============================================================================
OmniBuf-Py Benchmark Suite (Python 3.14.5 / darwin-arm64 / Apple M1 Pro)
Workload: 11-field PeerUpdate (366B OmniBuf, 341B Protobuf) + >16 KiB stress cases
Methodology: Full decode reads all 11 fields; mutations check city and increment/grow.
============================================================================

### 1. Full Message Encode Speed

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | Wire Size (B) | vs Baseline |
|---|---:|---:|---:|---:|
| OmniBuf (Generated) | 782 ns | 1,278,457 | 366 | 1.00x (baseline) |
| OmniBuf (@dataclass) | 897 ns | 1,115,246 | 366 | 1.15x slower |
| protobuf-py | 1,281 ns | 780,629 | 341 | 1.64x slower |
| google.protobuf (upb) | 568 ns | 1,759,873 | 341 | 1.38x faster |
| msgspec | 824 ns | 1,214,299 | 634 | 1.05x slower |
| msgpack | 2,196 ns | 455,352 | 634 | 2.81x slower |
| json (stdlib) | 6,276 ns | 159,348 | 874 | 8.02x slower |

### 2. Full Message Decode Speed

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | vs Baseline |
|---|---:|---:|---:|
| OmniBuf (Generated) | 2,777 ns | 360,124 | 1.00x (baseline) |
| OmniBuf (@dataclass) | 2,856 ns | 350,166 | 1.00x (baseline) |
| protobuf-py | 4,220 ns | 236,984 | 1.52x slower |
| google.protobuf (upb) | 5,982 ns | 167,165 | 2.15x slower |
| msgspec | 2,817 ns | 354,952 | 1.00x (baseline) |
| msgpack | 5,299 ns | 188,727 | 1.91x slower |
| json (stdlib) | 8,024 ns | 124,631 | 2.89x slower |

### 3. Read a Single Nested Field

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | vs Baseline |
|---|---:|---:|---:|
| OmniBuf (Generated) | 521 ns | 1,917,890 | 1.00x (baseline) |
| protobuf-py | 2,845 ns | 351,555 | 5.46x slower |
| google.protobuf (upb) | 1,037 ns | 964,141 | 1.99x slower |
| msgspec | 1,571 ns | 636,740 | 3.01x slower |
| msgpack | 3,845 ns | 260,058 | 7.37x slower |
| json (stdlib) | 6,358 ns | 157,284 | 12.19x slower |

### 4. Selective Single-Field Mutation

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | vs Baseline |
|---|---:|---:|---:|
| OmniBuf (Generated) | 1,312 ns | 762,090 | 1.00x (baseline) |
| protobuf-py | 4,287 ns | 233,279 | 3.27x slower |
| google.protobuf (upb) | 1,783 ns | 560,776 | 1.36x slower |
| msgspec | 2,394 ns | 417,631 | 1.82x slower |
| msgpack | 6,072 ns | 164,692 | 4.63x slower |
| json (stdlib) | 12,995 ns | 76,952 | 9.90x slower |

### 5. Large Embedded Message Encode

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | Wire Size (B) | vs Baseline |
|---|---:|---:|---:|---:|
| OmniBuf (Generated) | 3,644 ns | 274,456 | 39256 | 1.00x (baseline) |
| OmniBuf (@dataclass) | 3,764 ns | 265,657 | 39256 | 1.03x slower |
| protobuf-py | 4,334 ns | 230,736 | 39227 | 1.19x slower |
| google.protobuf (upb) | 2,829 ns | 353,539 | 39227 | 1.29x faster |
| msgspec | 2,263 ns | 441,913 | 39516 | 1.61x faster |
| msgpack | 3,973 ns | 251,668 | 39516 | 1.09x slower |
| json (stdlib) | 87,019 ns | 11,492 | 39752 | 23.88x slower |

### 6. Embedded Message Growth

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | vs Baseline |
|---|---:|---:|---:|
| OmniBuf (Generated) | 4,606 ns | 217,090 | 1.00x (baseline) |
| protobuf-py | 19,046 ns | 52,504 | 4.13x slower |
| google.protobuf (upb) | 6,120 ns | 163,391 | 1.33x slower |
| msgspec | 9,301 ns | 107,517 | 2.02x slower |
| msgpack | 24,959 ns | 40,065 | 5.42x slower |
| json (stdlib) | 84,737 ns | 11,801 | 18.40x slower |
