Performance
Preuve à l’échelle.
Benchmarks with discipline.
BOUNDA keeps the RC16 performance core and adds clearer CTO/CFO/Board interpretation. Benchmarks internes must be validated on client-specific read-only scope.
Runtime & proof engine.
50.47M/s
Policy/action boundary evaluations.
<11µs p99
Decision-path latency compatible with pre-effect gates.
61.5×
Batch replay throughput for historical audit.
38×
Evidence compression vs raw JSON.
11.3M/s
Éléments de preuve vérifiés en externe.
CPU-only
On-prem, sovereign and air-gapped deployment.
0 GPU
No dependency on expensive accelerators.
100%
Zero-side-effect harness in internal tests.
Benchmarks internes. BOUNDA measures client value; it does not guarantee ROI.
Synthetic 100K multi-surface detection run.
| Surface | Leak detected | Volume | Severity | Gate placement |
|---|---|---|---|---|
| Payment / provider | Objects after decline | 412 | P0 | Before provider.create() |
| Queue / jobs | Jobs after timeout | 298 | P1 | Freeze outbox |
| AI Agent | Tool calls after reject | 287 | P1 | Before tool.dispatch() |
| Physical | Stale physical command | 188 | P0 | Fail-closed on stale |
| DB / State | Mutations after cancel | 156 | P1 | Gate before commit |
| Data Export | Signed URL after deny | 134 | P0 | Before URL create |
| IAM | Credentials after deny | 89 | P0 | Before key.issue() |
1,847 leaks
Total detected over 100K events.
6.88%
Leak rate in synthetic run.
93.12%
Zero-delta PASS.
Finance readiness.
| Axis | Result | Reading |
|---|---|---|
| PaySim transactions | 6,362,620 | Massive payment boundary test. |
| PaySim recall global | 99.428% | Strong risk capture capability. |
| PaySim friction global | 0.213% | Low HOLD/REJECT rate on volume. |
| PaySim split final | 99.694% recall · 96.74% precision | Robust chronological split. |
| BAF regulated | 92.02% recall | Account-opening / KYC case. |
| BAF max safety | 94.82% recall · 74 false releases | Safety-oriented option. |