Rule Configuration
Rules are the mechanisms that parse content entering Nima and distribute it for moderation, either through automated enforcement or by routing it to human moderators.
Types of rules
Nima supports two main types of rules, configurable in the Rule Engine:
- Proactive Detection Rules to automatically ingest and analyze any type of content uploaded with AI models and triage it based on the thresholds (confidence levels) you set to:
- Automatically enforce policy violations for high confidence cases 🔴
- Send content to the configured human moderation queues for cases flagged as 🟠
- Dismiss content flagged as 🟢 if no issues detected.
- Queue Routing Rules (only for Human Moderation) to parse content coming from sources other than proactive detection (user reports, user report illegal content, LEA or Trusted Flagger notices) into queues for manual review, where moderators apply policies.
Configuring rules
Proactive detection rules
A proactive detection rule is associated to one of your Policy ConfigurationPolicy Configuration and displayed on the same screen as each policy.
Steps to configure:
- Source is pre-selected to proactive detection
- Select other trigger criteria - Choose additional conditions such as content type(mandatory),label, custom attributes, and all the available variables.
- Set up AI detection by selecting the Provider and the model(s) to utilise. For multiple models, users can use the AND / OR operator. After this, users can select the different thresholds for detection, resulting in:
- 🟢 for cases to dismiss
- 🟠 for cases to send to queues
- 🔴 for cases where the selected policy will be automatically enforced.
- Assign Policy - choose which policy/sub-policy applies for 🔴 cases (pre-filled if set from the Policies screen).
- Assign Queue for cases that will be selected as 🟠
Trigger criteria support matching operators including equals, does not equal, contains, and does not contain. The negated operators (does not equal, does not contain) let you build rules that exclude a specific value or a list of values — for example, matching content that is not in a given set of languages.
On screen the three zones are labelled by what they do, rather than by colour:
Zone | Label in Nima | Outcome |
|---|---|---|
🟢 | No Policy Violation | The case is dismissed |
🟠 | Requires human review | The case is sent to the queue you assign |
🔴 | Automated policy reinforcement | The assigned policy is enforced automatically |
The three zones are set with a single slider that has two handles. Drag both handles to the same point to remove a zone, leaving only two. Depending on your goals, you can use only:
- 🟢 and 🟠: dismiss non-issues or send flagged content to queues
- 🟢 and 🔴: dismiss non-issues or automatically enforce policy
- 🟠 and 🔴: route all flagged content either to queues or automatic enforcement
Queue routing rules
Queue routing rules are associated to one of your QueuesQueues and displayed on the same screen.
Steps to configure:
- Select the Source among the different options, including User Report, User Report Illegal Content, LEA, Trusted Flaggers.
- Select other trigger criteria: Choose additional conditions such as content type, label, custom attributes, and all the available variables.
- Assign the queue to which the content matching the criteria will be redirected
In case you want to pre-assign policies --> Enable the toggle "Automatically apply policy or workflow action" to automatically apply a policy on intake.
How to prioritise rules in Nima
As multiple rules can have the exact same criteria, Nima allows users to choose how they are prioritised in the Rules tab, by simply dragging and dropping the different rules.
By ranking rules, you are setting what rules with same criteria are hit first by content entering Nima.
Order only matters where criteria overlap. Two rules that can never match the same content are unaffected by their relative position. Where they can both match, the higher rule wins and the lower one never runs.
Example: one rule routes all image content to a general review queue, and a second routes image content carrying the label nudity to a specialist queue. If the general rule sits higher, the specialist rule never fires — an image with that label matches the general rule first. Placing the more specific rule above the general one gives the outcome you expect.
Your ranking is saved as soon as you drop a rule — there is no separate save step. Nima confirms with "Rule order updated successfully".
The default rule always runs last. Every instance has a default rule that catches content matching no other rule. It sits at the bottom of the list, cannot be dragged, reordered or edited, and stays there however you rearrange the rules above it. Treat it as the fallback rather than something you tune.
From the same page, users can also create new rules and edit pre-existing rules.
Finally, the Rules tab is where users can get a full overview of how automated rules have been mapped to policies.
To control the order in which cases already sitting in a queue are served to moderators — a separate question from which rule matched — see Prioritisation EnginePrioritisation Engine.
Testing rules before you go live
The Test Rules button on the Rules page opens a playground where you can submit sample content and see exactly how your configuration would handle it, without sending anything through your live pipeline.
Choose a content type, provide the text or a content URL, and run the test. Nima shows you:
- the outcome — No Rule Violation, Sent to queue, or Automated Enforcement
- the rule that matched, and which of its criteria matched
- the AI provider that ran and the confidence score it returned
- the policy that would be applied
This is the fastest way to confirm a new rule behaves as intended, and to diagnose the common surprises: a rule that never fires because a broader rule sits above it, or a threshold set tighter than the scores your provider actually returns.
Test Rules is in Beta and is not switched on for every instance. If you do not see the button, contact the Tremau Support team.
