Skip to Content

Changes for page FAQ

Last modified by Robert Schaub on 2026/02/08 21:20
From version 2.3
edited by Robert Schaub
on 2025/12/24 20:30
Change comment: Update document after refactoring.
To version 2.7
edited by Robert Schaub
on 2026/01/20 20:29
Change comment: Renamed back-links.

Summary

Details

Page properties
Content
... ... @@ -12,6 +12,7 @@
12 12  **What**: System dynamically researches claims using AKEL (AI Knowledge Extraction Layer)
13 13  
14 14  **Process**:
15 +
15 15  * Extracts claims from submitted text
16 16  * Generates structured sub-queries
17 17  * Performs **mandatory contradiction search** (actively seeks counter-evidence, not just confirmations)
... ... @@ -39,6 +39,7 @@
39 39  **What**: Sampling audits where experts review AI-generated content
40 40  
41 41  **Rates**:
43 +
42 42  * High-risk (Tier A): 30-50% sampling
43 43  * Medium-risk (Tier B): 10-20% sampling
44 44  * Low-risk (Tier C): 5-10% sampling
... ... @@ -50,6 +50,7 @@
50 50  === 1.4 Why All Three Matter ===
51 51  
52 52  **Complementary Strengths**:
55 +
53 53  * **AI research**: Scale and speed for emerging claims
54 54  * **Expert authoring**: Authority and precision for critical domains
55 55  * **Audit feedback**: Continuous quality improvement
... ... @@ -57,6 +57,7 @@
57 57  **Expert Time Optimization**:
58 58  
59 59  Experts can choose where to focus their time:
63 +
60 60  * Author high-priority content directly
61 61  * Validate and edit AI-generated outputs
62 62  * Audit samples to improve system-wide AI performance
... ... @@ -75,6 +75,7 @@
75 75  FactHarbor includes multiple safeguards against echo chambers and filter bubbles:
76 76  
77 77  **Mandatory Contradiction Search**:
82 +
78 78  * AI must actively search for counter-evidence, not just confirmations
79 79  * System checks for echo chamber patterns in source clusters
80 80  * Flags tribal or ideological source clustering
... ... @@ -81,53 +81,59 @@
81 81  * Requires diverse perspectives across political/ideological spectrum
82 82  
83 83  **Multiple Scenarios**:
89 +
84 84  * Claims are evaluated under different interpretations
85 85  * Reveals how assumptions change conclusions
86 86  * Makes disagreements understandable, not divisive
87 87  
88 88  **Transparent Reasoning**:
95 +
89 89  * All assumptions, definitions, and boundaries are explicit
90 90  * Evidence chains are traceable
91 91  * Uncertainty is quantified, not hidden
92 92  
93 93  **Audit System**:
101 +
94 94  * Human auditors check for bubble patterns
95 95  * Feedback loop improves AI search diversity
96 96  * Community can flag missing perspectives
97 97  
98 98  **Federation**:
107 +
99 99  * Multiple independent nodes with different perspectives
100 100  * No single entity controls "the truth"
101 101  * Cross-node contradiction detection
102 102  
103 -
104 104  == 3. How does FactHarbor handle claims that are "true in one context but false in another"? ==
105 105  
106 106  This is exactly what FactHarbor is designed for:
107 107  
108 108  **Scenarios capture contexts**:
117 +
109 109  * Each scenario defines specific boundaries, definitions, and assumptions
110 110  * The same claim can have different verdicts in different scenarios
111 111  * Example: "Coffee is healthy" depends on:
112 - ** Definition of "healthy" (reduces disease risk? improves mood? affects specific conditions?)
113 - ** Population (adults? pregnant women? people with heart conditions?)
114 - ** Consumption level (1 cup/day? 5 cups/day?)
115 - ** Time horizon (short-term? long-term?)
121 +** Definition of "healthy" (reduces disease risk? improves mood? affects specific conditions?)
122 +** Population (adults? pregnant women? people with heart conditions?)
123 +** Consumption level (1 cup/day? 5 cups/day?)
124 +** Time horizon (short-term? long-term?)
116 116  
117 117  **Truth Landscape**:
127 +
118 118  * Shows all scenarios and their verdicts side-by-side
119 119  * Users see *why* interpretations differ
120 120  * No forced consensus when legitimate disagreement exists
121 121  
122 122  **Explicit Assumptions**:
133 +
123 123  * Every scenario states its assumptions clearly
124 124  * Users can compare how changing assumptions changes conclusions
125 125  * Makes context-dependence visible, not hidden
126 126  
127 -
128 128  == 4. What makes FactHarbor different from traditional fact-checking sites? ==
129 129  
130 130  **Traditional Fact-Checking**:
141 +
131 131  * Binary verdicts: True / Mostly True / False
132 132  * Single interpretation chosen by fact-checker
133 133  * Often hides legitimate contextual differences
... ... @@ -134,6 +134,7 @@
134 134  * Limited ability to show *why* people disagree
135 135  
136 136  **FactHarbor**:
148 +
137 137  * **Multi-scenario**: Shows multiple valid interpretations
138 138  * **Likelihood-based**: Ranges with uncertainty, not binary labels
139 139  * **Transparent assumptions**: Makes boundaries and definitions explicit
... ... @@ -141,10 +141,10 @@
141 141  * **Contradiction search**: Actively seeks opposing evidence
142 142  * **Federated**: No single authority controls truth
143 143  
144 -
145 145  == 5. How do you prevent manipulation or coordinated misinformation campaigns? ==
146 146  
147 147  **Quality Gates**:
159 +
148 148  * Automated checks before AI-generated content publishes
149 149  * Source quality verification
150 150  * Mandatory contradiction search
... ... @@ -151,11 +151,13 @@
151 151  * Bubble detection for coordinated campaigns
152 152  
153 153  **Audit System**:
166 +
154 154  * Stratified sampling catches manipulation patterns
155 155  * Expert auditors validate AI research quality
156 156  * Failed audits trigger immediate review
157 157  
158 158  **Transparency**:
172 +
159 159  * All reasoning chains are visible
160 160  * Evidence sources are traceable
161 161  * AKEL involvement clearly labeled
... ... @@ -162,22 +162,24 @@
162 162  * Version history preserved
163 163  
164 164  **Moderation**:
179 +
165 165  * Moderators handle abuse, spam, coordinated manipulation
166 166  * Content can be flagged by community
167 167  * Audit trail maintained even if content hidden
168 168  
169 169  **Federation**:
185 +
170 170  * Multiple nodes with independent governance
171 171  * No single point of control
172 172  * Cross-node contradiction detection
173 173  * Trust model prevents malicious node influence
174 174  
175 -
176 176  == 6. What happens when new evidence contradicts an existing verdict? ==
177 177  
178 178  FactHarbor is designed for evolving knowledge:
179 179  
180 180  **Automatic Re-evaluation**:
196 +
181 181  1. New evidence arrives
182 182  2. System detects affected scenarios and verdicts
183 183  3. AKEL proposes updated verdicts
... ... @@ -186,26 +186,29 @@
186 186  6. Old versions remain accessible
187 187  
188 188  **Version History**:
205 +
189 189  * Every verdict has complete history
190 190  * Users can see "as of date X, what did we know?"
191 191  * Timeline shows how understanding evolved
192 192  
193 193  **Transparent Updates**:
211 +
194 194  * Reason for re-evaluation documented
195 195  * New evidence clearly linked
196 196  * Changes explained, not hidden
197 197  
198 198  **User Notifications**:
217 +
199 199  * Users following claims are notified of updates
200 200  * Can compare old vs new verdicts
201 201  * Can see which evidence changed conclusions
202 202  
203 -
204 204  == 7. Who can submit claims to FactHarbor? ==
205 205  
206 206  **Anyone** - even without login:
207 207  
208 208  **Readers** (no login required):
227 +
209 209  * Browse and search all published content
210 210  * Submit text for analysis
211 211  * New claims added automatically unless duplicates exist
... ... @@ -212,6 +212,7 @@
212 212  * System deduplicates and normalizes
213 213  
214 214  **Contributors** (logged in):
234 +
215 215  * Everything Readers can do
216 216  * Submit evidence sources
217 217  * Suggest scenarios
... ... @@ -218,6 +218,7 @@
218 218  * Participate in discussions
219 219  
220 220  **Workflow**:
241 +
221 221  1. User submits text (as Reader or Contributor)
222 222  2. AKEL extracts claims
223 223  3. Checks for existing duplicates
... ... @@ -227,12 +227,12 @@
227 227  7. Runs quality gates
228 228  8. Publishes as AI-Generated (Mode 2) if passes
229 229  
230 -
231 231  == 8. What are "risk tiers" and why do they matter? ==
232 232  
233 233  Risk tiers determine review requirements and publication workflow:
234 234  
235 235  **Tier A (High Risk)**:
256 +
236 236  * **Domains**: Medical, legal, elections, safety, security, major financial
237 237  * **Publication**: AI can publish with warnings, expert review required for "Human-Reviewed" status
238 238  * **Audit rate**: Recommendation 30-50%
... ... @@ -239,6 +239,7 @@
239 239  * **Why**: Potential for significant harm if wrong
240 240  
241 241  **Tier B (Medium Risk)**:
263 +
242 242  * **Domains**: Complex policy, science causality, contested issues
243 243  * **Publication**: AI can publish immediately with clear labeling
244 244  * **Audit rate**: Recommendation 10-20%
... ... @@ -245,6 +245,7 @@
245 245  * **Why**: Nuanced but lower immediate harm risk
246 246  
247 247  **Tier C (Low Risk)**:
270 +
248 248  * **Domains**: Definitions, established facts, historical data
249 249  * **Publication**: AI publication default
250 250  * **Audit rate**: Recommendation 5-10%
... ... @@ -251,14 +251,15 @@
251 251  * **Why**: Well-established, low controversy
252 252  
253 253  **Assignment**:
277 +
254 254  * AKEL suggests tier based on domain, keywords, impact
255 255  * Moderators and Experts can override
256 256  * Risk tiers reviewed based on audit outcomes
257 257  
258 -
259 259  == 9. How does federation work and why is it important? ==
260 260  
261 261  **Federation Model**:
285 +
262 262  * Multiple independent FactHarbor nodes
263 263  * Each node has own database, AKEL, governance
264 264  * Nodes exchange claims, scenarios, evidence, verdicts
... ... @@ -265,6 +265,7 @@
265 265  * No central authority
266 266  
267 267  **Why Federation Matters**:
292 +
268 268  * **Resilience**: No single point of failure or censorship
269 269  * **Autonomy**: Communities govern themselves
270 270  * **Scalability**: Add nodes to handle more users
... ... @@ -272,6 +272,7 @@
272 272  * **Trust diversity**: Multiple perspectives, not single truth source
273 273  
274 274  **How Nodes Exchange Data**:
300 +
275 275  1. Local node creates versions
276 276  2. Builds signed bundle
277 277  3. Pushes to trusted neighbor nodes
... ... @@ -280,26 +280,29 @@
280 280  6. Local re-evaluation if needed
281 281  
282 282  **Trust Model**:
309 +
283 283  * Trusted nodes → auto-import
284 284  * Neutral nodes → import with review
285 285  * Untrusted nodes → manual only
286 286  
287 -
288 288  == 10. Can experts disagree in FactHarbor? ==
289 289  
290 290  **Yes - and that's a feature, not a bug**:
291 291  
292 292  **Multiple Scenarios**:
319 +
293 293  * Experts can create different scenarios with different assumptions
294 294  * Each scenario gets its own verdict
295 295  * Users see *why* experts disagree (different definitions, boundaries, evidence weighting)
296 296  
297 297  **Parallel Verdicts**:
325 +
298 298  * Same scenario, different expert interpretations
299 299  * Both verdicts visible with expert attribution
300 300  * No forced consensus
301 301  
302 302  **Transparency**:
331 +
303 303  * Expert reasoning documented
304 304  * Assumptions stated explicitly
305 305  * Evidence chains traceable
... ... @@ -306,16 +306,17 @@
306 306  * Users can evaluate competing expert opinions
307 307  
308 308  **Federation**:
338 +
309 309  * Different nodes can have different expert conclusions
310 310  * Cross-node branching allowed
311 311  * Users can see how conclusions vary across nodes
312 312  
313 -
314 314  == 11. What prevents AI from hallucinating or making up facts? ==
315 315  
316 316  **Multiple Safeguards**:
317 317  
318 318  **Quality Gate 4: Structural Integrity**:
348 +
319 319  * Fact-checking against sources
320 320  * No hallucinations allowed
321 321  * Logic chain must be valid and traceable
... ... @@ -322,6 +322,7 @@
322 322  * References must be accessible and verifiable
323 323  
324 324  **Evidence Requirements**:
355 +
325 325  * Primary sources required
326 326  * Citations must be complete
327 327  * Sources must be accessible
... ... @@ -328,11 +328,13 @@
328 328  * Reliability scored
329 329  
330 330  **Audit System**:
362 +
331 331  * Human auditors check AI-generated content
332 332  * Hallucinations caught and fed back into training
333 333  * Patterns of errors trigger system improvements
334 334  
335 335  **Transparency**:
368 +
336 336  * All reasoning chains visible
337 337  * Sources linked
338 338  * Users can verify claims against sources
... ... @@ -339,16 +339,17 @@
339 339  * AKEL outputs clearly labeled
340 340  
341 341  **Human Oversight**:
375 +
342 342  * Tier A requires expert review for "Human-Reviewed" status
343 343  * Audit sampling catches errors
344 344  * Community can flag issues
345 345  
346 -
347 347  == 12. How does FactHarbor make money / is it sustainable? ==
348 348  
349 349  [ToDo: Business model and sustainability to be defined]
350 350  
351 351  Potential models under consideration:
385 +
352 352  * Non-profit foundation with grants and donations
353 353  * Institutional subscriptions (universities, research organizations, media)
354 354  * API access for third-party integrations
... ... @@ -361,12 +361,11 @@
361 361  == 13. Related Pages ==
362 362  
363 363  * [[Requirements (Roles)>>FactHarbor.Specification.Requirements.WebHome]]
364 -* [[AKEL (AI Knowledge Extraction Layer)>>FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]]
365 -* [[Automation>>FactHarbor.Specification.Automation.WebHome]]
366 -* [[Federation & Decentralization>>FactHarbor.Specification.Federation & Decentralization.WebHome]]
367 -* [[Mission & Purpose>>FactHarbor.Organisation.Core Problems FactHarbor Solves.WebHome]]
398 +* [[AKEL (AI Knowledge Extraction Layer)>>Archive.FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]]
399 +* [[Automation>>Archive.FactHarbor.Specification.Automation.WebHome]]
400 +* [[Federation & Decentralization>>Archive.FactHarbor.Specification.Federation & Decentralization.WebHome]]
401 +* [[Mission & Purpose>>Archive.FactHarbor.Organisation.Core Problems FactHarbor Solves.WebHome]]
368 368  
369 -
370 370  == 20. Glossary / Key Terms ==
371 371  
372 372  === Phase 0 vs POC v1 ===
... ... @@ -381,6 +381,7 @@
381 381  === Beta 0 ===
382 382  
383 383  The next development stage after POC, featuring:
417 +
384 384  * External testers
385 385  * Basic federation experiments
386 386  * Enhanced automation
... ... @@ -388,7 +388,7 @@
388 388  === Release 1.0 ===
389 389  
390 390  The first public release featuring:
425 +
391 391  * Full federation support
392 392  * 2000+ concurrent users
393 393  * Production-grade infrastructure
394 -