Skip to Content

Wiki source code of FAQ

Last modified by Robert Schaub on 2025/12/18 12:03
Show last authors
1 = Frequently Asked Questions (FAQ) =
2 Common questions about FactHarbor's design, functionality, and approach.
3 == 1. How do claims get evaluated in FactHarbor? ==
4 === 1.1 User Submission ===
5 **Who**: Anyone can submit claims
6 **Process**: User submits claim text + source URLs
7 **Speed**: Typically <20 seconds to verdict
8 === 1.2 AKEL Processing (Automated) ===
9 **What**: AI Knowledge Extraction Layer analyzes claim
10 **Steps**:
11 * Parse claim into testable components
12 * Extract evidence from provided sources
13 * Score source credibility
14 * Generate verdict with confidence level
15 * Assign risk tier
16 * Publish automatically
17 **Authority**: AKEL makes all content decisions
18 **Scale**: Can process millions of claims
19 === 1.3 Continuous Improvement (Human Role) ===
20 **What**: Humans improve the system, not individual verdicts
21 **Activities**:
22 * Monitor aggregate performance metrics
23 * Identify systematic errors
24 * Propose algorithm improvements
25 * Update policies and rules
26 * Test changes before deployment
27 **NOT**: Reviewing individual claims for approval
28 **Focus**: Fix the system, not the data
29 === 1.4 Exception Handling ===
30 **When AKEL flags for review**:
31 * Low confidence verdict
32 * Detected manipulation attempt
33 * Unusual pattern requiring attention
34 **Moderator role**:
35 * Reviews flagged items
36 * Takes action on abuse/manipulation
37 * Proposes detection improvements
38 * Does NOT override verdicts
39 === 1.5 Why This Model Works ===
40 **Scale**: Automation handles volume humans cannot
41 **Consistency**: Same rules applied uniformly
42 **Transparency**: Algorithms can be audited
43 **Improvement**: Systematic fixes benefit all claims
44 == 2. What prevents FactHarbor from becoming another echo chamber? ==
45 FactHarbor includes multiple safeguards against echo chambers and filter bubbles:
46 **Mandatory Contradiction Search**:
47 * AI must actively search for counter-evidence, not just confirmations
48 * System checks for echo chamber patterns in source clusters
49 * Flags tribal or ideological source clustering
50 * Requires diverse perspectives across political/ideological spectrum
51 **Multiple Scenarios**:
52 * Claims are evaluated under different interpretations
53 * Reveals how assumptions change conclusions
54 * Makes disagreements understandable, not divisive
55 **Transparent Reasoning**:
56 * All assumptions, definitions, and boundaries are explicit
57 * Evidence chains are traceable
58 * Uncertainty is quantified, not hidden
59 **Audit System**:
60 * Human auditors check for bubble patterns
61 * Feedback loop improves AI search diversity
62 * Community can flag missing perspectives
63 **Federation**:
64 * Multiple independent nodes with different perspectives
65 * No single entity controls "the truth"
66 * Cross-node contradiction detection
67 == 3. How does FactHarbor handle claims that are "true in one context but false in another"? ==
68 This is exactly what FactHarbor is designed for:
69 **Scenarios capture contexts**:
70 * Each scenario defines specific boundaries, definitions, and assumptions
71 * The same claim can have different verdicts in different scenarios
72 * Example: "Coffee is healthy" depends on:
73 ** Definition of "healthy" (reduces disease risk? improves mood? affects specific conditions?)
74 ** Population (adults? pregnant women? people with heart conditions?)
75 ** Consumption level (1 cup/day? 5 cups/day?)
76 ** Time horizon (short-term? long-term?)
77 **Truth Landscape**:
78 * Shows all scenarios and their verdicts side-by-side
79 * Users see *why* interpretations differ
80 * No forced consensus when legitimate disagreement exists
81 **Explicit Assumptions**:
82 * Every scenario states its assumptions clearly
83 * Users can compare how changing assumptions changes conclusions
84 * Makes context-dependence visible, not hidden
85 == 4. What makes FactHarbor different from traditional fact-checking sites? ==
86 **Traditional Fact-Checking**:
87 * Binary verdicts: True / Mostly True / False
88 * Single interpretation chosen by fact-checker
89 * Often hides legitimate contextual differences
90 * Limited ability to show *why* people disagree
91 **FactHarbor**:
92 * **Multi-scenario**: Shows multiple valid interpretations
93 * **Likelihood-based**: Ranges with uncertainty, not binary labels
94 * **Transparent assumptions**: Makes boundaries and definitions explicit
95 * **Version history**: Shows how understanding evolves
96 * **Contradiction search**: Actively seeks opposing evidence
97 * **Federated**: No single authority controls truth
98 == 5. How do you prevent manipulation or coordinated misinformation campaigns? ==
99 **Quality Gates**:
100 * Automated checks before AI-generated content publishes
101 * Source quality verification
102 * Mandatory contradiction search
103 * Bubble detection for coordinated campaigns
104 **Audit System**:
105 * Stratified sampling catches manipulation patterns
106 * Trusted Contributor auditors validate AI research quality
107 * Failed audits trigger immediate review
108 **Transparency**:
109 * All reasoning chains are visible
110 * Evidence sources are traceable
111 * AKEL involvement clearly labeled
112 * Version history preserved
113 **Moderation**:
114 * Moderators handle abuse, spam, coordinated manipulation
115 * Content can be flagged by community
116 * Audit trail maintained even if content hidden
117 **Federation**:
118 * Multiple nodes with independent governance
119 * No single point of control
120 * Cross-node contradiction detection
121 * Trust model prevents malicious node influence
122 == 6. What happens when new evidence contradicts an existing verdict? ==
123 FactHarbor is designed for evolving knowledge:
124 **Automatic Re-evaluation**:
125 1. New evidence arrives
126 2. System detects affected scenarios and verdicts
127 3. AKEL proposes updated verdicts
128 4. Contributors/experts validate
129 5. New verdict version published
130 6. Old versions remain accessible
131 **Version History**:
132 * Every verdict has complete history
133 * Users can see "as of date X, what did we know?"
134 * Timeline shows how understanding evolved
135 **Transparent Updates**:
136 * Reason for re-evaluation documented
137 * New evidence clearly linked
138 * Changes explained, not hidden
139 **User Notifications**:
140 * Users following claims are notified of updates
141 * Can compare old vs new verdicts
142 * Can see which evidence changed conclusions
143 == 7. Who can submit claims to FactHarbor? ==
144 **Anyone** - even without login:
145 **Readers** (no login required):
146 * Browse and search all published content
147 * Submit text for analysis
148 * New claims added automatically unless duplicates exist
149 * System deduplicates and normalizes
150 **Contributors** (logged in):
151 * Everything Readers can do
152 * Submit evidence sources
153 * Suggest scenarios
154 * Participate in discussions
155 **Workflow**:
156 1. User submits text (as Reader or Contributor)
157 2. AKEL extracts claims
158 3. Checks for existing duplicates
159 4. Normalizes claim text
160 5. Assigns risk tier
161 6. Generates scenarios (draft)
162 7. Runs quality gates
163 8. Publishes as AI-Generated (Mode 2) if passes
164 == 8. What are "risk tiers" and why do they matter? ==
165 Risk tiers determine review requirements and publication workflow:
166 **Tier A (High Risk)**:
167 * **Domains**: Medical, legal, elections, safety, security, major financial
168 * **Publication**: AI can publish with warnings, expert review required for "AKEL-Generated" status
169 * **Audit rate**: Recommendation 30-50%
170 * **Why**: Potential for significant harm if wrong
171 **Tier B (Medium Risk)**:
172 * **Domains**: Complex policy, science causality, contested issues
173 * **Publication**: AI can publish immediately with clear labeling
174 * **Audit rate**: Recommendation 10-20%
175 * **Why**: Nuanced but lower immediate harm risk
176 **Tier C (Low Risk)**:
177 * **Domains**: Definitions, established facts, historical data
178 * **Publication**: AI publication default
179 * **Audit rate**: Recommendation 5-10%
180 * **Why**: Well-established, low controversy
181 **Assignment**:
182 * AKEL suggests tier based on domain, keywords, impact
183 * Moderators and Trusted Contributors can override
184 * Risk tiers reviewed based on audit outcomes
185 == 9. How does federation work and why is it important? ==
186 **Federation Model**:
187 * Multiple independent FactHarbor nodes
188 * Each node has own database, AKEL, governance
189 * Nodes exchange claims, scenarios, evidence, verdicts
190 * No central authority
191 **Why Federation Matters**:
192 * **Resilience**: No single point of failure or censorship
193 * **Autonomy**: Communities govern themselves
194 * **Scalability**: Add nodes to handle more users
195 * **Specialization**: Domain-focused nodes (health, energy, etc.)
196 * **Trust diversity**: Multiple perspectives, not single truth source
197 **How Nodes Exchange Data**:
198 1. Local node creates versions
199 2. Builds signed bundle
200 3. Pushes to trusted neighbor nodes
201 4. Remote nodes validate signatures and lineage
202 5. Accept or branch versions
203 6. Local re-evaluation if needed
204 **Trust Model**:
205 * Trusted nodes → auto-import
206 * Neutral nodes → import with review
207 * Untrusted nodes → manual only
208 == 10. Can experts disagree in FactHarbor? ==
209 **Yes - and that's a feature, not a bug**:
210 **Multiple Scenarios**:
211 * Trusted Contributors can create different scenarios with different assumptions
212 * Each scenario gets its own verdict
213 * Users see *why* experts disagree (different definitions, boundaries, evidence weighting)
214 **Parallel Verdicts**:
215 * Same scenario, different expert interpretations
216 * Both verdicts visible with expert attribution
217 * No forced consensus
218 **Transparency**:
219 * Trusted Contributor reasoning documented
220 * Assumptions stated explicitly
221 * Evidence chains traceable
222 * Users can evaluate competing expert opinions
223 **Federation**:
224 * Different nodes can have different expert conclusions
225 * Cross-node branching allowed
226 * Users can see how conclusions vary across nodes
227 == 11. What prevents AI from hallucinating or making up facts? ==
228 **Multiple Safeguards**:
229 **Quality Gate 4: Structural Integrity**:
230 * Fact-checking against sources
231 * No hallucinations allowed
232 * Logic chain must be valid and traceable
233 * References must be accessible and verifiable
234 **Evidence Requirements**:
235 * Primary sources required
236 * Citations must be complete
237 * Sources must be accessible
238 * Reliability scored
239 **Audit System**:
240 * Human auditors check AI-generated content
241 * Hallucinations caught and fed back into training
242 * Patterns of errors trigger system improvements
243 **Transparency**:
244 * All reasoning chains visible
245 * Sources linked
246 * Users can verify claims against sources
247 * AKEL outputs clearly labeled
248 **Human Oversight**:
249 * Tier A marked as highest risk
250 * Audit sampling catches errors
251 * Community can flag issues
252 == 12. How does FactHarbor make money / is it sustainable? ==
253 [ToDo: Business model and sustainability to be defined]
254 Potential models under consideration:
255 * Non-profit foundation with grants and donations
256 * Institutional subscriptions (universities, research organizations, media)
257 * API access for third-party integrations
258 * Premium features for power users
259 * Federated node hosting services
260 Core principle: **Public benefit** mission takes priority over profit.
261 == 13. Related Pages ==
262 * [[Requirements (Roles)>>FactHarbor.Specification.Requirements.WebHome]]
263 * [[AKEL (AI Knowledge Extraction Layer)>>FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]]
264 * [[Automation>>FactHarbor.Specification.Automation.WebHome]]
265 * [[Federation & Decentralization>>FactHarbor.Specification.Federation & Decentralization.WebHome]]
266 * [[Mission & Purpose>>FactHarbor.Organisation.Core Problems FactHarbor Solves.WebHome]]
267 == 20. Glossary / Key Terms ==
268 === Phase 0 vs POC v1 ===
269 These terms refer to the same stage of FactHarbor's development:
270 * **Phase 0** - Organisational perspective: Pre-alpha stage with founder-led governance
271 * **POC v1** - Technical perspective: Proof of Concept demonstrating AI-generated publication
272 Both describe the current development stage where the platform is being built and initially validated.
273 === Beta 0 ===
274 The next development stage after POC, featuring:
275 * External testers
276 * Basic federation experiments
277 * Enhanced automation
278 === Release 1.0 ===
279 The first public release featuring:
280 * Full federation support
281 * 2000+ concurrent users
282 * Production-grade infrastructure