diff --git a/notes/pdftract-3g6ne.md b/notes/pdftract-3g6ne.md
new file mode 100644
index 0000000..05373a1
--- /dev/null
+++ b/notes/pdftract-3g6ne.md
@@ -0,0 +1,91 @@
+# pdftract-3g6ne: CMap Codespace Range Parser
+
+## Bead Summary
+
+Implemented the CMap codespace range parser for extracting byte-width boundaries from `begincodespacerange` / `endcodespacerange` PostScript blocks.
+
+## Implementation Location
+
+- Module: `crates/pdftract-core/src/font/codespace.rs`
+- Exported from: `crates/pdftract-core/src/font/mod.rs`
+
+## Structures Implemented
+
+### CodespaceRange
+```rust
+pub struct CodespaceRange {
+    pub lo: [u8; 4],   // Low bound (big-endian, 4-byte storage)
+    pub hi: [u8; 4],   // High bound (big-endian, 4-byte storage)
+    pub width: u8,     // Byte width (1-4)
+}
+```
+
+### CodespaceRanges
+```rust
+pub struct CodespaceRanges {
+    pub ranges: SmallVec<[CodespaceRange; 8]>,
+}
+```
+
+### CodespaceParser
+PostScript-style tokenizer that:
+- Recognizes `begincodespacerange` / `endcodespacerange` keywords
+- Parses hex string pairs `<lo> <hi>`
+- Validates width matching (lo.len() == hi.len())
+- Emits diagnostics on malformed entries
+- Continues parsing after errors (recovery)
+
+## Acceptance Criteria Status
+
+| Criterion | Status | Test |
+|-----------|--------|------|
+| Parse <00> <7F> → 1 range, width=1 | PASS | `test_parse_single_range_one_byte` |
+| Parse <00> <7F> <8000> <FFFF> in one block → 2 ranges | PASS | `test_parse_two_ranges_mixed_width` |
+| Width inference: 2-char hex → width=1; 4-char hex → width=2 | PASS | `test_width_inference` |
+| Case-insensitive hex (<C0> and <c0> equivalent) | PASS | `test_case_insensitive_hex` |
+| Malformed range (width mismatch) → diagnostic + skipped | PASS | `test_malformed_range_width_mismatch` |
+| Empty CMap → empty ranges | PASS | `test_empty_cmap`, `test_no_codespace_block` |
+| Round-trip with Identity-H CMap fixture | N/A | No standalone CMap fixtures exist; tests cover parsing logic |
+
+### Additional Tests
+
+- `test_jis_range`: JIS lead/trail 2-byte pattern `<8140> <FEFE>`
+- `test_three_byte_range`: 3-byte codespace support
+- `test_four_byte_range`: 4-byte codespace support
+- `test_invalid_width_too_large`: Rejects 5+ byte ranges
+- `test_find_range`: Utility to match byte sequences to ranges
+- `test_comment_in_block`: PostScript comment stripping
+- `test_hex_string_with_whitespace`: Internal whitespace handling
+- `test_odd_length_hex_string`: Dangling nibble padding
+- `test_recovery_on_error`: Continues after malformed entries
+- `test_convenience_function`: Public API entry points
+
+## Public API
+
+```rust
+// Parse without diagnostics (for internal use)
+pub fn parse_codespace_ranges(input: &[u8]) -> CodespaceRanges
+
+// Parse with diagnostics (for error reporting)
+pub fn parse_codespace_ranges_with_diags(input: &[u8]) -> (CodespaceRanges, Vec<Diagnostic>)
+```
+
+## Design Decisions
+
+1. **4-byte storage for bounds**: Ranges up to 4 bytes are stored in fixed `[u8; 4]` arrays with leading zeros, simplifying comparison logic
+2. **SmallVec capacity 8**: Most predefined CMaps (Identity-H/V, UTF-16 variants) have 1-2 ranges; 8 provides stack allocation for typical cases without overflow
+3. **Recovery over hard failure**: Malformed entries emit diagnostics but don't stop parsing; subsequent valid ranges are still collected
+4. **Case-insensitive hex**: Both `<C0>` and `<c0>` parse to `0xC0` per PDF spec
+5. **Width validation**: Rejects ranges where lo.len() != hi.len() or width > 4
+
+## Integration Notes
+
+- Module is imported in `font/mod.rs` but not yet exported at crate level
+- Sibling tokenizer bead will consume `CodespaceRanges` for multi-byte walking
+- Coordinator `pdftract-19oy` (CMap parser + tokenizer) depends on this module
+
+## Commit
+
+- Hash: `1dfaf73`
+- Message: `feat(pdftract-3g6ne): implement CMap codespace range parser`
+- Pushed: `forgejo main`