> ## Documentation Index
> Fetch the complete documentation index at: https://api.thesapientcompany.com/llms.txt
> Use this file to discover all available pages before exploring further.

# How the model works

> The pipeline from a piece of content to a brain response to the four lenses.

A scan moves through a fixed pipeline. Content goes in, a predicted brain response comes out, and that response is reduced step by step into something you can read at a glance. Each step is a faithful summary of the one before it, and nothing is invented along the way.

<Steps>
  <Step title="Content comes in">
    You submit a piece of content by URL: a video or an audio clip. This is the raw material the model will respond to. (Image and raw-text inputs were read by Mary, which is temporarily unavailable; Qualia, the active model, takes video or audio.)
  </Step>

  <Step title="Content becomes features">
    The content is turned into machine-readable features: what's being said, heard, and seen over time. Speech is transcribed, sound is characterized, and motion and imagery are tracked frame by frame. These features are the model's senses.
  </Step>

  <Step title="The model predicts a brain response">
    From those features, the model predicts a full brain response: activity across roughly 20,000 points on the cortex, for every moment of the content. This is the ground-truth measurement everything else is built on.
  </Step>

  <Step title="The brain is reduced to 7 networks per second">
    The cortex-wide response is summarized into **7 well-established brain networks**, one value each, every second. Each network corresponds to a plain-English function: seeing, moving, focusing, alerting, feeling, reasoning, and reflecting.
  </Step>

  <Step title="Networks blend into KPIs">
    The seven per-second networks are blended into a set of **KPIs**: intermediate measures that combine networks into more specific signals.
  </Step>

  <Step title="KPIs roll up into composites">
    The KPIs are grouped into a handful of **composites**: higher-level summaries of how the content is performing.
  </Step>

  <Step title="Everything resolves into lenses and a score">
    Finally, the response resolves into the **4 lenses** (attention, memory, buy / not buy, and manipulation) plus an overall **score** and **grade**. These are the human-readable answers.
  </Step>
</Steps>

## Why the reduction matters

Each layer is a lossy but honest summary of the one beneath it. The deepest layer, the predicted brain response, is the most precise and the least readable. The shallowest layer, the score, is the easiest to act on and the least detailed. You choose how deep to go depending on the question you're asking.

<Columns cols={2}>
  <Card title="The 7 networks" href="/knowledge-base/the-7-networks">
    Each brain network and what it tells you.
  </Card>

  <Card title="From networks to lenses" href="/knowledge-base/from-networks-to-lenses">
    How per-second networks become the four lenses.
  </Card>
</Columns>