PROOF

Qualification at system scale, not a wall of isolated test counters.

SnapSpace evidence spans bounded software, distributed software/SIL, real GPU, physical simulation and autonomous-system environments with repeated qualification, adversarial cases, deterministic replay and sealed evidence.

409,600 / 15represented servers / kernels

Exact software/SIL control-equivalence stress point for two frozen cases. Not physical capacity, a scaling law or a production-throughput claim.

48 / 48 + 9 / 9federation + sealed artifacts

Pinned multi-domain software composition preserved local authority while the recorded provenance set verified cleanly.

32 / 32 ×2authenticated end-to-end

Two independent authenticated runs preserved the same semantic sequence across decision, authority, execution, STOP, reconciliation and receipt.

58 / 58 ×2institutional rule profile

Frozen offline qualification repeated twice; receiver acceptance remains an external decision rather than being invented by the control layer.

40 / 40 → 44 / 44institutional handoff

The synthetic handoff repeated byte-identically, then passed a later STOP-enhanced regression without widening the clinical claim.

20 / 20 ×2autonomy simulator

Closed-loop underwater physical-simulator qualification repeated with the same semantic result; field and hardware claims remain open.

WHY THESE NUMBERS MATTER

The evidence is organised by scale, adversarial depth, environment and reproducibility.

SnapSpace does not manufacture one giant test total by adding inherited regressions and repeated runs together. Each figure keeps its own environment and claim ceiling.

Repeated qualification

Critical suites are rerun independently rather than represented by one successful execution.

Adversarial cases

Stale authority, replay, malformed inputs, changed material, forged STOP, partial effects and lost-response states are tested explicitly where applicable.

Deterministic replay

The two-domain fabric records REPLAY_MATCH plus an independent verifier result of VERIFIED / EVIDENCE_GRAPH_VALID.

Sealed evidence

Canonical convergence records 48/48 PASS plus 9/9 sealed-artifact verification; hashes bind frozen evidence to exact candidate material.

Real accelerator execution

A sealed real two-GPU CUDA/NCCL distributed-training qualification closes part of the gap between pure SIL and actual accelerator execution on one physical host.

Negative evidence retained

Earlier failed or incomplete runs remain visible. Later PASS results do not rewrite the qualification history.