Communications trials on an instrumented range.

The mission

An evaluation laboratory is asked the question every programme eventually asks: how does this communications architecture actually behave when links degrade — and can you show the same behaviour twice? A trial aircraft flies a scripted profile over an instrumented range; the ground station carries two or three radio links of different types to it; and a degradation schedule — cuts, added delay, rising loss — is applied to those links at known times. The deliverable is not a demonstration that ends in applause. It is a dataset: what each link did, what the system decided, and what the traffic experienced, at a resolution fine enough to compare against the next configuration, the next antenna, the next vendor.

What breaks

Conventionally, the measurement is a second project bolted onto the first. Packet capture boxes at both ends, GPS-disciplined clocks to align them, spectrum gear, and a week of post-processing to reconstruct what happened — and even then the reconstruction shows what the packets did, not why the system did it. The decision layer is a black box: when traffic moved from one link to another, the trial can see the effect but must guess at the cause, which makes root-causing a failure an argument instead of a lookup.

Reproducibility suffers most. Two trial days differ in weather, geometry and traffic; without a machine-readable record of both the stimulus and the response, “it performed better on Tuesday” is as precise as the conclusion gets. Procurement decisions deserve harder evidence than that.

The architecture on this mission

On this mission the system under test and the measurement system are the same software. Every Atlas node probes every link roughly four times a second — round-trip time, loss, jitter — as part of carrying traffic at all, so the trial’s finest-grained sensor is already installed on every platform, at both ends of every link, with no extra apparatus and no clock-alignment project. The event journal records the stimulus side as the links experienced it: when the scheduled cut landed, how the delay ramp looked from inside, when the link died and when it returned.

The decision log opens the black box. When the scheduler reweights traffic away from a degrading link — typically within half to three-quarters of a second of the measured change — the log records what it saw and what it chose, timestamped against the same clock as the link records. Cause and effect sit in one dataset: the analyst can trace a specific packet loss spike to a specific probe result to a specific rerouting decision, and disagreements about “why did it switch?” become lookups. Behaviours that are otherwise hard to assess — adaptive redundancy escalating from one to two to three simultaneous paths as conditions worsen, or the redundancy arithmetic itself, where three independent links each losing 2% of packets combine to an effective 0.0008% — become measurable, repeatable exhibits rather than slideware.

Reproducibility falls out of the same design. A scene is a scripted degradation schedule plus a flight profile; because both the stimulus and the response are recorded machine-readably, the same scene replays across configurations, carriers, antennas and software versions, and the comparison is run on identical terms. This is the format Nexus Atlas itself was field-validated in — controlled link-degradation trials on an 8-node mixed fleet at Hemus 2026, links cut, delayed and degraded live while video and control traffic kept flowing.

How the trial day unfolds

1. Scene definition. The evaluators fix the day’s matrix: flight profile, link set, and a degradation schedule per scene — which link is cut, delayed or degraded, when, and by how much. The schedule is a file, and the file is part of the record.
2. Baseline pass. The aircraft flies the profile with no degradation applied. The journal captures the range’s honest baseline — every link’s round-trip, loss and jitter along the whole geometry — against which every later scene is read.
3. The scenes. Pass after pass, the schedule runs: radio A cut at the turn point, delay ramping on radio B down the return leg. Traffic reweights within fractions of a second of each measured change, and every probe and every decision lands in the log as it happens.
4. The hard scene. All links degraded at once, loss climbing everywhere. Adaptive redundancy escalates to multiple simultaneous paths; the class system sheds bulk while command holds its floor. The behaviour is not observed from outside — it is recorded from inside, decision by decision.
5. Same-day replay. A configuration change — a different antenna, a different weighting — and the identical scene runs again. Two datasets, one stimulus, comparable line by line before the aircraft is packed.
6. The deliverable. The trial report is generated from the journals and decision logs: per-scene, per-link, per-decision, timestamped on a common clock. The next laboratory can rerun the same scenes and check the numbers.

What each mechanism contributes

• Continuous per-link measurement — round-trip, loss and jitter probed ~4× per second on every link, at both ends, as a side effect of operating at all. Shipped.
• The event journal — the stimulus as the system experienced it: cuts, ramps, recoveries, timestamped. Shipped.
• The scheduler’s decision log — what was seen and what was chosen, on the same clock as the link records; the black box, opened. Shipped.
• The behaviours under test — sub-second reweighting, adaptive redundancy, class floors and ordered degradation — themselves shipped product, so the trial measures the same software a deployment would run. Classes are opt-in.

The honest boundary: Atlas measures the links it operates — it is not a spectrum analyser, and RF-level instrumentation of the environment, the degradation apparatus itself, range safety and the experimental design remain the test centre’s. What it removes is the gap between the system under test and the evidence about it: the claims and the record come from the same clock.

What a pilot should prove

• Instrumentation parity: the built-in journal against the range’s independent reference measurement on the same scene, link by link.
• Reproducibility: the same scene run twice, with the recorded response matching within stated tolerances.
• Traceability: one induced failure traced end to end — probe, decision, reroute, traffic effect — entirely from the logs.
• A comparison deliverable: two configurations flown against the identical schedule, reported from the journals alone.

One range, one aircraft, one degradation matrix, two days. The evaluation format covers the structure.