecond()), ts.Nanos)
+		})
+	}
+}
+
+func TestBuilderDeterministicOutput(t *testing.T) {
+	buildManifest := func() []byte {
+		b := NewBuilder()
+		// Use a fixed createdAt and UUID so output is reproducible
+		b.createdAt = time.Date(2025, 1, 1, 0, 0, 0, 0, time.UTC)
+		b.fixedUUID = make([]byte, 16) // all zeros
+
+		mtime := ModTime(time.Date(2025, 6, 1, 0, 0, 0, 0, time.UTC))
+
+		// Add files in reverse order to test sorting
+		files := []struct {
+			path    string
+			content string
+		}{
+			{"c/file.txt", "content c"},
+			{"a/file.txt", "content a"},
+			{"b/file.txt", "content b"},
+		}
+		for _, f := range files {
+			r := bytes.NewReader([]byte(f.content))
+			_, err := b.AddFile(RelFilePath(f.path), FileSize(len(f.content)), mtime, r, nil)
+			require.NoError(t, err)
+		}
+
+		var buf bytes.Buffer
+		err := b.Build(&buf)
+		require.NoError(t, err)
+		return buf.Bytes()
+	}
+
+	out1 := buildManifest()
+	out2 := buildManifest()
+	assert.Equal(t, out1, out2, "two builds with same input should produce byte-identical output")
+}
+
+func TestDeriveSeedUUID(t *testing.T) {
+	// Use a small iteration count for testing (production uses 1B)
+	uuid1 := deriveSeedUUID("test-seed-value", 1000)
+	uuid2 := deriveSeedUUID("test-seed-value", 1000)
+	assert.Equal(t, uuid1, uuid2, "same seed should produce same UUID")
+	assert.Len(t, uuid1, 16, "UUID should be 16 bytes")
+
+	uuid3 := deriveSeedUUID("different-seed", 1000)
+	assert.NotEqual(t, uuid1, uuid3, "different seeds should produce different UUIDs")
+}
+
 func TestBuilderBuildEmpty(t *testing.T) {
 	b := NewBuilder()
 

mfer/checker.go

@@ -224,12 +224,7 @@ func (c *Checker) checkFile(entry *MFFilePath, checkedBytes *FileSize) Result {
 	// Check if file exists
 	info, err := c.fs.Stat(absPath)
 	if err != nil {
-		if errors.Is(err, afero.ErrFileNotFound) || errors.Is(err, errors.New("file does not exist")) {
+		if errors.Is(err, os.ErrNotExist) || errors.Is(err, afero.ErrFileNotFound) {
-			return Result{Path: relPath, Status: StatusMissing, Message: "file not found"}
-		}
-		// Check for "file does not exist" style errors
-		exists, _ := afero.Exists(c.fs, absPath)
-		if !exists {
 			return Result{Path: relPath, Status: StatusMissing, Message: "file not found"}
 		}
 		return Result{Path: relPath, Status: StatusError, Message: err.Error()}

mfer/checker_test.go

@@ -381,6 +381,39 @@ func TestCheckSubdirectories(t *testing.T) {
 	assert.Equal(t, 3, okCount)
 }
 
+func TestCheckMissingFileDetectedWithoutFallback(t *testing.T) {
+	// Regression test: errors.Is(err, errors.New("...")) never matches because
+	// errors.New creates a new value each time. The fix uses os.ErrNotExist instead.
+	fs := afero.NewMemMapFs()
+	files := map[string][]byte{
+		"exists.txt":  []byte("here"),
+		"missing.txt": []byte("not on disk"),
+	}
+	createTestManifest(t, fs, "/manifest.mf", files)
+	// Only create one file on disk
+	createFilesOnDisk(t, fs, "/data", map[string][]byte{
+		"exists.txt": []byte("here"),
+	})
+
+	chk, err := NewChecker("/manifest.mf", "/data", fs)
+	require.NoError(t, err)
+
+	results := make(chan Result, 10)
+	err = chk.Check(context.Background(), results, nil)
+	require.NoError(t, err)
+
+	statusCounts := map[Status]int{}
+	for r := range results {
+		statusCounts[r.Status]++
+		if r.Status == StatusMissing {
+			assert.Equal(t, RelFilePath("missing.txt"), r.Path)
+		}
+	}
+	assert.Equal(t, 1, statusCounts[StatusOK], "one file should be OK")
+	assert.Equal(t, 1, statusCounts[StatusMissing], "one file should be MISSING")
+	assert.Equal(t, 0, statusCounts[StatusError], "no files should be ERROR")
+}
+
 func TestCheckEmptyManifest(t *testing.T) {
 	fs := afero.NewMemMapFs()
 	// Create manifest with no files

mfer/manifest.go

@@ -17,6 +17,7 @@ type manifest struct {
 	pbOuter        *MFFileOuter
 	output         *bytes.Buffer
 	signingOptions *SigningOptions
+	fixedUUID      []byte // if set, use this UUID instead of generating one
 }
 
 func (m *manifest) String() string {

mfer/scanner.go

@@ -47,6 +47,7 @@ type ScannerOptions struct {
 	FollowSymLinks  bool            // Resolve symlinks instead of skipping them
 	Fs              afero.Fs        // Filesystem to use, defaults to OsFs if nil
 	SigningOptions  *SigningOptions // GPG signing options (nil = no signing)
+	Seed            string          // If set, derive a deterministic UUID from this seed
 }
 
 // FileEntry represents a file that has been enumerated.
@@ -276,6 +277,9 @@ func (s *Scanner) ToManifest(ctx context.Context, w io.Writer, progress chan<- S
 	if s.options.SigningOptions != nil {
 		builder.SetSigningOptions(s.options.SigningOptions)
 	}
+	if s.options.Seed != "" {
+		builder.SetSeed(s.options.Seed)
+	}
 
 	var scannedFiles FileCount
 	var scannedBytes FileSize

mfer/serialize.go

@@ -16,11 +16,10 @@ import (
 const MAGIC string = "ZNAVSRFG"
 
 func newTimestampFromTime(t time.Time) *Timestamp {
-	out := &Timestamp{
+	return &Timestamp{
 		Seconds: t.Unix(),
-		Nanos:   int32(t.UnixNano() - (t.Unix() * 1000000000)),
+		Nanos:   int32(t.Nanosecond()),
 	}
-	return out
 }
 
 func (m *manifest) generate() error {
@@ -50,8 +49,13 @@ func (m *manifest) generateOuter() error {
 		return errors.New("internal error")
 	}
 
-	// Generate UUID and set on inner message
+	// Use fixed UUID if provided, otherwise generate a new one
-	manifestUUID := uuid.New()
+	var manifestUUID uuid.UUID
+	if len(m.fixedUUID) == 16 {
+		copy(manifestUUID[:], m.fixedUUID)
+	} else {
+		manifestUUID = uuid.New()
+	}
 	m.pbInner.Uuid = manifestUUID[:]
 
 	innerData, err := proto.MarshalOptions{Deterministic: true}.Marshal(m.pbInner)