Moderate with policy you version, change it without a deploy, replay any disputed call.

The moderation lifecycle, on one engine

Content comes in. Swiftward computes signals (classifier models for text and images, plus your own), evaluates policy, and routes the result: allow, block, rate-limit, or send to human review. Reputation and velocity are tracked as state, so a first offense and a tenth offense are treated differently. Every decision is traced, so when a takedown is disputed you replay exactly what was decided and why.

Policy your team owns

Edit a rule and ship it without a release train. Test the change in shadow mode against live traffic before it affects a single user, or A/B two versions and compare. New harm pattern appears? You respond in minutes, in policy, not in a sprint.

What a rule looks like

Reputation is state, so a first offense and a fourth are treated differently. Versioned, replayable, changed without a deploy.

rules:
  # A toxic post from a user who has already crossed the line three times
  # this month is rejected, and the strike is counted.
  repeat_offender:
    all:
      - path: "event.type"
        op: eq
        value: "ugc.post.created"
      - path: "signals.toxicity_score"
        op: gte
        value: 0.9
      - path: "state.user.counters.violations_30d"
        op: gte
        value: 3
    effects:
      verdict: rejected
      priority: 100
      state_changes:
        user:
          set_labels:
            repeat_offender:
          change_counters:
            violations_30d: 1

Built for the people who review

The human queue is a real review surface, not an afterthought: bulk actions, escalation paths, SLAs, and the screens and permissions configured to fit how your team works. Reviewer protection is built into the workflow, not bolted on: blurring and grayscale by default on graphic queues, exposure caps per shift, and mandatory rotation off the hardest content, all set as policy. An appeals path is its own workflow, distinct from first-pass review, on the same audit trail. On detection we are straight with you: Swiftward orchestrates the detection signals you bring - classifiers and hash-match lists, your own included - rather than claiming a best-in-class detector for every modality. We are the control plane; you keep the detectors that fit your platform. The same pattern closes the mandatory reporting workflow on top of it: a flagged item routes to a child-safety queue, a reviewer applies the industry classification, and the decision drives the report to the NCMEC CyberTipline for CSAM, fires the preservation hold, and lands the whole chain on one tamper-evident audit trail.

Versus assembling open source

You can stand up open-source moderation building blocks and run them yourself, across several services, with no SSO, no access control, no audit, and no support line. Swiftward is the single, on-prem control plane where moderation is one policy pack on a foundation that already passes an enterprise security review. If you are launching a platform, including one built for AI-generated content, you get the T&S engine and the enterprise wrapping together. See how we compare to Hive, ActiveFence, and ROOST.

Map your decision system first

Before you change anything, see the whole picture. Our Trust & Safety Decision System Map is a vendor-neutral, seven-layer reference - decisions, signals, policy logic, enforcement, human ops, change control, and audit. Use it to audit your own system and find the layers you are missing.