VM PERFORMANCE RECEIPT

This is DBYTE release-to-release regression evidence: v12.2.1 versus clean v12.3.0. It is not a cross-language marketing benchmark.

00 identity

BASELINE COMMIT   3843f7b3cf29951f10fe732754434b41ac53fc7f
CANDIDATE COMMIT  9b4ccfdfec197e607325792785d266cccbe611bf
CANDIDATE DIRTY   false
PACKAGE RELEASE   v12.3.0
PACKAGE SHA256    0ad84b022810080897a323dee7ba103c64c0dedc58f3c9ca251ca712c1823b53
MACHINE CPU       12th Gen Intel(R) Core(TM) i5-12400F
MACHINE OS        Microsoft Windows NT 10.0.26200.0
RUSTC              rustc 1.95.0 (59807616e 2026-04-14)

10 exact command

powershell -NoProfile -ExecutionPolicy Bypass -File .\scripts\verify_vm_performance.ps1

The verifier builds an exact v12.2.1^{} baseline from Git archive, builds the detached v12.3.0 candidate, checks source and output hashes, discards three warmups, records nine measurements and rejects unstable receipts above 5 percent relative MAD.

20 measured result

CASES             15
GEOMETRIC MEAN    1.415x
RELEASE GATE      >= 1.25x
PER-CASE FLOOR    >= 0.90x
RESULT            PASS

Case	v12.2.1 median ms	v12.3 median ms	base MAD	v12.3 MAD	speedup
`binary_read_u32`	251.097	237.645	1.958%	0.910%	1.057x
`buffer_replace`	228.346	183.087	1.399%	2.445%	1.247x
`bytes_find`	174.039	155.826	0.611%	0.437%	1.117x
`bytes_find_single`	119.967	102.885	2.180%	0.948%	1.166x
`function_call`	684.325	367.059	0.931%	1.447%	1.864x
`function_call_chain`	1001.926	428.024	1.506%	1.610%	2.341x
`function_call_int`	843.446	456.775	1.499%	1.482%	1.847x
`function_call_loop_return`	944.245	447.375	1.204%	0.954%	2.111x
`function_call_many_args`	827.238	423.729	0.841%	1.811%	1.952x
`function_call_nested`	592.262	364.765	3.487%	1.307%	1.624x
`int_compare_loop`	505.092	400.810	3.043%	1.125%	1.260x
`loop_sum`	258.213	221.599	1.697%	0.919%	1.165x
`loop_sum_large`	520.417	513.753	0.722%	2.748%	1.013x
`nested_int_loop`	264.579	221.968	1.098%	1.002%	1.192x
`patch_workflow`	201.137	180.677	1.718%	0.685%	1.113x

30 raw samples and source

The receipt contains every baseline and candidate sample, source SHA-256, output SHA-256, median, min, max, MAD and runtime profile. The corpus and runner source are published beside it so the summary above is not the only copy of the evidence.

RECEIPT SHA256  50f36733faa2556810eadc39562881e17945d63a52b7b116faead4b91a3b7007
VERIFIER SHA256 c25cab65fa6504f903b92c267156d435a35f9fdf57af1f181e8fa3796876548c
CORPUS SHA256   ee473abe5771665d9ecf745a95ebdf771bb0eab6e4f34e9cde511989360cf697
RUNNER SHA256   5d706ca5cd65ad7e2e38b6ecbcc9536ce11ba478048cb79007175c5c3c6351b3

RAW RECEIPT JSON · VERIFIER SOURCE · CORPUS MANIFEST · RUNNER SOURCE · BENCHMARK SOURCE

function_call.dbyDOWNLOAD OPEN RAW

fn work(n: int) -> int:
    return n * 2

let i = 0
while i < 1000000:
    work(i)
    i = i + 1

DBY · 104 BSHA-256 516a04883caa1c173e3b2871c6997457576d392e6e48ead564e2533c399fc53b

NO TROPHY CASE. The numbers are useful because the raw receipt is here. Without the receipt they are decoration.