Trust & Safety¶

SCOPE_ITEM: Content moderation, fraud detection, review authenticity, seller verification enforcement, and reporting systems that maintain marketplace integrity and user trust.

Decision Tree¶

IF: Marketplace allows user-generated content (listings, messages, reviews). THEN: Include content moderation (at minimum: manual review queue + reporting).

IF: Marketplace handles payments >EUR 500 per transaction. THEN: Include fraud detection signals and enhanced seller verification.

IF: Marketplace has >100 active sellers. THEN: Include automated moderation rules to reduce manual review load.

IF: Marketplace operates in EU. THEN: Include P2B Regulation compliance (transparent policies, complaint handling).

Content Moderation¶

Listing Moderation¶

SCOPE_ITEM: Ensure listings meet platform quality and policy standards.

Pre-Publish Checks (Automated)¶

INCLUDES: - Required field validation (title, description, price, category, images). - Prohibited words/phrases filter (configurable blocklist). - Price sanity check (flag if price is 0 or unreasonably high/low for category). - Duplicate detection (flag listings with >90% title similarity from same seller). - Image count minimum (at least 1 image required).

OPTIONAL: - AI image moderation (NSFW detection via cloud API or on-premises model). - AI text moderation (policy violation detection, contact detail extraction). - Category-specific validation rules (e.g., required attributes for electronics). - Plagiarism detection (description copied from other listings).

Manual Review Queue¶

INCLUDES: - Admin queue for listings flagged by automated checks. - Admin queue for listings reported by users. - Review actions: approve, reject (with reason), request changes. - Rejection notification to seller with specific policy violation cited. - Listing status: pending_review, approved, rejected, changes_requested.

OPTIONAL: - Priority queue (high-value listings, reported listings first). - Reviewer assignment and workload balancing. - Review SLA tracking (review within 24h target).

Message Moderation¶

SCOPE_ITEM: Prevent abuse and off-platform transaction attempts.

INCLUDES: - Contact detail detection (regex for phone numbers, email addresses, URLs). - Detected contact details masked or flagged (configurable: mask vs warn). - Spam rate limiting (max 20 new conversations per hour per user). - Report message button (with reason: spam, harassment, scam, other).

OPTIONAL: - AI-powered scam detection (known scam patterns, urgency language). - Automated warnings when suspicious content detected. - Admin message review queue for reported conversations.

Review Moderation¶

SCOPE_ITEM: Ensure review quality and authenticity.

INCLUDES: - Only verified purchasers can leave reviews (order must be completed). - One review per order (prevent review bombing). - Prohibited content filter (profanity, personal attacks). - Report review button (with reason: fake, inappropriate, not relevant). - Admin review queue for reported reviews.

OPTIONAL: - Review edit window (48h after submission, then locked). - Seller response moderation (same content policy applies). - Review removal by admin with documented reason.

Fraud Detection¶

Account-Level Signals¶

SCOPE_ITEM: Detect fraudulent accounts at registration and during use.

INCLUDES: - Email domain validation (reject disposable email domains). - Rate limiting on account creation (3 per IP per hour). - Duplicate account detection (same email, phone, or payment details). - Account age factor in trust calculations (new accounts = higher risk).

OPTIONAL: - Device fingerprinting (detect multiple accounts from same device). - IP geolocation anomaly (registration country vs ID country mismatch). - Velocity checks (sudden spike in listings or transactions). - Cross-reference with fraud databases (e.g., SEON, Sift Science).

Transaction Fraud Signals¶

SCOPE_ITEM: Detect suspicious transaction patterns.

INCLUDES: - Payment failure rate monitoring per buyer (>3 failures = flag). - Unusual transaction patterns (sudden high-value purchases from new account). - Refund rate monitoring per buyer (>30% refund rate = flag). - Dispute rate monitoring per seller (>5% dispute rate = flag).

OPTIONAL: - Real-time risk scoring per transaction (weighted signal combination). - Automatic hold on high-risk transactions pending manual review. - Chargeback pattern detection. - Coordinated fraud detection (multiple accounts, same payment method or shipping address, purchasing from same seller).

Seller-Specific Fraud Signals¶

INCLUDES: - Listing pricing anomaly (significantly below market for category). - Shipping delay patterns (consistently late or no tracking provided). - Review pattern anomaly (sudden burst of 5-star reviews). - Inventory mismatch (listing shows "in stock" but orders consistently cancelled).

OPTIONAL: - Fake inventory detection (listings never fulfilled). - Counterfeit indicator flags (brand name + unrealistic price). - Shadow profile monitoring (suspended seller creating new account).

Review Authenticity¶

SCOPE_ITEM: Prevent fake, incentivised, or manipulated reviews.

Prevention Measures¶

INCLUDES: - Verified purchase requirement (review tied to completed order). - One review per order per reviewer. - Minimum order completion delay before review allowed (e.g., 24h after delivery). - Review text minimum length (50 characters) to prevent low-effort fake reviews.

OPTIONAL: - Incentivised review detection (flag reviews from users who received discounts or free products — requires seller disclosure). - Review timing analysis (burst of reviews in short period = suspicious). - Review text similarity analysis (copy-paste reviews across sellers). - Reviewer reputation score (reviews from established buyers weighted higher). - Self-review prevention (buyer and seller from same IP/device = flag).

Detection Signals¶

Signal	Indicator	Action
Review burst	>5 reviews for same seller in 24h	Flag for manual review
New account reviewer	Account <7 days old, first review	Lower display priority
Copy-paste text	>80% similarity to another review	Flag for manual review
Rating mismatch	5-star text with negative sentiment	Flag for manual review
Cross-review	Buyer and seller review each other	Flag for manual review

Seller Verification Enforcement¶

SCOPE_ITEM: Ongoing verification beyond initial KYC onboarding.

Verification Levels¶

Level	Requirements	Badge	Benefits
Basic	Email verified, terms accepted	None	Can list, limited visibility
Identity Verified	PSP KYC completed (ID + selfie)	Checkmark	Full listing visibility
Business Verified	KYC + CoC/KVK confirmed	Business badge	Priority in search
Enhanced	All above + proof of address + bank verified	Gold badge	Lowest commission tier

Ongoing Monitoring¶

INCLUDES: - Periodic re-verification trigger (PSP may request updated documents). - Automatic suspension if PSP verification lapses. - Business registration status check (annual, via KVK API if available).

OPTIONAL: - Professional certification verification (for service marketplaces). - Insurance verification (for high-liability services). - Background check integration (via third-party provider).

Reporting System¶

SCOPE_ITEM: Enable users to report policy violations.

User Reporting Flow¶

1. User clicks "Report" on listing / message / review / profile
2. Select reason:
   ├── Spam or misleading
   ├── Counterfeit or illegal
   ├── Offensive content
   ├── Scam or fraud
   ├── Harassment
   └── Other (free text)
3. Optional: additional details (text field)
4. Submit → confirmation message
5. Report enters admin queue
6. Admin reviews → action (warn, remove content, suspend account)
7. Reporter notified of outcome (optional)

Report Data Model¶

CREATE TABLE reports (
  id UUID PRIMARY KEY DEFAULT gen_random_uuid(),
  reporter_id UUID REFERENCES users(id),
  target_type TEXT NOT NULL,       -- listing, message, review, user
  target_id UUID NOT NULL,
  reason TEXT NOT NULL,
  details TEXT,
  status TEXT DEFAULT 'pending',   -- pending, reviewing, resolved, dismissed
  resolution TEXT,                 -- warned, content_removed, account_suspended, dismissed
  resolved_by UUID REFERENCES users(id),
  resolved_at TIMESTAMPTZ,
  created_at TIMESTAMPTZ DEFAULT now()
);

Admin Moderation Queue¶

INCLUDES: - Unified queue for all report types. - Priority sorting (fraud/scam reports first). - Report clustering (multiple reports on same target = higher priority). - Quick actions: warn user, remove content, suspend account. - Resolution templates (standard responses for common violations). - Resolution audit trail (who decided what, when).

OPTIONAL: - SLA tracking (resolve reports within 48h target). - Moderation team assignment and workload balancing. - Escalation to senior moderator for edge cases.

Policy Framework¶

SCOPE_ITEM: Clear policies that define acceptable platform behaviour.

INCLUDES: - Acceptable use policy (what can and cannot be listed). - Community guidelines (expected behaviour for buyers and sellers). - Review policy (what constitutes a fair and authentic review). - Intellectual property policy (counterfeit reporting process). - Consequences matrix (warning → temporary suspension → permanent ban).

COMPLIANCE: EU Platform-to-Business Regulation (P2B). - Transparent terms and conditions for sellers. - Clear explanation of ranking factors. - Clear explanation of any differential treatment. - Complaint handling mechanism for sellers (internal, accessible). - 30-day notice before T&C changes.

COMPLIANCE: Digital Services Act (DSA). - Illegal content reporting mechanism for all users. - Notice-and-action procedure for reported content. - Transparency reporting (annual, if >10M EU users). - Trusted flaggers (if applicable).

Scoping Questions¶

CHECK: What content types need moderation (listings, messages, reviews)? CHECK: Is AI-assisted content moderation desired or manual-only? CHECK: What are the prohibited listing categories? CHECK: Is seller verification beyond PSP KYC needed? CHECK: What is the dispute resolution process? CHECK: What are the consequences for policy violations (warning, suspension, ban)? CHECK: Is fraud detection beyond basic signals needed? CHECK: Does the platform need to comply with DSA (>10M EU users)? CHECK: What is the expected volume of reports/flags per day?