============================================================================
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) | 379 ns | 2,636,864 | 366 | 1.00x (baseline) |
| OmniBuf (@dataclass) | 385 ns | 2,600,173 | 366 | 1.00x (baseline) |
| protobuf-py | 968 ns | 1,033,429 | 341 | 2.55x slower |
| google.protobuf (upb) | 443 ns | 2,258,339 | 341 | 1.17x slower |
| msgspec | 632 ns | 1,581,625 | 634 | 1.67x slower |
| msgpack | 1,674 ns | 597,494 | 634 | 4.41x slower |
| json (stdlib) | 4,876 ns | 205,073 | 874 | 12.86x slower |

### 2. Full Message Decode Speed

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | vs Baseline |
|---|---:|---:|---:|
| OmniBuf (Generated) | 2,036 ns | 491,272 | 1.00x (baseline) |
| OmniBuf (@dataclass) | 2,043 ns | 489,541 | 1.00x (baseline) |
| protobuf-py | 3,158 ns | 316,636 | 1.55x slower |
| google.protobuf (upb) | 4,622 ns | 216,348 | 2.27x slower |
| msgspec | 2,205 ns | 453,615 | 1.08x slower |
| msgpack | 3,991 ns | 250,570 | 1.96x slower |
| json (stdlib) | 6,121 ns | 163,359 | 3.01x slower |

### 3. Read a Single Nested Field

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | vs Baseline |
|---|---:|---:|---:|
| OmniBuf (Generated) | 506 ns | 1,974,478 | 1.00x (baseline) |
| protobuf-py | 2,168 ns | 461,357 | 4.28x slower |
| google.protobuf (upb) | 775 ns | 1,290,370 | 1.53x slower |
| msgspec | 1,171 ns | 853,616 | 2.31x slower |
| msgpack | 2,713 ns | 368,659 | 5.36x slower |
| json (stdlib) | 4,827 ns | 207,180 | 9.53x slower |

### 4. Selective Single-Field Mutation

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | vs Baseline |
|---|---:|---:|---:|
| OmniBuf (Generated) | 821 ns | 1,218,234 | 1.00x (baseline) |
| protobuf-py | 3,227 ns | 309,888 | 3.93x slower |
| google.protobuf (upb) | 1,354 ns | 738,586 | 1.65x slower |
| msgspec | 1,837 ns | 544,510 | 2.24x slower |
| msgpack | 4,472 ns | 223,592 | 5.45x slower |
| json (stdlib) | 9,891 ns | 101,103 | 12.05x slower |

### 5. Large Embedded Message Encode

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | Wire Size (B) | vs Baseline |
|---|---:|---:|---:|---:|
| OmniBuf (Generated) | 2,636 ns | 379,395 | 39256 | 1.00x (baseline) |
| OmniBuf (@dataclass) | 2,563 ns | 390,225 | 39256 | 1.00x (baseline) |
| protobuf-py | 3,371 ns | 296,611 | 39227 | 1.28x slower |
| google.protobuf (upb) | 2,260 ns | 442,429 | 39227 | 1.17x faster |
| msgspec | 1,866 ns | 535,989 | 39516 | 1.41x faster |
| msgpack | 3,177 ns | 314,780 | 39516 | 1.21x slower |
| json (stdlib) | 67,520 ns | 14,810 | 39752 | 25.62x slower |

### 6. Embedded Message Growth

| Codec / Operation | Latency (ns/op) | Throughput (ops/sec) | vs Baseline |
|---|---:|---:|---:|
| OmniBuf (Generated) | 2,486 ns | 402,181 | 1.00x (baseline) |
| protobuf-py | 13,682 ns | 73,089 | 5.50x slower |
| google.protobuf (upb) | 4,458 ns | 224,325 | 1.79x slower |
| msgspec | 7,198 ns | 138,924 | 2.89x slower |
| msgpack | 18,526 ns | 53,978 | 7.45x slower |
| json (stdlib) | 63,944 ns | 15,639 | 25.72x slower |
