← Help Center

Reference

Glossary

Every term and concept you'll encounter across the Checkstep platform and API - defined in one place.

A

Appeal Platform: A user's request to reverse a moderation decision. When content is removed, the author can submit an appeal through the Transparency Portal (or via the headless API). Appeals create a new incident for moderator review, resulting in either upholding or overturning the original decision.

Author API: The user who created a piece of content, identified by a unique ID from your platform. Designating authors enables Checkstep to build user profiles, track repeated offences, and take author-level actions (suspend, terminate) in addition to content-level enforcement.

C

Community Report Platform: A report submitted by a platform user flagging content they believe violates policies. Community reports are valuable signals that AI may miss - they provide human context, and high volumes can indicate emergencies. Reports are associated with content cases and visible to moderators alongside AI evaluations.

Complex Type API: A content category that describes what kind of user-generated content you're sending - such as profile, post, comment, chat, product, or review. Each complex type can contain multiple fields of different simple types. You can also define custom complex types with Checkstep's team.

Confidence Score Concept: A number from 0 to 1 (or 0–100%) expressing how certain a strategy is that its label applies to a piece of content. Higher scores mean higher certainty. Your rules use confidence thresholds to determine what happens - auto-enforce, send for review, or trust. Different models produce different confidence distributions, so the same threshold number can mean different things across strategies.

Content API: An instance of a complex type - a specific piece of user-generated content on your platform. Content can be created, updated, deleted, and moderated. Each piece of content is uniquely identified by its complex type and ID.

Content Case Platform: A compiled record of all evidence related to a piece of content: different versions with analysis results, decisions, appeals, community reports, and the full incident log. The content case is the unified view moderators and ModBot work from when making decisions.

Content Field API: A typed component of a piece of content - for example, a profile's avatar (image) and username (text). Fields are identified by a unique ID and a simple type. Fields can be nested (an image with a caption) and can carry their own metadata.

D

DSA (Digital Services Act) Concept: EU regulation setting rules for how platforms must handle content moderation, user notifications, and transparency reporting. Checkstep automates DSA compliance - moderation actions automatically generate statements of reason and user notifications via the Transparency Portal.

E

Evaluation Concept: The output of a strategy after scanning content - a label and a score in the range [0.0, 1.0]. Evaluations are the raw signals that rules use to make routing decisions. You can also send your own external evaluations alongside ingested content if you have classification systems outside Checkstep.

I

Incident Platform: Created when scanned content triggers a policy rule. An incident requires actioning - either by a human moderator, ModBot, or automation. Incidents track their cause (AI evaluation, community report, manual review, appeal, or inquiry) and their resolution (enforced, dismissed, overturned, upheld, terminated).

Inquiry API: A manually opened incident on content that wasn't flagged by automation. Useful when you want moderators to review specific content based on external signals - for example, an internal report or a customer complaint that came in through a support channel.

Internal Guidelines Platform: The second layer of a policy - additional context, definitions, edge-case examples, and escalation criteria for moderators. Not visible to end users. If ModBot is enabled, it reads internal guidelines to make nuanced decisions. The quality of your guidelines directly affects ModBot's decision quality.

L

Label Concept: A tag applied to content by a strategy - describing what was detected, such as "nudity," "hate symbol," "financial spam," or "Nazi imagery." Labels are the output of strategies and the input to rules. You can observe labels appearing on content before creating rules for them.

M

Metadata API: Key-value pairs attached to content or individual fields - additional information for moderators like user plan type, region, or account age. Metadata is forwarded back in webhook payloads, making it useful for passing internal identifiers through the Checkstep pipeline and back to your systems.

ModBot Platform: An advanced AI moderation layer that reads your full policy text and internal guidelines, examines flagged content in context, and produces decisions with written rationale. ModBot sits after top-of-funnel scanning - it reviews the 1–2% of content flagged for review, making nuanced decisions that simple threshold rules can't. Decisions include policy citations and reasoning visible in the content timeline.

O

Organization Platform: The top-level container in Checkstep. All configuration and data are housed under an organization. Team members are invited to an organization, and each organization can provision multiple platforms.

OSA (Online Safety Act) Concept: UK regulation governing how platforms handle illegal and harmful content. Like the DSA, Checkstep automates OSA compliance by generating required reports and user notifications when moderation actions are taken.

P

Platform Platform: A provisioned environment within an organization, representing an online platform where user-generated content is hosted. Each organization can have multiple platforms, each with its own API keys, policies, and configuration.

Policy Concept: A content standard you enforce - like "Hate Speech," "Sexual Content," or "Financial Spam." Each policy has three layers: a public description (what users see in the Transparency Portal), internal guidelines (what moderators and ModBot reference), and rules (what the AI acts on). Policies are your source of truth - everything else in Checkstep enforces them.

R

Rule Concept: The connection between a label and an action, defined inside a policy. A rule says: "When this label is detected with this confidence, take this action." Without a rule, a label has no effect - content is tagged but nothing happens. Rules can be scoped to specific content types using tags.

S

Severity API: A classification of how serious a policy violation is - typically low, medium, or high. Returned in webhook payloads alongside policy violations to help your platform prioritize enforcement actions.

Simple Type API: The media type of a content field. Defined set: text, image, audio, video, uri, file. Simple types determine which strategies can scan a given field - image recognition won't run on text fields, and text classifiers won't run on images.

Statement of Reason (SoR) Concept: The explanation provided to a user when their content is removed - including which policy was violated and the text of that policy. Required by the DSA. Checkstep generates SoRs automatically and presents them through the Transparency Portal.

Strategy Concept: An AI model or detection method that scans incoming content and produces evaluations (labels + scores). Four types exist: pre-trained models (e.g., AWS Rekognition), prompt-based LLM classifiers (e.g., OpenAI), example-driven models (e.g., Cohere), and keyword lists. Strategies are configured globally under Settings → Strategies and produce labels that rules consume.

T

Tags API: Labels attached to content for classification and queue routing - for example, #featured or #profile-photo. Tags must start with #. Rules can be scoped to content with specific tags, allowing different enforcement behavior for different parts of your platform.

Threshold Concept: The confidence score boundaries that determine what happens to flagged content. Each rule has a lower threshold (content at or above this is sent for human review) and a higher threshold (content at or above this is auto-enforced). Content below the lower threshold is trusted.

Transparency Portal Platform: The user-facing interface where people whose content was removed can see the removed content, which policy was violated, the policy text, and submit an appeal. Hosted by Checkstep (with custom domain support). Required for DSA and OSA compliance. All content is auto-generated from moderation actions.

W

Webhook API: An HTTP callback from Checkstep to your platform when events occur - content analysed, decisions taken, incidents closed. Four types: decision, author-decision, incident-closed, and analysed-content. Webhooks can also be delivered via Amazon SQS. Payloads can be signed with HMAC for verification.