diff --git a/benchmarks/baseline.json b/benchmarks/baseline.json
index 4cdf98f23b..5479f0ca67 100644
--- a/benchmarks/baseline.json
+++ b/benchmarks/baseline.json
@@ -1,199 +1,465 @@
 {
-  "commit": "53f0f29",
-  "generated_at": "2026-06-09T21:44:30Z",
-  "notes": "Refreshed 2026-06-09 from workflow_dispatch run 27236538959 (macos-14, main @ 53f0f29bb) \u2014 the first run after #4857 un-vacuumed the gate (current.json had never been written since the comparison was added; see compare.sh --json-out fix). Numbers are the median of 3 on one runner; macos-14 runner-to-runner speed variance was observed at +10-35% on CPU-bound rows between same-week runs, so treat single-row speed flags near the 20% threshold as suspect before chasing a code regression \u2014 the perry/node ratio column in output.txt normalizes runner speed and is the better cross-run signal. This run measured on the slower end (e.g. 02_loop_overhead 111ms vs 99-101ms on June 4/5 runs), which makes it a conservative gate baseline. Strip the correctness payloads when hand-editing; the gate only reads perry_ms/perry_rss_kb.",
+  "commit": "98c88a2ec",
+  "generated_at": "2026-06-13T09:40:52Z",
+  "notes": "Refreshed 2026-06-13 on commit 98c88a2ec via `./benchmarks/compare.sh --update-baseline --full --runs 5` (macos-14-class host, Node 24, median of 5). The prior baseline (53f0f29, 2026-06-09) had gone badly stale: it recorded 100x-3000x regressions on the array/numeric rows (e.g. 03_array_write 4143ms, 04_array_read 4067ms, 16_matrix_multiply 10221ms, bench_numeric_array_downgrade 21902ms) that no longer reproduce after the intervening array/codegen work - most of those rows are now at or below Node. The numbers here are the honest current state. Rows where Perry still trails Node materially and are worth real investigation: 09_method_calls (~490x - monomorphic method dispatch), bench_numeric_array_downgrade (~780x - dynamic ops on a mixed any[] array), bench_object_property (~70x), bench_numeric_array_numeric (~50x), 16_matrix_multiply / 10_nested_loops / 04_array_read (~15-17x - inner loops not vectorized to V8's level). Note macos runner-to-runner speed variance of +10-35% on CPU-bound rows: treat single-row flags near the 20% gate threshold as suspect and prefer the perry/node ratio column as the cross-run signal. Strip correctness payloads when hand-editing; the gate only reads perry_ms/perry_rss_kb.",
   "benchmarks": {
     "02_loop_overhead": {
-      "perry_ms": 111,
-      "perry_rss_kb": 3665,
-      "node_ms": 91,
-      "node_rss_kb": 17716,
-      "speed_ratio": 1.22,
-      "memory_ratio": 0.207
+      "perry_ms": 97,
+      "perry_rss_kb": 4609,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "sum:100000000"
+        ],
+        "expected_lines": [
+          "sum:100000000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 54,
+      "node_rss_kb": 33332,
+      "speed_ratio": 1.796,
+      "memory_ratio": 0.138
     },
     "03_array_write": {
-      "perry_ms": 4143,
-      "perry_rss_kb": 245601,
+      "perry_ms": 3,
+      "perry_rss_kb": 92993,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:9999999"
+        ],
+        "expected_lines": [
+          "checksum:9999999"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
       "node_ms": 8,
-      "node_rss_kb": 326885,
-      "speed_ratio": 517.875,
-      "memory_ratio": 0.751
+      "node_rss_kb": 361678,
+      "speed_ratio": 0.375,
+      "memory_ratio": 0.257
     },
     "04_array_read": {
-      "perry_ms": 4067,
-      "perry_rss_kb": 245585,
-      "node_ms": 13,
-      "node_rss_kb": 327621,
-      "speed_ratio": 312.846,
-      "memory_ratio": 0.75
+      "perry_ms": 167,
+      "perry_rss_kb": 93009,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "sum:49999995000000"
+        ],
+        "expected_lines": [
+          "sum:49999995000000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 11,
+      "node_rss_kb": 362173,
+      "speed_ratio": 15.182,
+      "memory_ratio": 0.257
     },
     "05_fibonacci": {
-      "perry_ms": 377,
-      "perry_rss_kb": 3745,
-      "node_ms": 1236,
-      "node_rss_kb": 17684,
-      "speed_ratio": 0.305,
-      "memory_ratio": 0.212
+      "perry_ms": 317,
+      "perry_rss_kb": 4609,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "fib(40):102334155"
+        ],
+        "expected_lines": [
+          "fib(40):102334155"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 1010,
+      "node_rss_kb": 33188,
+      "speed_ratio": 0.314,
+      "memory_ratio": 0.139
     },
     "06_math_intensive": {
-      "perry_ms": 55,
-      "perry_rss_kb": 3745,
-      "node_ms": 54,
-      "node_rss_kb": 18741,
-      "speed_ratio": 1.019,
-      "memory_ratio": 0.2
+      "perry_ms": 51,
+      "perry_rss_kb": 4593,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "result:19.30474921829397"
+        ],
+        "expected_lines": [
+          "result:19.30474921829397"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 51,
+      "node_rss_kb": 33989,
+      "speed_ratio": 1.0,
+      "memory_ratio": 0.135
     },
     "07_object_create": {
       "perry_ms": 2,
-      "perry_rss_kb": 3697,
-      "node_ms": 6,
-      "node_rss_kb": 19493,
-      "speed_ratio": 0.333,
-      "memory_ratio": 0.19
+      "perry_rss_kb": 4657,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "sum:1000000000000"
+        ],
+        "expected_lines": [
+          "sum:1000000000000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 5,
+      "node_rss_kb": 35589,
+      "speed_ratio": 0.4,
+      "memory_ratio": 0.131
     },
     "08_string_concat": {
       "perry_ms": 3,
-      "perry_rss_kb": 3937,
+      "perry_rss_kb": 4897,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "length:100000"
+        ],
+        "expected_lines": [
+          "length:100000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
       "node_ms": 5,
-      "node_rss_kb": 22470,
+      "node_rss_kb": 41209,
       "speed_ratio": 0.6,
-      "memory_ratio": 0.175
+      "memory_ratio": 0.119
     },
     "09_method_calls": {
-      "perry_ms": 13368,
-      "perry_rss_kb": 3777,
+      "perry_ms": 5408,
+      "perry_rss_kb": 6449,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "value:10000000"
+        ],
+        "expected_lines": [
+          "value:10000000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
       "node_ms": 11,
-      "node_rss_kb": 17716,
-      "speed_ratio": 1215.273,
-      "memory_ratio": 0.213
+      "node_rss_kb": 33653,
+      "speed_ratio": 491.636,
+      "memory_ratio": 0.192
     },
     "10_nested_loops": {
-      "perry_ms": 4579,
-      "perry_rss_kb": 5425,
-      "node_ms": 24,
-      "node_rss_kb": 18949,
-      "speed_ratio": 190.792,
-      "memory_ratio": 0.286
+      "perry_ms": 306,
+      "perry_rss_kb": 6433,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "sum:26991000000"
+        ],
+        "expected_lines": [
+          "sum:26991000000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 19,
+      "node_rss_kb": 34005,
+      "speed_ratio": 16.105,
+      "memory_ratio": 0.189
     },
     "11_prime_sieve": {
-      "perry_ms": 658,
-      "perry_rss_kb": 23521,
-      "node_ms": 7,
-      "node_rss_kb": 50893,
-      "speed_ratio": 94.0,
-      "memory_ratio": 0.462
+      "perry_ms": 47,
+      "perry_rss_kb": 24833,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "primes:78498"
+        ],
+        "expected_lines": [
+          "primes:78498"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 10,
+      "node_rss_kb": 70629,
+      "speed_ratio": 4.7,
+      "memory_ratio": 0.352
     },
     "12_binary_trees": {
       "perry_ms": 3,
-      "perry_rss_kb": 3681,
-      "node_ms": 9,
-      "node_rss_kb": 19605,
-      "speed_ratio": 0.333,
-      "memory_ratio": 0.188
+      "perry_rss_kb": 4673,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "sum:1500001500000"
+        ],
+        "expected_lines": [
+          "sum:1500001500000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 6,
+      "node_rss_kb": 35717,
+      "speed_ratio": 0.5,
+      "memory_ratio": 0.131
     },
     "13_factorial": {
-      "perry_ms": 106,
-      "perry_rss_kb": 3745,
-      "node_ms": 134,
-      "node_rss_kb": 18149,
-      "speed_ratio": 0.791,
-      "memory_ratio": 0.206
+      "perry_ms": 98,
+      "perry_rss_kb": 4592,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "sum:49950000000"
+        ],
+        "expected_lines": [
+          "sum:49950000000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 101,
+      "node_rss_kb": 34613,
+      "speed_ratio": 0.97,
+      "memory_ratio": 0.133
     },
     "14_closure": {
-      "perry_ms": 60,
-      "perry_rss_kb": 3761,
-      "node_ms": 64,
-      "node_rss_kb": 21365,
-      "speed_ratio": 0.938,
-      "memory_ratio": 0.176
+      "perry_ms": 49,
+      "perry_rss_kb": 4609,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "sum:2500000000000000"
+        ],
+        "expected_lines": [
+          "sum:2500000000000000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 52,
+      "node_rss_kb": 34549,
+      "speed_ratio": 0.942,
+      "memory_ratio": 0.133
     },
     "15_mandelbrot": {
-      "perry_ms": 24,
-      "perry_rss_kb": 3633,
-      "node_ms": 30,
-      "node_rss_kb": 18821,
-      "speed_ratio": 0.8,
-      "memory_ratio": 0.193
+      "perry_ms": 23,
+      "perry_rss_kb": 4609,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "total_iter:8011148"
+        ],
+        "expected_lines": [
+          "total_iter:8011148"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 25,
+      "node_rss_kb": 34293,
+      "speed_ratio": 0.92,
+      "memory_ratio": 0.134
     },
     "16_matrix_multiply": {
-      "perry_ms": 10221,
-      "perry_rss_kb": 7265,
-      "node_ms": 48,
-      "node_rss_kb": 23671,
-      "speed_ratio": 212.938,
-      "memory_ratio": 0.307
+      "perry_ms": 671,
+      "perry_rss_kb": 8273,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:41079519680"
+        ],
+        "expected_lines": [
+          "checksum:41079519680"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 39,
+      "node_rss_kb": 39639,
+      "speed_ratio": 17.205,
+      "memory_ratio": 0.209
     },
     "bench_gc_pressure": {
-      "perry_ms": 92,
-      "perry_rss_kb": 23393,
-      "node_ms": 25,
-      "node_rss_kb": 27000,
-      "speed_ratio": 3.68,
-      "memory_ratio": 0.866
+      "perry_ms": 70,
+      "perry_rss_kb": 24305,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:249999500000"
+        ],
+        "expected_lines": [
+          "checksum:249999500000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 19,
+      "node_rss_kb": 44056,
+      "speed_ratio": 3.684,
+      "memory_ratio": 0.552
     },
     "bench_json_roundtrip": {
-      "perry_ms": 626,
-      "perry_rss_kb": 319346,
-      "node_ms": 568,
-      "node_rss_kb": 77930,
-      "speed_ratio": 1.102,
-      "memory_ratio": 4.098
+      "perry_ms": 432,
+      "perry_rss_kb": 320033,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:53735550"
+        ],
+        "expected_lines": [
+          "checksum:53735550"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 454,
+      "node_rss_kb": 129496,
+      "speed_ratio": 0.952,
+      "memory_ratio": 2.471
     },
     "bench_object_property": {
-      "perry_ms": 1120,
-      "perry_rss_kb": 89713,
-      "node_ms": 17,
-      "node_rss_kb": 19029,
-      "speed_ratio": 65.882,
-      "memory_ratio": 4.715
+      "perry_ms": 1127,
+      "perry_rss_kb": 90785,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:1999990000"
+        ],
+        "expected_lines": [
+          "checksum:1999990000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 16,
+      "node_rss_kb": 35173,
+      "speed_ratio": 70.438,
+      "memory_ratio": 2.581
     },
     "bench_int_arithmetic": {
-      "perry_ms": 561,
-      "perry_rss_kb": 3713,
-      "node_ms": 104,
-      "node_rss_kb": 17892,
-      "speed_ratio": 5.394,
-      "memory_ratio": 0.208
+      "perry_ms": 596,
+      "perry_rss_kb": 4657,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:5760000"
+        ],
+        "expected_lines": [
+          "checksum:5760000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 71,
+      "node_rss_kb": 33620,
+      "speed_ratio": 8.394,
+      "memory_ratio": 0.139
     },
     "bench_buffer_readwrite": {
-      "perry_ms": 970,
-      "perry_rss_kb": 4721,
-      "node_ms": 103,
-      "node_rss_kb": 18180,
-      "speed_ratio": 9.417,
-      "memory_ratio": 0.26
+      "perry_ms": 933,
+      "perry_rss_kb": 5665,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:12749385600"
+        ],
+        "expected_lines": [
+          "checksum:12749385600"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 88,
+      "node_rss_kb": 33460,
+      "speed_ratio": 10.602,
+      "memory_ratio": 0.169
     },
     "bench_array_grow": {
-      "perry_ms": 501,
-      "perry_rss_kb": 40305,
-      "node_ms": 14,
-      "node_rss_kb": 84724,
-      "speed_ratio": 35.786,
-      "memory_ratio": 0.476
+      "perry_ms": 141,
+      "perry_rss_kb": 41537,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "length:2000000",
+          "checksum:2998500000"
+        ],
+        "expected_lines": [
+          "length:2000000",
+          "checksum:2998500000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 16,
+      "node_rss_kb": 96916,
+      "speed_ratio": 8.812,
+      "memory_ratio": 0.429
     },
     "bench_string_heavy": {
       "perry_ms": 60,
-      "perry_rss_kb": 59569,
-      "node_ms": 50,
-      "node_rss_kb": 20998,
-      "speed_ratio": 1.2,
-      "memory_ratio": 2.837
+      "perry_rss_kb": 60497,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:21063000"
+        ],
+        "expected_lines": [
+          "checksum:21063000"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 48,
+      "node_rss_kb": 37478,
+      "speed_ratio": 1.25,
+      "memory_ratio": 1.614
     },
     "bench_numeric_array_numeric": {
-      "perry_ms": 3546,
-      "perry_rss_kb": 22993,
-      "node_ms": 6,
-      "node_rss_kb": 40286,
-      "speed_ratio": 591.0,
-      "memory_ratio": 0.571
+      "perry_ms": 248,
+      "perry_rss_kb": 25041,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:6500625"
+        ],
+        "expected_lines": [
+          "checksum:6500625"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 5,
+      "node_rss_kb": 55854,
+      "speed_ratio": 49.6,
+      "memory_ratio": 0.448
     },
     "bench_numeric_array_downgrade": {
-      "perry_ms": 21902,
-      "perry_rss_kb": 21425,
-      "node_ms": 7,
-      "node_rss_kb": 40766,
-      "speed_ratio": 3128.857,
-      "memory_ratio": 0.526
+      "perry_ms": 4690,
+      "perry_rss_kb": 24705,
+      "correctness": {
+        "status": "pass",
+        "reference": "node",
+        "actual_lines": [
+          "checksum:6500875"
+        ],
+        "expected_lines": [
+          "checksum:6500875"
+        ],
+        "reason": "all 5 Perry sample(s) matched Node semantic output"
+      },
+      "node_ms": 6,
+      "node_rss_kb": 55774,
+      "speed_ratio": 781.667,
+      "memory_ratio": 0.443
     }
   }
 }