# Automotive Cybersecurity Dashboard - Data Release

## Purpose

This data release provides the structured data behind the Automotive Cybersecurity Dashboard.

The central concept is the **Threat Register**. External evidence is connected to individual threats so that a risk assessment can be examined against observable technical, academic and strategic information.

The data is intended for research, analysis, validation and further exploration. It is not an automated risk decision system and should not replace expert judgement.

> A risk analysis describes what could happen. Connecting threats to observable evidence makes the risk tangible, examinable and open to challenge.

---

## Recommended directory structure

```text
ncrc-data/
├── threats/
│   ├── threat_register.json
│   └── register.json
├── evidence/
│   ├── threat_correlations.json
│   ├── evidence_index.json
│   └── correlation_review.json
├── sources/
│   ├── cves.json
│   ├── papers.json
│   ├── github.json
│   ├── signals.json
│   └── incidents.json
└── README.md
```

The exact filenames may differ from the working environment. The structure above is intended to keep authoritative threat information, source collections and derived correlation output clearly separated.

---

## 1. Threat data

### `threats/threat_register.json`

The base Threat Register. Threat identifiers use the form:

```text
T-001
T-002
...
T-069
```

These identifiers are the primary keys for connecting risk information with the evidence and correlation layers.

Depending on the exported version, threat records may contain information such as title, category, risk classification, domains, affected areas, rationale, tags and matching profiles.

### `threats/register.json`

The enriched Threat Register used by the evidence and dashboard pipeline. It can contain resolved matching information and evidence already associated with a threat, including repository mappings.

When analysing relationships, use the `T-xxx` threat identifier rather than relying on titles, because titles may evolve while the identifier provides a stable relationship key within a release.

---

## 2. Source collections

The source files contain collected evidence before, or independently from, the final threat-correlation decision.

### `sources/cves.json`

Automotive-relevant vulnerability records collected from public vulnerability information. Records can include technical descriptions, severity information, matched terminology and detected attack surfaces.

The collection includes vehicle technology as well as relevant adjacent infrastructure such as EV charging where it intersects with automotive cybersecurity.

### `sources/papers.json`

Academic and research material relevant to automotive cybersecurity. Records can include titles, abstracts or descriptions, publication metadata, topics, matched terminology, source links and CVE references where available.

### `sources/github.json`

Public repositories associated with automotive cybersecurity research, implementations, tools and relevant technical attack surfaces.

Repository presence is evidence of publicly available technical knowledge or capability. It should not automatically be interpreted as evidence that a corresponding attack has occurred in practice.

### `sources/signals.json`

Strategic or emerging signals that may be relevant to the changing threat landscape.

A signal is an observation to be considered in context. It is not automatically evidence that a specific threat has materialised.

### `sources/incidents.json`

Incident and intelligence records used to connect reported or observed real-world developments with threats in the register.

Some releases may also contain authoritative assessments or related manually curated intelligence records in this layer or in separate source files.

---

## 3. Derived evidence and correlations

### `evidence/threat_correlations.json`

The main threat-centred correlation dataset.

This file connects evidence records to individual `T-xxx` threats. A correlation can contain fields describing, among other things:

- evidence type;
- evidence identifier;
- correlation status;
- relevance/correlation score;
- matched terminology or technical features;
- explanation of the relationship;
- source URL where available.

This is usually the most useful file when the analytical question starts with a threat, for example:

> Which external evidence is associated with T-002?

### `evidence/evidence_index.json`

An evidence-centred view of the correlation results.

Where `threat_correlations.json` starts with the Threat Register, the evidence index can be used to approach the same relationships from the evidence side: an item can be inspected together with the threats to which it has been matched.

### `evidence/correlation_review.json`

Review-oriented correlation output, where present in the release. This is intended to support human assessment of candidate relationships rather than silently converting every automated match into an accepted relationship.

---

## 4. Correlation status

The distinction between correlation states is fundamental to correct interpretation of the dataset.

### Accepted

`accepted` indicates that the relationship between an evidence item and a specific threat meets the configured acceptance criteria of the correlation pipeline.

Accepted does **not** mean that the source proves every aspect of the threat, that the threat has materialised, or that the risk level is correct.

### Review

`review` identifies a potentially relevant relationship that should be examined before being treated as accepted evidence for the specific threat.

A review item is not necessarily a weak source. Source quality and correlation status are different concepts.

### Rejected

Some analytical outputs may contain or support rejected matches. Rejected candidates are relationships that did not meet the configured criteria for presentation as accepted or review evidence. They may be omitted from dashboard views depending on the generated output.

---

## 5. Relevance and correlation scores

A `relevance_score` or correlation score describes the strength of an automatically identified relationship according to the configured matching model.

The model may use combinations of information such as:

- lexical overlap;
- technical features;
- assets;
- protocols;
- attack vectors;
- impacts;
- contextual rules or penalties.

A score must **not** be interpreted as:

- probability that an attack will occur;
- probability that an evidence item is true;
- CVSS severity;
- threat risk level;
- proof of causality.

It is a ranking and correlation aid for analysis.

---

## 6. Evidence strength versus correlation strength

These two dimensions should be kept separate.

**Evidence strength** asks questions such as:

> How authoritative, direct or technically meaningful is this source?

**Correlation strength** asks:

> How strongly does this evidence item appear to relate to this particular threat?

An authoritative assessment may therefore still require review for a particular threat. Conversely, a technically close correlation does not automatically make the underlying source authoritative.

---

## 7. Evidence types

The correlation layer can combine different kinds of evidence because each source type answers a different analytical question.

| Evidence type | Analytical role |
|---|---|
| CVE | Concrete publicly documented vulnerability |
| Paper | Academic research, experiments and technical analysis |
| Repository | Public tooling, implementation or security research |
| Signal | Emerging or strategic observation |
| Incident | Reported or observed real-world development |
| Assessment | Authoritative analysis or security/policy assessment |
| Narrative | Synthesis across multiple observations or evidence items |

Not every release or every threat will contain all evidence types.

---

## 8. Cross-domain interpretation

Automotive cybersecurity is not limited to components physically located inside a vehicle.

Relevant relationships can cross boundaries such as:

```text
Vehicle
   <-> Charging infrastructure
   <-> Backend / cloud
   <-> Telecommunications
   <-> Energy infrastructure
   <-> Digital services
```

This is deliberate. An adjacent system can be relevant to an automotive threat even where ownership, regulation, operation or supervision lies in another domain.

EV charging is a useful example: CVEs, academic papers and public repositories can provide evidence about EVSE, OCPP, ISO 15118, Plug & Charge and V2G-related technology while the corresponding risk can span both mobility and energy infrastructure.

---

## 9. Important analytical cautions

### Correlation is not proof

A correlation identifies a relationship worth considering. It does not establish causality and does not demonstrate that the complete threat scenario has occurred.

### Absence of evidence is not evidence of absence

A threat with little matched evidence should not automatically be considered unlikely or unimportant. Public evidence can be incomplete, unevenly distributed or unavailable.

### Volume is not risk

A threat with many papers or CVEs is not automatically more severe than a threat with fewer evidence items. Evidence volume, evidence diversity, impact, likelihood and source quality are different dimensions.

### Public repositories require context

The existence of a tool, proof of concept or implementation does not establish that it has been used maliciously.

### Review candidates require judgement

Large review sets are deliberately retained to support expert assessment and discovery. They should not be counted as confirmed evidence without further review.

---

## 10. Useful ways to analyse the release

The data model supports analyses beyond the dashboard itself. Examples include:

- evidence per threat;
- evidence diversity per threat;
- CVEs by technology, supplier or manufacturer where supported by source metadata;
- development of CVEs, papers or repositories over time where source dates are available;
- protocols or attack surfaces shared by apparently unrelated threats;
- threats with large review backlogs;
- threats with strong research coverage but limited vulnerability evidence;
- threats with concrete vulnerabilities but little academic coverage;
- cross-domain dependencies such as vehicle, charging, backend and energy infrastructure;
- unexpected relationships emerging from shared technologies or evidence.

These are analytical possibilities, not precomputed conclusions. Any quantitative result should be derived from the files in the specific release being analysed.

---

## 11. Traceability and provenance

Where available, source URLs, evidence identifiers, generation timestamps and matching explanations should be retained when redistributing or transforming the data.

This makes it possible to move from:

```text
Threat
  -> Correlation
      -> Evidence item
          -> Original public source
```

Traceability is a core design principle of the dataset.

---

## 12. Versioning

This dataset evolves as:

- new public evidence is collected;
- source records are updated;
- matching and correlation logic is refined;
- review decisions are made;
- the Threat Register itself evolves.

For reproducible analysis, record at least:

- the date of the exported data release;
- the Threat Register version/source used;
- the correlation-output generation timestamp where available;
- the filenames or hashes of the analysed JSON files.

Do not compare counts from different releases without first checking whether collection scope, filtering or correlation logic changed.

---

## 13. Research status

The Automotive Cybersecurity Dashboard and this data release are a research and analytical prototype.

The purpose is to explore how cybersecurity risk analysis can be made more evidence-driven, transparent, traceable and continuously examinable while retaining human judgement in the assessment process.

The data and correlations should therefore be treated as analytical support, not as formal findings, regulatory decisions or automated risk determinations.

---

## Quick start

For most analyses, start with these three layers:

1. `threats/register.json` for the enriched threat definitions;
2. `evidence/threat_correlations.json` for threat-to-evidence relationships;
3. the files under `sources/` when the original evidence metadata is needed.

The stable conceptual join is:

```text
T-xxx threat identifier
      <->
threat correlation
      <->
evidence identifier
      <->
source record / source URL
```

That structure makes it possible to analyse the dataset from either direction: start with a risk and inspect the evidence, or start with an evidence item and investigate which risks it may inform.
