# What is Engine 2

## Four components, one direction

Neurologyca's platform has four parts, and data flows through them in one direction:

| Component | Runs | Does | Does not |
| --- | --- | --- | --- |
| **Capture (@neurologyca/capture)** | **On-device (real time)** (**Preview 0.45**): On the person's device, in the browser | Extracts facial geometry, expression coefficients and head pose as you capture. **Leaves:** landmarks, expression coefficients and pose to Engine; the transcript and extra signals, only if you turn them on. Video and audio never leave the device. | Interpret anything |
| **Capture (@neurologyca/capture)** | **Deferred (server-side)** (**Preview**): Records in the browser; processed by our processing service | Records the video and a step manifest and uploads them; our processing service handles them shortly after, so video and audio leave the device. **Leaves:** video and audio to your receiver and the processing service. | Interpret anything |
| **Engine** | Neurologyca | Runs the measurement protocol, the proprietary models and the calibration | Own the meaning of a result |
| **Konstellation** | Neurologyca | Resolves the case deterministically, grounds it in evidence and writes the narrative | Process signal or recompute measures |
| **Your application** | You | Owns the interface, the identity of your users, storage and workflow; runs your backend half of the SDK (`@neurologyca/capture-server`): the proxy, the token and enrolments | — |

Each arrow is a versioned contract.

The capture SDK itself has two halves: `@neurologyca/capture` runs in the browser and `@neurologyca/capture-server` runs in your backend. See [Install](/docs/start/install/).

Which mode to use, and how to combine the SDK with the HTTP API: [Integration paths](/docs/start/paths/).

## You only talk to Engine

Your backend and your page call **Engine**, and nothing else. Engine calls Konstellation for you when you ask for an interpretation. In deferred, your backend also talks to Neurologyca's processing service, which calls your backend back with a signed webhook ([Processing service](/docs/sdk/processing/)).

## Contexts: what your organization can measure

A **context** is a use case (for example, an athlete's wellness during the competitive week) with its protocols and, usually, a **taxonomy** that tells the interpretation what to look at.

Neurologyca assigns contexts to each organization. `GET /catalog/contexts` and `GET /catalog/protocols` return **your** organization's catalogue, so two organizations see different lists. The organization behind these examples is only an example.

| If the context… | A capture returns | Typical use |
| --- | --- | --- |
| has a taxonomy (`taxonomy.constellation_context_id`) | HCIs **and** an interpretation | Wellness, sport, hiring |
| has none (**Measurement**) | HCIs only, per frame and per step | "Just measure": you own the meaning |

Measurement is universal in the test environment: every organization has its context, and sees it without a taxonomy. The production Engine does not offer it yet ([Your environments](/docs/start/environments/)).

```json
[{ "name": "Measurement", "taxonomy": null }]
```

A context with a taxonomy carries its id (illustrative):

```json
[
  {
    "name": "<your context>",
    "taxonomy": {
      "constellation_context_id": "<your-context-id>"
    }
  }
]
```

Missing a context? The admin of your organization asks Neurologyca for it. You can list yours at any time with the two calls above, or in the [API reference](/docs/reference/).

Neurologyca provides a guide for each of your contexts: it comes from Neurologyca with access to the context. A context guide covers:

- what the context measures and for whom;
- its protocols, positions and cadence;
- its taxonomy: the context, the target indicator and the subcontext of each mechanic;
- its steps and media: who ends each step, and when to send `mediaId`;
- which steps to transcribe, and in which language;
- the audiences available for the interpretation, and the roles of its users.

## Words you will see

| Word | What it is |
| --- | --- |
| Context | A use case your organization has been given; it decides which protocols you can enrol people in |
| Protocol | What is measured and when: an ordered list of positions |
| Position | One slot of the protocol, the capture of one moment; each one runs a mechanic made of steps |
| Enrol | Sign a person up to a protocol (`POST /user-protocols`); the answer's `userProtocolId` names that enrolment |
| The pair | `userProtocolId` and `protocolMechanicId`: one enrolment and the position to capture. Your backend gives it to the page; it is used once |
| Session | One capture of a position, from its first frame to its results; Engine opens one per capture |
| Baseline | A short capture per person that Engine uses as their reference; recommended before any protocol, not required |
| Kit | In [Your integration](/integration), per protocol: its client configuration (`neuro.config.json`), your backend's environment (`.env.example`) and a ready-to-run starter per mode |

More levels and their API names: [Protocols and their hierarchy](/docs/concepts/protocols/).

## What Engine returns

- **Human Context Indicators (HCIs)** for every captured frame, and their mean per step.
- **Live events** while a capture is processed (Server-Sent Events).
- **An interpretation** of a completed capture, written for the audience you choose, when the context has a taxonomy.

## What it does not return

- No aggregate analysis across a whole protocol.
- No webhooks from Engine: you use the live stream or poll for results (in deferred, the processing service does send one).
- No audio reaches Engine: on-device, transcription runs in the browser and only its text is sent; deferred, the audio goes to the processing service, which transcribes it.
- No intent, deception or diagnosis. Indicators describe the interaction, not the person.

## Next

Follow the [Quickstart](/docs/start/quickstart/) to get from your API key to your first indicators, or choose your way first in [Integration paths](/docs/start/paths/).
