Skip to Content

Workflows

Version 1.1 by Robert Schaub on 2025/12/22 13:26

Workflows

Version: 0.9.70  
Last Updated: December 21, 2025  
Status: CORRECTED - Automation Philosophy Consistent

This page describes FactHarbor's core workflows with the automation-first philosophy.

1. Core Workflow Principles

  • Automation First: 90%+ content published automatically
  • No Approval Bottlenecks: No centralized review queues
  • Quality Gates: Automated validation before publication
  • Sampling Audits: Pattern analysis for system improvement
  • Transparent Confidence: All outputs labeled with confidence scores

2. Claim Submission Workflow

2.1 Claim Extraction

When users submit content (text, articles, web pages), FactHarbor first extracts individual verifiable claims:

Input Types:

  • Single claim: "The Earth is flat"
  • Text with multiple claims: "Climate change is accelerating. Sea levels rose 3mm in 2023. Arctic ice decreased 13% annually."
  • URLs: Web pages analyzed for factual claims

Extraction Process:

  • LLM analyzes submitted content
  • Identifies distinct, verifiable factual claims
  • Separates claims from opinions, questions, or commentary
  • Each claim becomes independent for processing

Output:

  • List of claims with context
  • Each claim assigned unique ID
  • Original context preserved for reference

This extraction ensures:

  • Each claim receives focused analysis
  • Multiple claims in one submission are all processed
  • Claims are properly isolated for independent verification
  • Context is preserved for accurate interpretation

Flow:
```
User submits → Duplicate detection → Categorization → Processing queue → User receives ID
```

Timeline: Seconds  
No approval needed: Instant processing

3. Automated Analysis Workflow

Complete Pipeline:

```
Claim from queue
↓
Evidence gathering (AKEL)
↓
Source evaluation (track record check)
↓
Scenario generation
↓
Verdict synthesis
↓
Risk assessment
↓
Quality gates validation
↓
Decision: PUBLISH or BLOCK
```

Timeline: 10-30 seconds  
Automation Rate: 90%+ published automatically

3.1 Quality Gates Decision

Gate Validation:

  1. Gate 1: Source Quality ✓
    2. Gate 2: Contradiction Search ✓
    3. Gate 3: Uncertainty Quantification ✓
    4. Gate 4: Structural Integrity ✓

If ALL gates PASS:
→ Publish immediately (Mode 2: AI-Generated)
→ Apply appropriate risk tier label
→ Display confidence score
→ Make available for sampling audit

If ANY gate FAILS:
→ Block publication (Mode 1: Draft-Only)
→ Log failure reason
→ Analyze failure pattern
→ Queue system improvement task
→ May re-process after improvements

CRITICAL: No human approval step - gates are automated.

4. Publication Workflow

V0.9.70 CLARIFIED: Risk tiers affect LABELS and AUDIT FREQUENCY, NOT approval requirements.

Standard Flow (90%+)

```
Pass quality gates
↓
Determine risk tier (A/B/C)
↓
Apply appropriate labels
↓
PUBLISH IMMEDIATELY
↓
Add to audit sampling pool
```

No delays, no approval queues

High-Risk Content (Tier A - <10%)

V0.9.70 CORRECTION:

```
Pass quality gates
↓
Identified as Tier A (medical/legal/safety)
↓
PUBLISH IMMEDIATELY with prominent warnings
↓
Higher sampling audit frequency (50%)
```

What changed from V0.9.69:
- ❌ REMOVED: "Risk > 80% → Moderator review"
- ✅ ADDED: "Risk > 80% → Publish with WARNING labels"

Philosophy: Publish with strong warnings, monitor closely through sampling.

Warning Labels for Tier A:
```
⚠️ HIGH-IMPACT TOPIC
AI-Generated Analysis

This claim involves [medical/legal/financial/safety] topics.
- Confidence: [X]%
- Last Updated: [timestamp]
- This is NOT professional advice
- Consult qualified professionals for decisions

[View Evidence] [See Methodology] [Report Issue]
```

Low Quality Content (<10%)

```
FAIL quality gates
↓
Confidence < threshold OR structural issues
↓
BLOCK (Mode 1: Draft-Only)
↓
Log failure patterns
↓
Queue for system improvement
```

NOT: Send for human review  
IS: Improve prompts/algorithms based on failure patterns

5. User Contribution Workflow

Philosophy: Wikipedia-style immediate application + audit trail

```
Contributor edits published content
↓
System validates (basic checks)
↓
Applied IMMEDIATELY
↓
Logged in version history
↓
Reputation earned
↓
May be selected for sampling audit
```

No approval required: Changes apply instantly  
Quality control: Through sampling audits and reputation system

New contributors (<50 reputation): Limited to minor edits

6. Sampling Audit Workflow

Purpose: Improve system quality through pattern analysis

6.1 Selection Process

```
Published content
↓
Stratified sampling (by risk tier, confidence, traffic)
↓
Selected for audit (Tier A: 50%, B: 20%, C: 5%)
↓
Added to audit queue
```

6.2 Audit Execution

```
Auditor receives sample
↓
Reviews against quality standards
↓
Identifies issues/patterns
↓
Logs findings
↓
System improvement tasks created
```

What auditors DO:

  • ✅ Analyze patterns across multiple outputs
  • ✅ Identify systematic issues
  • ✅ Recommend algorithm/prompt improvements
  • ✅ Track accuracy trends

What auditors DON'T DO:

  • ❌ Approve individual outputs before publication
  • ❌ Manually fix individual outputs
  • ❌ Act as gatekeepers
  • ❌ Override quality gates

6.3 Improvement Loop

```
Audit findings aggregated
↓
Patterns identified
↓
System improvements proposed
↓
Implemented and tested
↓
Deployed
↓
Metrics monitored
```

Examples of Improvements:

  • Refine evidence search queries
  • Adjust source reliability weights
  • Enhance contradiction detection
  • Improve claim extraction prompts
  • Recalibrate risk tier thresholds

7. Flagging Workflow

Two types of flags:

7.1 Quality Issues

```
User flags quality issue
↓
Categorized automatically
↓
Added to sampling audit pool (priority)
↓
Pattern analysis
↓
System improvement if pattern found
```

NOT: Manual correction of individual claim  
IS: Improve system to prevent similar issues

7.2 Abuse/Spam

```
User flags abuse/spam
↓
Automated pre-moderation check
↓
Moderator review (if needed)
↓
Action taken (hide/ban)
```

Moderator role: Handle abuse/spam, NOT content quality

8. Moderation Workflow

V0.9.70 CLARIFIED: Moderators handle ABUSE, not content quality

8.1 Content Moderation (Abuse/Spam)

Moderator Queue Contains:

  • Flagged abusive content
  • Spam detection alerts
  • Harassment reports
  • Privacy violations
  • Terms of service violations

Moderator Actions:

  • Hide abusive content
  • Ban repeat offenders
  • Handle appeals
  • Escalate to governing team

Moderators DO NOT:

  • ❌ Approve content for publication
  • ❌ Review content quality before publication
  • ❌ Act as editorial gatekeepers
  • ❌ Manually fix AI outputs

8.2 Appeal Process

```
User disagrees with moderation
↓
Appeals to different moderator
↓
If still disagrees, escalates to Governing Team
↓
Governing Team decision (final)
```

9. Time Evolution Workflow

Automatic Re-evaluation:

```
Published claim
↓
Monitoring for triggers:
  - New evidence published
  - Source retractions
  - Significant events
  - Scheduled review (6-12 months)
↓
Trigger detected
↓
AKEL re-processes claim
↓
Quality gates validate
↓
If verdict changes: Correction workflow
↓
If passes: Update published analysis
```

Correction Workflow (New in V0.9.70):

```
Verdict changed significantly
↓
Generate correction notice
↓
Publish correction banner (30 days)
↓
Update corrections log
↓
Notify users (email, RSS, API)
↓
Update ClaimReview schema
```

10. Contributor Journey

  1. Visitor – Explores platform, reads documentation
    2. New Contributor – Submits first improvements (typo fixes, clarifications)
    3. Contributor – Contributes regularly, follows conventions
    4. Trusted Contributor – Track record of quality work
    5. Reviewer – Participates in sampling audits (pattern analysis)
    6. Moderator – Handles abuse/spam (not content quality)
    7. Expert (optional) – Provides domain expertise for contested claims

All contributions apply immediately - no approval workflow

11. Related Pages

V0.9.70 CHANGES:

REMOVED:
- ❌ "High Risk → Moderator review" (was approval workflow)
- ❌ "Review queue" language for publication
- ❌ Any implication that moderators approve content quality

ADDED/CLARIFIED:
- ✅ Risk tiers affect warnings and audit frequency, NOT approval
- ✅ High-risk content publishes immediately with prominent warnings
- ✅ Quality gate failures → Block + improve system (not human review)
- ✅ Clear distinction: Sampling audits (improvement) vs. Content moderation (abuse)
- ✅ Moderator role clarified: Abuse only, NOT content quality
- ✅ User contributions apply immediately (Wikipedia model)
- ✅ Correction workflow for significant verdict changes
- ✅ Time evolution and re-evaluation workflow