Transparency

Benchmark databank

The measured multi-language chain of sc-neurocore 3.15.35, exactly as the platform published it: every measured cell is traceable to a source artifact digest, the whole databank is sealed, and the seal is recorded in the transparency log. Where the measurements contradict the declared dispatch order, the contradiction is shown — not hidden.

Studio sc-neurocore 3.15.35
Measured 65 of 70 cells
Backends 6 across 8 capabilities
Transparency entry 5 of 5 in the log

Publication seal

The databank bytes served at /benchmark/databank.json are sealed with the Hub's hybrid signature and recorded as entry 5 of the transparency log. The workbench re-verifies the seal and the log inclusion in the browser on every load.

Content digest
sha256:d10ec791581bd1ae419eecbb5f2822ec5436e5a40a89e2ca055791a2569059cb
Signature
Ed25519+ML-DSA-65
Key
sha256:8e59ea1452e127dd32e886b3e634cb68064f9c326c9f74682eb81389901d92bc

Where the measurements disagree with the declaration

The studio declares a dispatch order for its backends; the databank measures it. 8 of 8 capabilities measure an order that contradicts the declaration, and in some the undeclared Python reference outruns the first-dispatched port. These are findings, published as measured:

  • mixed_dense_kernel (median_call_ms): declared rust → mojo → julia → go, measured julia → mojo → rust → go.
    • rust is declared before mojo, but mojo measured faster (2.756443 vs 2.989141 ms).
    • rust is declared before julia, but julia measured faster (2.152609 vs 2.989141 ms).
    • mojo is declared before julia, but julia measured faster (2.152609 vs 2.756443 ms).
  • evo_substrate (genomic_distance_ns_per_call): declared rust → mojo → julia → go, measured mojo → julia → go → rust.
    • rust is declared before mojo, but mojo measured faster (18.77696 vs 257.5644900207408 ns).
    • rust is declared before julia, but julia measured faster (22.5 vs 257.5644900207408 ns).
    • rust is declared before go, but go measured faster (22.80301 vs 257.5644900207408 ns).
  • adc_to_spike_kernel (median_call_ms): declared rust → mojo → julia → go, measured mojo → rust → julia → go.
    • rust is declared before mojo, but mojo measured faster (0.947479 vs 1.177833 ms).
  • dcls_tent_kernel (median_call_ms): declared rust → mojo → julia → go, measured rust → julia → mojo → go.
    • mojo is declared before julia, but julia measured faster (1.390021 vs 1.53202 ms).
  • dimensionality (median_call_ms): declared rust → mojo → julia → go, measured rust → mojo → go → julia.
    • julia is declared before go, but go measured faster (84.96713 vs 593.578842 ms).
  • gpfa (median_call_ms): declared rust → mojo → julia → go, measured rust → mojo → go → julia.
    • julia is declared before go, but go measured faster (5.649015 vs 2180.965885 ms).
  • phi (median_call_ms): declared rust → mojo → julia → go, measured rust → julia → mojo → go.
    • mojo is declared before julia, but julia measured faster (0.511882 vs 0.62901 ms).
  • sorting_quality (median_call_ms): declared rust → mojo → julia → go, measured mojo → go → rust → julia.
    • rust is declared before mojo, but mojo measured faster (0.059804 vs 0.128726 ms).
    • rust is declared before go, but go measured faster (0.090446 vs 0.128726 ms).
    • julia is declared before go, but go measured faster (0.090446 vs 0.400581 ms).
  • mixed_dense_kernel (median_call_ms): python carries no dispatch rank yet measured faster than the first-dispatched rust ( 2.325683 vs 2.989141 ms).
  • dimensionality (median_call_ms): python carries no dispatch rank yet measured faster than the first-dispatched rust ( 6.280014 vs 12.022684 ms).

5 cells remain honestly unmeasured ( numpy) — they are listed as “not measured”, never estimated.

Measurements by capability

mixed_dense_kernel

Metric rustmojojuliagonumpypython
median_call_ms (ms) 2.989141 2.756443 2.152609 4.655122 not measured 2.325683
mac_per_s (mac/s) 1403180415.5 1521636403.7 1948474672.4 901008383.3 not measured 1803471912.9

Measured order (median_call_ms): julia → python → mojo → rust → go — contradicts the declared dispatch order (see findings above).

Source artifact digest: sha256:cf7533a298cc678069ebafb664dc479f0192c58ad946cf6e9f98f3c0db2d0a97

evo_substrate

Metric rustmojojuliagonumpypython
genomic_distance_ns_per_call (ns) 257.5644900207408 18.77696 22.5 22.80301 not measured 5992.319560027681
crossover_uniform_ns_per_call (ns) 481.35791992535815 150.50048 45.8 42.1763 not measured 1093.8634499325417
point_mutation_ns_per_call (ns) 432.43831998552196 151.20289 295.4 46.74153 not measured 3984.3632400152273

Measured order (genomic_distance_ns_per_call): mojo → julia → go → rust → python — contradicts the declared dispatch order (see findings above).

Source artifact digest: sha256:dd49db2c94298253e4d15f65491cec9cb017fe9f8266575fa0763699c5078c17

adc_to_spike_kernel

Metric rustmojojuliagopython
median_call_ms (ms) 1.177833 0.947479 1.758218 5.692024 618.964942
samples_per_s (samples/s) 445129336.9 553350498.1 298192841.5 92109238.9 847039.9

Measured order (median_call_ms): mojo → rust → julia → go → python — contradicts the declared dispatch order (see findings above).

Source artifact digest: sha256:49eac5a1cdd94ac218c4e3bc25731e35408f8749cc3cf75fd71f842dc10d502e

dcls_tent_kernel

Metric rustmojojuliagopython
median_call_ms (ms) 1.312861 1.53202 1.390021 1.647302 112.135801
elements_per_s (elements/s) 199673826.5 171110032.4 188589948.2 159135355.6 2337736.9

Measured order (median_call_ms): rust → julia → mojo → go → python — contradicts the declared dispatch order (see findings above).

Source artifact digest: sha256:2ed3a8c2b1ae51a5dd315780addd86c685c37514b568afab7b73384ae8520e29

dimensionality

Metric rustmojojuliagopython
median_call_ms (ms) 12.022684 54.576435 593.578842 84.96713 6.280014

Measured order (median_call_ms): python → rust → mojo → go → julia — contradicts the declared dispatch order (see findings above).

Source artifact digest: sha256:4692ab690d33d3d6ce4885922c840bfbee9fdf2262678e92082fc3e0f1acc222

gpfa

Metric rustmojojuliagopython
median_call_ms (ms) 2.014656 3.326446 2180.965885 5.649015 2.884177

Measured order (median_call_ms): rust → python → mojo → go → julia — contradicts the declared dispatch order (see findings above).

Source artifact digest: sha256:80d119a6b51a3481818e335dcd7478bd4039737fb1b78e752342bbd1e0e58bb8

phi

Metric rustmojojuliagopython
median_call_ms (ms) 0.319231 0.62901 0.511882 0.732037 2.229384

Measured order (median_call_ms): rust → julia → mojo → go → python — contradicts the declared dispatch order (see findings above).

Source artifact digest: sha256:3089a59f6b6646a2e0fc1fe70e0e8f61981b244ab43ecfa635f7c06e5ba24753

sorting_quality

Metric rustmojojuliagopython
median_call_ms (ms) 0.128726 0.059804 0.400581 0.090446 0.216364

Measured order (median_call_ms): mojo → go → rust → python → julia — contradicts the declared dispatch order (see findings above).

Source artifact digest: sha256:740f403cf1b138d19b6e9b81dd6badf3c01977beadebd35948227d8b92a44ec3

How the databank is used

The platform aggregates each studio's real benchmark artifacts into a databank, seals it with the Hub's publication seal, and appends the seal digest to the transparency log. A measured cell must cite the digest of the artifact it came from; an unmeasured cell must stay empty — fabricating either fails the contract. When a studio declares a dispatch order for its backends, the databank reconciles that declaration against the measured values, and a contradiction is published as a finding.

Environment: cpu 11th Gen Intel(R) Core(TM) i5-11600K @ 3.90GHz · platform Linux-6.17.0-35-generic-x86_64-with-glibc2.39 · python 3.12.3 · numpy 2.2.6