Validation

Evidence, by tier

Every claim on this page is labeled by what kind of evidence it is — real telemetry, public-dataset characterization, simulation-supported research, or not yet available — so nothing here overstates what's been measured.

Real telemetry, measured

A single lab-operated reference link, continuously collected, carrying two recorded incidents end to end.

Two-panel chart of real telemetry from 2026-06-25: the top panel shows device-reported Link Quality Index dropping sharply at 12:29 and again at 14:04, while the bottom panel shows ping latency and packet loss — what conventional monitoring watches — staying flat and healthy-looking until a latency and loss spike at the 16:54 cascade.
REAL TELEMETRY — MEASURED

Outdoor 5 GHz point-to-point reference link, lab-operated. Continuous SNMP/MIB + ICMP collection, 2026-06-25 — the incident behind the figures below. Link Quality Index (top) degrades hours before ping latency or packet loss (bottom) show anything a threshold-based monitor would flag.

Mandatory limitation

Single link, single incident. Enough to demonstrate the mechanism end to end. Not yet enough to claim detection accuracy — multi-link validation is the pilot objective.

01

4h 25m — Decision Window

First weak signal to cascade escalation.

02

2h 50m — False Calm

Conventional monitoring reported healthy while degradation was underway.

03

2 Recorded Incidents · 144 Collection Cycles · 1 Production-Representative Link

Continuous SNMP/MIB collection from an outdoor 5 GHz point-to-point reference link, lab-operated. The collector is standard SNMP/MIB — the same path reads carrier microwave network elements.

04

Specificity Check

The same unmodified reasoning, run on a deliberate load test on the same link, identifies traffic-induced saturation rather than radio degradation.

Resource envelope

103 MB peak resident memory — 1.4% of an 8 GB edge node.

Measured on an NVIDIA Jetson Orin Nano Developer Kit (8 GB) against a 2,500-row demonstration dataset. Behaviour at operator scale is not measured.


PUBLIC DATASET — CHARACTERIZATION

Third-party datasets, characterized — not validation

PMADS, SNOB5G and SIAE are third-party public research datasets, characterized to sharpen the telemetry request Velorona makes of an operator. No Velorona system has run against them.


SIMULATED — NOT OPERATOR DATA

Simulation supports algorithm development — it is not evidence

Algorithm development is supported by a standards-informed, physics-based simulation environment. For point-to-point microwave this is derived from ITU-R propagation models. It is not calibrated against operational data, and no validated claim rests on it.

See the simulation research on AID Edge Research

OPERATOR PILOT — NOT YET AVAILABLE

Operator pilot — not yet available

No operator-network results yet. A read-only historical replay is how we get them.

Discuss validation criteria for your network