Evidence & validation status
What is implemented, what is simulated, what is experimental, and what has not been tested yet. This page is the authoritative status statement for CarpatDome. Where any other page disagrees with it, this page is correct.
Last reviewed .
Capability matrix
| Capability | Evidence today | Next gate |
|---|---|---|
| Remote ID / OpenDroneID parsing | Software validated, automated tests | Outdoor decode against a known airframe |
| Remote ID over Bluetooth Long Range | Software validated on captured packets | Outdoor decode against a known airframe |
| Sub-GHz RF energy scan and classification | Software validated, bench tier | Receiver sensitivity and false-alarm characterisation |
| Acoustic detection and bearing (RF-silent drones) | Synthetic audio only | Outdoor capture against a known airframe; bearing error |
| MQTT transport, versioned JSON contracts | Software validated | Third-party C2 integration demonstration |
| Integration port (partner detections in, cues out) | Software validated, contract tests | Third-party sensor or C2 interoperability demonstration |
| Multi-node fusion and track association | Validated on synthetic scenarios | Association accuracy against ground truth |
| Fused picture (one contact per drone) | Validated on synthetic multi-source scenarios | Association accuracy against ground truth |
| TDoA cross-correlation | Synthetic IQ only | Simultaneous raw IQ capture across physical nodes |
| Multilateration | Synthetic IQ only | Localisation error against ground truth (CEP50 / CEP90) |
| TDoA end-to-end (capture → fix) | Synthetic emitter recovered in software | Real IQ capture across physical nodes |
| Engagement ladder (detect → intercept) | Scripted C2 simulation, no hardware commanded | Not applicable — no effector is in scope |
| Physical receiver node | Components received — assembly not started | First nodes assembled and time-checked |
| Outdoor Remote ID detection | Not tested | Field experiment #001 |
| Non-cooperative RF detection in the field | Not tested | Field characterisation campaign |
| TDoA localisation on real signal | Not tested | Multi-node field validation |
The test suite, precisely
194 automated tests are collected; 192 pass. Two are deselected by default — they carry hardware and integration markers and require a physical radio and a running MQTT broker. They will be enabled by the field phase, not before. We report the denominator because the two excluded tests are exactly the ones a defence reviewer would care about.
Maturity
We assess the software demonstrator at TRL 3 — analytical and experimental proof of concept. TRL 4, component validation in a laboratory environment, is the objective of the first multi-node field campaign. It is not a current claim, and we will not make it before the measurements exist.
Band coverage
The current sub-GHz chain is RTL-SDR based and covers 24 MHz to 1.77 GHz. It does not cover the 2.4/5.8 GHz bands most FPV video and control links use; Remote ID on those bands is decoded in software over a dual-band Wi-Fi adapter and a Bluetooth Long Range receiver; the hardware has been received and is not yet assembled. This is a scope boundary, not an oversight — and it is the limit most likely to matter to anyone evaluating the FPV scenarios in the demonstrator.
What Field Note #001 will publish
The next credibility step is not another feature. It is a measured number. When the first physical nodes run their first controlled trial, we will publish the following — or state plainly that a figure was not achieved:
- Remote ID detection range and packet acquisition rate
- Bands monitored, receiver sensitivity and detection threshold
- Probability of detection and false-alarm rate at a stated threshold
- Track association accuracy and end-to-end latency through fusion
- Timing error between nodes
- Acoustic detection range and bearing error against a known airframe
- TDoA localisation error (CEP50 / CEP90), line-of-sight and non-line-of-sight
A single honest line — median localisation error X m at Y km, first outdoor experiment — is worth more to a consortium lead than another hundred software tests. We will publish the figures we get, including the ones that disappoint.