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

# Guardrails

> The three layers of safeguards: system, agentic, and human

Focaldata AI is governed by layered safeguards that work together: system controls for secure operation, agent guardrails for research best practice, and human checkpoints for high-impact decisions.

## The three layers

<CardGroup cols={1}>
  <Card title="Layer 1 — System-level guardrails" icon="server">
    Controls that support secure operation and auditability. They govern access, project isolation, and operational constraints that prevent accidental cross-project leakage and limit exposure of sensitive data.
  </Card>

  <Card title="Layer 2 — Agentic guardrails" icon="robot">
    Research best practice encoded into artefact generation and workflow behaviour. They constrain variability where inconsistency creates risk — for example, neutral guide wording, evidence-linked reporting, and enforced interview quality controls.
  </Card>

  <Card title="Layer 3 — Human judgment" icon="user-check">
    Human judgment concentrated at a small number of high-impact checkpoints, rather than spread thinly across every action.
  </Card>
</CardGroup>

## Human approval checkpoints

Projects advance through explicit checkpoints. These concentrate judgment where it matters and reduce accidental commitment.

<Steps>
  <Step title="Objective and context review (recommended)">
    Validate that the brief captures the decision and constraints.
  </Step>

  <Step title="Audience and screener approval (required)">
    Confirm who will be recruited and how eligibility is operationalised.
  </Step>

  <Step title="Discussion guide approval (recommended)">
    Confirm topic flow, neutrality, and coverage against objectives.
  </Step>

  <Step title="Launch confirmation (required)">
    Approve the consolidated project summary before any spend is committed.
  </Step>
</Steps>

<Note>
  The platform standardises a high baseline of rigor, but it does not remove professional responsibility. You remain accountable for research design, interpretation, and all client-facing conclusions.
</Note>