Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.helix.ax/llms.txt

Use this file to discover all available pages before exploring further.

Conventional sensitive-content systems are usually detector-first. They find labels, attach scores, and leave the operating decision to thresholds and manual tuning. Helix is decision-first. The system may produce labels, distributions, traces, and supporting evidence, but the product is the operational call:
  • escalate when the evidence supports action.
  • suppress when the evidence supports not acting.
  • abstain when the system does not have enough confidence to decide.

​
Why this matters

False positives create alert fatigue, broken workflows, and lost trust. False negatives create security, privacy, and compliance risk. The useful product is therefore not “more labels.” It is a better decision boundary on messy, ambiguous content.

​
Human-reviewer model

A strong reviewer asks five questions:
  1. What does this look like?
  2. What else could it be?
  3. What context is it in?
  4. Is it actually risky here?
  5. Am I confident enough to act?
Helix mirrors that process through candidate proposal, confounder evidence, context signals, risk evidence, calibration, and abstention.