> For the complete documentation index, see [llms.txt](https://cogna8-io.gitbook.io/docs/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://cogna8-io.gitbook.io/docs/understanding-cogna8/how-it-works.md).

# How It Works

Cogna8 operates as an authority runtime in the action path. It evaluates information as it naturally arrives, maintains governed state, and determines whether proposed actions are authorized to proceed based on current evidence, policy, and unresolved conflicts.

***

## The Authority Pipeline

{% stepper %}
{% step %}
**Input Ingestion**

A user, operator, or upstream agent produces information through the chat interface, an API call, a webhook, or another integrated path.
{% endstep %}

{% step %}
**State Extraction and Classification**

Cogna8 identifies information that carries operational weight and classifies it into **typed state items** with governance characteristics appropriate to each category of information.
{% endstep %}

{% step %}
**Normalization and Canonical Keying**

Extracted values are normalized into comparable forms and assigned **canonical keys** - unique identifiers that allow the system to recognize when two statements refer to the same underlying concept.
{% endstep %}

{% step %}
**Conflict Detection**

New state is compared against active state in the applicable scope. Contradictions are surfaced as explicit **conflict records** with type classification and severity metadata.
{% endstep %}

{% step %}
**Policy Enforcement**

Deterministic policy rules evaluate state transitions and action prerequisites. Policies produce repeatable, inspectable outcomes - not probabilistic assessments.
{% endstep %}

{% step %}
**Action Gating**

When an action request is present, Cogna8 evaluates whether the action is authorized: required state is **present**, **in scope**, **sufficiently consistent**, and **policy-compliant**. The gate returns **allow**, **warn**, or **block** with a structured authorization receipt.
{% endstep %}

{% step %}
**Trace and Audit Recording**

State changes, conflict events, gate evaluations, and their associated reasoning are recorded in a structured audit trail. Each trace entry is tied to specific state items and specific outcomes.
{% endstep %}
{% endstepper %}

***

## The Continuous Loop

{% hint style="info" %}
**Authorization is not a one-time validation step.** As new information arrives, integrity is re-evaluated continuously. A state item that was consistent when first recorded may become stale or conflicted three interactions later.
{% endhint %}

***

## Pipeline Visibility

The authority pipeline is **visible in the product interface**. Users and operators see the system performing integrity work at each stage - not just receiving a final answer.

> The pipeline doesn't just produce better outcomes. It produces **visible** better outcomes - which is what builds operational confidence.