87.5% observed reduction in human-visible alerts.
PROOF & METHODOLOGY
Observed evidence, with the limitations left intact.
Last updated: 10 September 2026. This page separates observed validation evidence from claims that have not been established.
All six critical events were surfaced or grouped; 0 observed misses.
Observed p50 decision latency in the tested runtime validation.
Additional built-in validation scenarios
The following results come from ADCON's deterministic built-in validation suite. They are product-validation scenarios, not customer case studies and not independent third-party certification.
| Scenario | Raw | Human-visible | Critical | Observed misses | Observed reduction |
|---|---|---|---|---|---|
| threshold_chatter | 107 | 0 | 0 | 0 | 100.0% |
| warmup_critical | 5 | 4 | 5 | 0 | 20.0% |
| critical_cluster | 17 | 2 | 17 | 0 | 88.2% |
| recovery_tail | 46 | 4 | 16 | 0 | 91.3% |
| mixed | 50 | 11 | 44 | 0 | 78.0% |
| vendor_format_replay_equivalent | 32 | 4 | 13 | 0 | 87.5% |
Repeatability testing reproduced the built-in scenario counts across repeated runs, and the custom boundary test passed in the tested validation run. These results expand product-validation coverage without presenting synthetic scenarios as customer outcomes.
Public replay methodology
The published replay contained 400 raw alert events. ADCON decision control produced 50 human-visible alerts, corresponding to an observed reduction of 87.5%. The replay contained no critical events, so it is not used to claim critical-event preservation.
Critical-safety evidence is separate
A separate safety validation pack contained six critical events. Six of six were surfaced or grouped and zero were observed as missed. This evidence is reported separately to avoid combining unrelated test populations.
Runtime observation
The tested build observed a p50 decision latency of 14.6 ms. This is a measurement from the validation environment, not a universal service-level guarantee. Windows and Linux validation suites passed for the tested build.
What this evidence does not prove
- It does not establish a universal alert-reduction percentage for every workload.
- It does not establish a universal latency SLA.
- It does not replace customer-specific shadow-mode or replay validation.
- It does not constitute an independent security certification.
Recommended evaluation path: start with a replay of sample or approved customer data, inspect reduction and safety behavior, then move to shadow mode before production routing.