Curation

Guide to operating Lighthouse as a content curator.

---

What is a Curator?

A curator is a trusted entity that:

• Reviews incoming content
• Applies moderation rulesets
• Signs verification decisions
• Publishes decisions for others to consume

Curators are the “human filters” in the Web of Trust model.

---

Becoming a Curator

Prerequisites

1. A Nostr identity (npub/nsec)
2. A Lighthouse instance
3. Understanding of the content you’re curating
4. Commitment to consistent moderation

Setup Steps

1. Configure Strict Trust

   trust:
     depth: 0  # Whitelist only
   

2. Enable Tag Filtering (optional)

   indexer:
     tag_filter_enabled: true
     tag_filter:
       - movies
       - tv
   

3. Import or Create Rulesets

4. Start Processing Content

---

Rulesets

Rulesets define moderation policies. There are two types:

Censoring Rulesets

Deterministic blocking rules. Rejections always take priority.

Code	Description
LEGAL_DMCA	Documented DMCA takedown
LEGAL_ILLEGAL	Manifestly illegal content
ABUSE_SPAM	Spam or flooding
ABUSE_MALWARE	Malicious content

Semantic Rulesets

Quality and classification rules. Aggregation policy applies.

Code	Description
SEM_DUPLICATE_EXACT	Exact duplicate (same infohash)
SEM_DUPLICATE_PROBABLE	Same files, different event
SEM_BAD_META	Incomplete metadata
SEM_LOW_QUALITY	Below quality threshold
SEM_CATEGORY_MISMATCH	Wrong category

---

Creating Rulesets

Ruleset Structure

{
  "name": "Movie Curation Ruleset",
  "type": "semantic",
  "version": "1.0.0",
  "description": "Quality rules for movie torrents",
  "author": "npub1...",
  "rules": [
    {
      "id": "min-size",
      "name": "Minimum Size",
      "type": "threshold",
      "field": "size",
      "operator": "gte",
      "value": 100000000,
      "reason_code": "SEM_LOW_QUALITY",
      "priority": 10
    }
  ]
}

Rule Types

Pattern Rules

Match against text fields.

{
  "type": "pattern",
  "field": "name",
  "operator": "contains",
  "value": "CAM",
  "reason_code": "SEM_LOW_QUALITY"
}

Operators: equals, contains, starts_with, ends_with, regex

Threshold Rules

Compare numeric values.

{
  "type": "threshold",
  "field": "size",
  "operator": "gte",
  "value": 100000000,
  "reason_code": "SEM_LOW_QUALITY"
}

Operators: eq, ne, gt, gte, lt, lte

List Rules

Match against lists.

{
  "type": "list",
  "field": "tags",
  "operator": "contains_any",
  "value": ["spam", "fake"],
  "reason_code": "ABUSE_SPAM"
}

Operators: contains_any, contains_all, contains_none

Available Fields

Field	Type	Description
name	string	Torrent name
size	integer	Size in bytes
category	integer	Torznab category
tags	array	Nostr tags
pubkey	string	Publisher’s npub
created_at	timestamp	Publication time

---

Importing Rulesets

Via UI

1. Go to Curation page
2. Click Import Ruleset
3. Paste JSON content
4. Review and save

Via API

curl -X POST http://localhost:9999/api/rulesets \
  -H "X-API-Key: your-key" \
  -H "Content-Type: application/json" \
  -d @ruleset.json

From URL

curl -X POST http://localhost:9999/api/rulesets/import \
  -H "X-API-Key: your-key" \
  -H "Content-Type: application/json" \
  -d '{"url": "https://example.com/ruleset.json"}'

---

Verification Decisions

When content is processed, a decision is created:

{
  "decision_id": "abc123",
  "decision": "accept",
  "reason_codes": [],
  "ruleset_type": "semantic",
  "ruleset_version": "1.0.0",
  "ruleset_hash": "sha256...",
  "target_event_id": "nostr_event_id",
  "target_infohash": "aabbccdd...",
  "curator_pubkey": "npub1...",
  "signature": "ed25519_sig...",
  "created_at": "2024-01-01T00:00:00Z"
}

Decision Flow

Content → Censoring Rules → Semantic Rules → Decision → Sign → Publish

Signing Decisions

Decisions are signed with your nsec:

signature = Ed25519.Sign(privateKey, decisionHash)

The signature proves:

• You made the decision
• The decision hasn’t been tampered with
• The exact ruleset version was used

---

Publishing Decisions

Decisions can be published to Nostr relays.

Enable Publishing

curator:
  publish_decisions: true
  publish_relays:
    - "wss://relay.example.com"

Decision Event Format

Decisions are published as Kind 30172 events:

{
  "kind": 30172,
  "content": "{decision_json}",
  "tags": [
    ["d", "decision_id"],
    ["infohash", "aabbccdd..."],
    ["decision", "accept"],
    ["ruleset", "semantic", "1.0.0", "sha256..."]
  ],
  "pubkey": "curator_pubkey",
  "sig": "signature"
}

---

Handling Reports

Curators receive reports from users.

Report Flow

User submits report → Report queued → Curator reviews → Decision updated

Review Process

1. Go to Reports page
2. Review pending reports
3. Acknowledge report (marks as seen)
4. Investigate if needed
5. Resolve or reject

Report Categories

Category	Description
dmca	DMCA takedown request
illegal	Illegal content
spam	Spam or flooding
malware	Malicious content
false_info	Incorrect metadata
duplicate	Duplicate content
other	Other issues

---

Appeals

Users can appeal decisions.

Appeal Flow

User submits appeal → Appeal queued → Curator reviews → Decision reconsidered

Handling Appeals

1. Review appeal evidence
2. Re-evaluate original decision
3. Update decision if warranted
4. Respond with resolution

---

Best Practices

Consistency

• Apply rules consistently
• Document your policies
• Use versioned rulesets
• Communicate changes

Transparency

• Publish your ruleset publicly
• Explain rejection reasons
• Respond to appeals fairly

Specialization

Consider focusing on specific content:

• Movie Curator - Film-focused rules
• Linux Curator - Open source software
• Music Curator - Audio releases

Collaboration

• Share rulesets with other curators
• Coordinate on legal takedowns
• Build reputation through quality

---

Ruleset Examples

Basic Quality Filter

{
  "name": "Basic Quality",
  "type": "semantic",
  "version": "1.0.0",
  "rules": [
    {
      "id": "min-size",
      "name": "Minimum Size (100MB)",
      "type": "threshold",
      "field": "size",
      "operator": "gte",
      "value": 100000000,
      "reason_code": "SEM_LOW_QUALITY"
    },
    {
      "id": "no-cam",
      "name": "No CAM Releases",
      "type": "pattern",
      "field": "name",
      "operator": "regex",
      "value": "\\bCAM\\b",
      "reason_code": "SEM_LOW_QUALITY"
    }
  ]
}

Anti-Spam Filter

{
  "name": "Anti-Spam",
  "type": "censoring",
  "version": "1.0.0",
  "rules": [
    {
      "id": "spam-tags",
      "name": "Spam Tags",
      "type": "list",
      "field": "tags",
      "operator": "contains_any",
      "value": ["spam", "fake", "virus"],
      "reason_code": "ABUSE_SPAM"
    },
    {
      "id": "spam-names",
      "name": "Spam Names",
      "type": "pattern",
      "field": "name",
      "operator": "regex",
      "value": "(\\bFREE\\b.*\\bDOWNLOAD\\b|\\bCLICK\\s+HERE\\b)",
      "reason_code": "ABUSE_SPAM"
    }
  ]
}

---

Metrics

Track your curation effectiveness:

Metric	Description
Decisions/day	Volume processed
Accept rate	% accepted
Appeal rate	% decisions appealed
Overturn rate	% appeals resulting in change

---

Next Steps

• Federation - Multi-curator deployment
• Web of Trust - Trust system
• API Reference - Automation