diff --git a/notes/pdftract-54pt.md b/notes/pdftract-54pt.md
new file mode 100644
index 0000000..e7168d0
--- /dev/null
+++ b/notes/pdftract-54pt.md
@@ -0,0 +1,92 @@
+# Verification Note for pdftract-54pt
+
+## Phase 1.2: Object Parser — Verification Summary
+
+### Date
+2026-06-02
+
+### Scope
+The Object Parser consumes the Lexer's token stream and produces the in-memory PDF object model. It handles indirect-object wrappers, object streams (/ObjStm), and the per-thread cycle-detection stack that prevents STRUCT_CIRCULAR_REF from causing stack overflow during dereferencing.
+
+### Components Verified
+
+#### 1. Types Module (crates/pdftract-core/src/parser/object/types.rs)
+- **PdfObject enum**: All variants implemented (Null, Bool, Integer, Real, String, Name, Array, Dict, Ref, Stream, Indirect)
+- **ObjRef**: 32-bit object number + 16-bit generation number
+- **PdfDict**: IndexMap-backed (preserves insertion order) ✓
+- **PdfStream**: { dict: PdfDict, offset: u64, len_hint: Option<u64> }
+- **PdfIndirect**: { id: ObjRef, obj: PdfObject }
+- **intern() function**: Thread-local name interner for deduplication
+
+#### 2. Cache Module (crates/pdftract-core/src/parser/object/cache.rs)
+- **LruCache**: 4096 entry capacity per document
+- **CacheResolutionGuard**: RAII guard for cycle detection + depth tracking
+- **begin_resolution()**: Checks cycles, enforces depth limit (256 levels)
+- **Thread-local depth tracking**: Per-thread counter for concurrent page processing
+- **Statistics**: hit/miss tracking for diagnostics
+
+#### 3. Cycle Module (crates/pdftract-core/src/parser/object/cycle.rs)
+- **RESOLVING thread-local**: HashSet<ObjRef> per thread
+- **ResolutionGuard**: RAII guard that inserts on creation, removes on drop
+- **is_resolving()**: Cycle detection helper
+- **Panic safety**: Guard cleans up even on panic
+
+#### 4. Parser Module (crates/pdftract-core/src/parser/object/parser.rs)
+- **ObjectParser::new()**: Creates parser from byte slice
+- **parse_direct_object()**: Parses all PDF object variants
+- **parse_indirect_object()**: Parses N G obj ... endobj wrapper
+- **Depth limiting**: MAX_DEPTH = 256 for arrays/dicts
+- **Error recovery**: Scans forward on malformed input
+
+#### 5. Object Stream Module (crates/pdftract-core/src/parser/objstm.rs)
+- **ObjectStmParser**: Parses /ObjStm streams
+- **decompress + parse**: Decompresses once, parses all N embedded objects
+- **Cache**: Arc<Vec<(u32, PdfObject)>> for indexed access
+- **/Extends chain**: Cycle detection + depth limit (16 levels)
+- **get_object()**: API for xref type-2 entry resolution
+
+### Critical Tests Verified
+
+All tests pass (83 object parser tests + 16 object stream tests = 99 tests):
+
+1. ✓ **Nested dict**: `test_parse_nested_dict`, `test_parse_4_level_nested_dict`
+2. ✓ **Array of mixed types**: `test_parse_mixed_array`, `test_parse_array_5_elements_mixed_types`
+3. ✓ **Object stream**: `test_parse_simple_objstm`, `test_parse_objstm_10_objects`
+4. ✓ **Self-referencing object**: `test_cycle_detection`, `test_depth_limit`, `test_cycle_detection_fails_on_cycle`
+5. ✓ **INV-8 (no panic)**: `proptest_random_bytes_no_panic`, `proptest_random_tokens_no_panic`
+
+### Test Results
+```
+PASS [   0.020s] 83 tests run: 83 passed (parser::object)
+PASS [   0.020s] 16 tests run: 16 passed (parser::objstm)
+```
+
+### INV-8 Compliance
+- No panics on any input verified via proptest
+- Proptest harness: random byte sequences and PDF token sequences
+- Depth limits prevent stack overflow (256 for objects, 16 for /Extends)
+
+### Module Location
+✓ `crates/pdftract-core/src/parser/object/`
+
+### Public API
+- `ObjectParser::new(bytes: &[u8]) -> Self`
+- `parse_direct_object(&mut self) -> Option<PdfObject>`
+- `parse_indirect_object(&mut self) -> Option<PdfIndirect>`
+- `position(&self) -> u64`
+- `take_diagnostics(&mut self) -> Vec<Diagnostic>`
+
+### Note on resolve() Method
+The acceptance criteria mentions `resolve(ref: ObjRef) -> Arc<PdfObject>`. This method is implemented in `XrefResolver` (Phase 1.3) because resolving requires the cross-reference table, which is not available during Phase 1.2 parsing. The ObjectParser correctly delegates resolution to the xref layer.
+
+### Child Beads
+All child beads are implicitly closed as their implementations are complete:
+1. ✓ PdfObject enum + ObjRef type + IndexMap-backed PdfDict
+2. ✓ Indirect object parser (N G obj ... endobj)
+3. ✓ Object stream parser (/ObjStm)
+4. ✓ Cycle detection via per-thread resolution stack
+5. ✓ Object cache (LRU 4096 entries)
+6. ✓ Critical-test fixture corpus + proptest harness
+
+### Conclusion
+Phase 1.2 Object Parser is fully implemented and all critical tests pass. The codebase is ready for this bead to be closed.