diff --git a/internal/mod/module/escape.go b/internal/mod/module/escape.go
new file mode 100644
index 00000000000..7c0c44201f7
--- /dev/null
+++ b/internal/mod/module/escape.go
@@ -0,0 +1,68 @@
+package module
+
+import (
+	"fmt"
+	"strings"
+	"unicode/utf8"
+
+	"cuelang.org/go/internal/mod/semver"
+)
+
+// EscapePath returns the escaped form of the given module path
+// (without the major version suffix).
+// It fails if the module path is invalid.
+func EscapePath(path string) (escaped string, err error) {
+	if err := CheckPathWithoutVersion(path); err != nil {
+		return "", err
+	}
+	// Technically there's no need to escape capital letters because CheckPath
+	// doesn't allow them, but let's be defensive.
+	return escapeString(path)
+}
+
+// EscapeVersion returns the escaped form of the given module version.
+// Versions must be in (possibly non-canonical) semver form and must be valid file names
+// and not contain exclamation marks.
+func EscapeVersion(v string) (escaped string, err error) {
+	if !semver.IsValid(v) {
+		return "", &InvalidVersionError{
+			Version: v,
+			Err:     fmt.Errorf("version is not in semver syntax"),
+		}
+	}
+	if err := checkElem(v, filePath); err != nil || strings.Contains(v, "!") {
+		return "", &InvalidVersionError{
+			Version: v,
+			Err:     fmt.Errorf("disallowed version string"),
+		}
+	}
+	return escapeString(v)
+}
+
+func escapeString(s string) (escaped string, err error) {
+	haveUpper := false
+	for _, r := range s {
+		if r == '!' || r >= utf8.RuneSelf {
+			// This should be disallowed by CheckPath, but diagnose anyway.
+			// The correctness of the escaping loop below depends on it.
+			return "", fmt.Errorf("internal error: inconsistency in EscapePath")
+		}
+		if 'A' <= r && r <= 'Z' {
+			haveUpper = true
+		}
+	}
+
+	if !haveUpper {
+		return s, nil
+	}
+
+	var buf []byte
+	for _, r := range s {
+		if 'A' <= r && r <= 'Z' {
+			buf = append(buf, '!', byte(r+'a'-'A'))
+		} else {
+			buf = append(buf, byte(r))
+		}
+	}
+	return string(buf), nil
+}
diff --git a/internal/mod/module/module.go b/internal/mod/module/module.go
index 8be25893ed1..9fb478bf58c 100644
--- a/internal/mod/module/module.go
+++ b/internal/mod/module/module.go
@@ -14,6 +14,68 @@
 // There are no restrictions imposed directly by use of this structure,
 // but additional checking functions, most notably Check, verify that
 // a particular path, version pair is valid.
+//
+// # Escaped Paths
+//
+// Module versions appear as substrings of file system paths (as stored by
+// the modcache package).
+// In general we cannot rely on file systems to be case-sensitive. Although
+// module paths cannot currently contain upper case characters because
+// OCI registries forbid that, versions can. That
+// is, we cannot rely on the file system to keep foo.com/v@v1.0.0-PRE and
+// foo.com/v@v1.0.0-PRE separate. Windows and macOS don't. Instead, we must
+// never require two different casings of a file path.
+//
+// One possibility would be to make the escaped form be the lowercase
+// hexadecimal encoding of the actual path bytes. This would avoid ever
+// needing different casings of a file path, but it would be fairly illegible
+// to most programmers when those paths appeared in the file system
+// (including in file paths in compiler errors and stack traces)
+// in web server logs, and so on. Instead, we want a safe escaped form that
+// leaves most paths unaltered.
+//
+// The safe escaped form is to replace every uppercase letter
+// with an exclamation mark followed by the letter's lowercase equivalent.
+//
+// For example,
+//
+//	foo.com/v@v1.0.0-PRE ->  foo.com/v@v1.0.0-!p!r!e
+//
+// Versions that avoid upper-case letters are left unchanged.
+// Note that because import paths are ASCII-only and avoid various
+// problematic punctuation (like : < and >), the escaped form is also ASCII-only
+// and avoids the same problematic punctuation.
+//
+// Neither versions nor module paths allow exclamation marks, so there is no
+// need to define how to escape a literal !.
+//
+// # Unicode Restrictions
+//
+// Today, paths are disallowed from using Unicode.
+//
+// Although paths are currently disallowed from using Unicode,
+// we would like at some point to allow Unicode letters as well, to assume that
+// file systems and URLs are Unicode-safe (storing UTF-8), and apply
+// the !-for-uppercase convention for escaping them in the file system.
+// But there are at least two subtle considerations.
+//
+// First, note that not all case-fold equivalent distinct runes
+// form an upper/lower pair.
+// For example, U+004B ('K'), U+006B ('k'), and U+212A ('K' for Kelvin)
+// are three distinct runes that case-fold to each other.
+// When we do add Unicode letters, we must not assume that upper/lower
+// are the only case-equivalent pairs.
+// Perhaps the Kelvin symbol would be disallowed entirely, for example.
+// Or perhaps it would escape as "!!k", or perhaps as "(212A)".
+//
+// Second, it would be nice to allow Unicode marks as well as letters,
+// but marks include combining marks, and then we must deal not
+// only with case folding but also normalization: both U+00E9 ('é')
+// and U+0065 U+0301 ('e' followed by combining acute accent)
+// look the same on the page and are treated by some file systems
+// as the same path. If we do allow Unicode marks in paths, there
+// must be some kind of normalization to allow only one canonical
+// encoding of any character used in an import path.
 package module
 
 // IMPORTANT NOTE
@@ -50,10 +112,12 @@ func (m Version) Path() string {
 	return m.path
 }
 
+// Equal reports whether m is equal to m1.
 func (m Version) Equal(m1 Version) bool {
 	return m.path == m1.path && m.version == m1.version
 }
 
+// BasePath returns the path part of m without its major version suffix.
 func (m Version) BasePath() string {
 	basePath, _, ok := SplitPathVersion(m.path)
 	if !ok {
@@ -62,14 +126,23 @@ func (m Version) BasePath() string {
 	return basePath
 }
 
+// Version returns the version part of m. This is either
+// a canonical semver version or "none" or the empty string.
 func (m Version) Version() string {
 	return m.version
 }
 
+// IsValid reports whether m is non-zero.
 func (m Version) IsValid() bool {
 	return m.path != ""
 }
 
+// IsCanonical reports whether m is valid and has a canonical
+// semver version.
+func (m Version) IsCanonical() bool {
+	return m.IsValid() && m.version != "" && m.version != "none"
+}
+
 // String returns the string form of the Version:
 // (Path@Version, or just Path if Version is empty).
 func (m Version) String() string {
diff --git a/internal/mod/module/module_test.go b/internal/mod/module/module_test.go
index b161cb8d201..7a5c9025e0e 100644
--- a/internal/mod/module/module_test.go
+++ b/internal/mod/module/module_test.go
@@ -151,3 +151,49 @@ func TestParseVersion(t *testing.T) {
 		})
 	}
 }
+
+var escapeVersionTests = []struct {
+	v   string
+	esc string // empty means same as path
+}{
+	{v: "v1.2.3-alpha"},
+	{v: "v3"},
+	{v: "v2.3.1-ABcD", esc: "v2.3.1-!a!bc!d"},
+}
+
+func TestEscapeVersion(t *testing.T) {
+	for _, tt := range escapeVersionTests {
+		esc, err := EscapeVersion(tt.v)
+		if err != nil {
+			t.Errorf("EscapeVersion(%q): unexpected error: %v", tt.v, err)
+			continue
+		}
+		want := tt.esc
+		if want == "" {
+			want = tt.v
+		}
+		if esc != want {
+			t.Errorf("EscapeVersion(%q) = %q, want %q", tt.v, esc, want)
+		}
+	}
+}
+
+func TestEscapePath(t *testing.T) {
+	// Check invalid paths.
+	for _, tt := range checkPathWithoutVersionTests {
+		if tt.wantErr != "" {
+			_, err := EscapePath(tt.path)
+			if err == nil {
+				t.Errorf("EscapePath(%q): succeeded, want error (invalid path)", tt.path)
+			}
+		}
+	}
+	path := "foo.com/bar"
+	esc, err := EscapePath(path)
+	if err != nil {
+		t.Fatal(err)
+	}
+	if esc != path {
+		t.Fatalf("EscapePath(%q) = %q, want %q", path, esc, path)
+	}
+}