Wiki source code of FAQ
Version 3.1 by Robert Schaub on 2025/12/15 16:56
Show last authors
| author | version | line-number | content |
|---|---|---|---|
| 1 | = Frequently Asked Questions (FAQ) = | ||
| 2 | |||
| 3 | Common questions about FactHarbor's design, functionality, and approach. | ||
| 4 | |||
| 5 | ---- | ||
| 6 | |||
| 7 | == How do facts get input into the system? == | ||
| 8 | |||
| 9 | FactHarbor uses a hybrid model combining three complementary approaches: | ||
| 10 | |||
| 11 | === 1. AI-Generated Content (Scalable) === | ||
| 12 | |||
| 13 | **What**: System dynamically researches claims using AKEL (AI Knowledge Extraction Layer) | ||
| 14 | |||
| 15 | **Process**: | ||
| 16 | * Extracts claims from submitted text | ||
| 17 | * Generates structured sub-queries | ||
| 18 | * Performs **mandatory contradiction search** (actively seeks counter-evidence, not just confirmations) | ||
| 19 | * Runs automated quality gates | ||
| 20 | * Publishes with clear "AI-Generated" labels | ||
| 21 | |||
| 22 | **Publication**: Mode 2 (public, AI-labeled) when quality gates pass | ||
| 23 | |||
| 24 | **Purpose**: Handles scale — emerging claims get immediate responses with transparent reasoning | ||
| 25 | |||
| 26 | === 2. Expert-Authored Content (Authoritative) === | ||
| 27 | |||
| 28 | **What**: Domain experts directly author, edit, and validate content | ||
| 29 | |||
| 30 | **Focus**: High-risk domains (medical, legal, safety-critical) | ||
| 31 | |||
| 32 | **Publication**: Mode 3 ("Human-Reviewed" status) with expert attribution | ||
| 33 | |||
| 34 | **Authority**: Tier A content requires expert approval | ||
| 35 | |||
| 36 | **Purpose**: Provides authoritative grounding for critical domains where errors have serious consequences | ||
| 37 | |||
| 38 | === 3. Audit-Improved Quality (Continuous) === | ||
| 39 | |||
| 40 | **What**: Sampling audits where experts review AI-generated content | ||
| 41 | |||
| 42 | **Rates**: | ||
| 43 | * High-risk (Tier A): 30-50% sampling | ||
| 44 | * Medium-risk (Tier B): 10-20% sampling | ||
| 45 | * Low-risk (Tier C): 5-10% sampling | ||
| 46 | |||
| 47 | **Impact**: Expert feedback systematically improves AI research quality | ||
| 48 | |||
| 49 | **Purpose**: Ensures AI quality evolves based on expert validation patterns | ||
| 50 | |||
| 51 | === Why All Three Matter === | ||
| 52 | |||
| 53 | **Complementary Strengths**: | ||
| 54 | * **AI research**: Scale and speed for emerging claims | ||
| 55 | * **Expert authoring**: Authority and precision for critical domains | ||
| 56 | * **Audit feedback**: Continuous quality improvement | ||
| 57 | |||
| 58 | **Expert Time Optimization**: | ||
| 59 | |||
| 60 | Experts can choose where to focus their time: | ||
| 61 | * Author high-priority content directly | ||
| 62 | * Validate and edit AI-generated outputs | ||
| 63 | * Audit samples to improve system-wide AI performance | ||
| 64 | |||
| 65 | This focuses expert time where domain expertise matters most while leveraging AI for scale. | ||
| 66 | |||
| 67 | === Current Status === | ||
| 68 | |||
| 69 | **POC v1**: Demonstrates the AI research pipeline (fully automated with transparent reasoning and quality gates) | ||
| 70 | |||
| 71 | **Full System**: Will support all three pathways with integrated workflow | ||
| 72 | |||
| 73 | ---- | ||
| 74 | |||
| 75 | == What prevents FactHarbor from becoming another echo chamber? == | ||
| 76 | |||
| 77 | FactHarbor includes multiple safeguards against echo chambers and filter bubbles: | ||
| 78 | |||
| 79 | **Mandatory Contradiction Search**: | ||
| 80 | * AI must actively search for counter-evidence, not just confirmations | ||
| 81 | * System checks for echo chamber patterns in source clusters | ||
| 82 | * Flags tribal or ideological source clustering | ||
| 83 | * Requires diverse perspectives across political/ideological spectrum | ||
| 84 | |||
| 85 | **Multiple Scenarios**: | ||
| 86 | * Claims are evaluated under different interpretations | ||
| 87 | * Reveals how assumptions change conclusions | ||
| 88 | * Makes disagreements understandable, not divisive | ||
| 89 | |||
| 90 | **Transparent Reasoning**: | ||
| 91 | * All assumptions, definitions, and boundaries are explicit | ||
| 92 | * Evidence chains are traceable | ||
| 93 | * Uncertainty is quantified, not hidden | ||
| 94 | |||
| 95 | **Audit System**: | ||
| 96 | * Human auditors check for bubble patterns | ||
| 97 | * Feedback loop improves AI search diversity | ||
| 98 | * Community can flag missing perspectives | ||
| 99 | |||
| 100 | **Federation**: | ||
| 101 | * Multiple independent nodes with different perspectives | ||
| 102 | * No single entity controls "the truth" | ||
| 103 | * Cross-node contradiction detection | ||
| 104 | |||
| 105 | ---- | ||
| 106 | |||
| 107 | == How does FactHarbor handle claims that are "true in one context but false in another"? == | ||
| 108 | |||
| 109 | This is exactly what FactHarbor is designed for: | ||
| 110 | |||
| 111 | **Scenarios capture contexts**: | ||
| 112 | * Each scenario defines specific boundaries, definitions, and assumptions | ||
| 113 | * The same claim can have different verdicts in different scenarios | ||
| 114 | * Example: "Coffee is healthy" depends on: | ||
| 115 | ** Definition of "healthy" (reduces disease risk? improves mood? affects specific conditions?) | ||
| 116 | ** Population (adults? pregnant women? people with heart conditions?) | ||
| 117 | ** Consumption level (1 cup/day? 5 cups/day?) | ||
| 118 | ** Time horizon (short-term? long-term?) | ||
| 119 | |||
| 120 | **Truth Landscape**: | ||
| 121 | * Shows all scenarios and their verdicts side-by-side | ||
| 122 | * Users see *why* interpretations differ | ||
| 123 | * No forced consensus when legitimate disagreement exists | ||
| 124 | |||
| 125 | **Explicit Assumptions**: | ||
| 126 | * Every scenario states its assumptions clearly | ||
| 127 | * Users can compare how changing assumptions changes conclusions | ||
| 128 | * Makes context-dependence visible, not hidden | ||
| 129 | |||
| 130 | ---- | ||
| 131 | |||
| 132 | == What makes FactHarbor different from traditional fact-checking sites? == | ||
| 133 | |||
| 134 | **Traditional Fact-Checking**: | ||
| 135 | * Binary verdicts: True / Mostly True / False | ||
| 136 | * Single interpretation chosen by fact-checker | ||
| 137 | * Often hides legitimate contextual differences | ||
| 138 | * Limited ability to show *why* people disagree | ||
| 139 | |||
| 140 | **FactHarbor**: | ||
| 141 | * **Multi-scenario**: Shows multiple valid interpretations | ||
| 142 | * **Likelihood-based**: Ranges with uncertainty, not binary labels | ||
| 143 | * **Transparent assumptions**: Makes boundaries and definitions explicit | ||
| 144 | * **Version history**: Shows how understanding evolves | ||
| 145 | * **Contradiction search**: Actively seeks opposing evidence | ||
| 146 | * **Federated**: No single authority controls truth | ||
| 147 | |||
| 148 | ---- | ||
| 149 | |||
| 150 | == How do you prevent manipulation or coordinated misinformation campaigns? == | ||
| 151 | |||
| 152 | **Quality Gates**: | ||
| 153 | * Automated checks before AI-generated content publishes | ||
| 154 | * Source quality verification | ||
| 155 | * Mandatory contradiction search | ||
| 156 | * Bubble detection for coordinated campaigns | ||
| 157 | |||
| 158 | **Audit System**: | ||
| 159 | * Stratified sampling catches manipulation patterns | ||
| 160 | * Expert auditors validate AI research quality | ||
| 161 | * Failed audits trigger immediate review | ||
| 162 | |||
| 163 | **Transparency**: | ||
| 164 | * All reasoning chains are visible | ||
| 165 | * Evidence sources are traceable | ||
| 166 | * AKEL involvement clearly labeled | ||
| 167 | * Version history preserved | ||
| 168 | |||
| 169 | **Moderation**: | ||
| 170 | * Moderators handle abuse, spam, coordinated manipulation | ||
| 171 | * Content can be flagged by community | ||
| 172 | * Audit trail maintained even if content hidden | ||
| 173 | |||
| 174 | **Federation**: | ||
| 175 | * Multiple nodes with independent governance | ||
| 176 | * No single point of control | ||
| 177 | * Cross-node contradiction detection | ||
| 178 | * Trust model prevents malicious node influence | ||
| 179 | |||
| 180 | ---- | ||
| 181 | |||
| 182 | == What happens when new evidence contradicts an existing verdict? == | ||
| 183 | |||
| 184 | FactHarbor is designed for evolving knowledge: | ||
| 185 | |||
| 186 | **Automatic Re-evaluation**: | ||
| 187 | 1. New evidence arrives | ||
| 188 | 2. System detects affected scenarios and verdicts | ||
| 189 | 3. AKEL proposes updated verdicts | ||
| 190 | 4. Reviewers/experts validate | ||
| 191 | 5. New verdict version published | ||
| 192 | 6. Old versions remain accessible | ||
| 193 | |||
| 194 | **Version History**: | ||
| 195 | * Every verdict has complete history | ||
| 196 | * Users can see "as of date X, what did we know?" | ||
| 197 | * Timeline shows how understanding evolved | ||
| 198 | |||
| 199 | **Transparent Updates**: | ||
| 200 | * Reason for re-evaluation documented | ||
| 201 | * New evidence clearly linked | ||
| 202 | * Changes explained, not hidden | ||
| 203 | |||
| 204 | **User Notifications**: | ||
| 205 | * Users following claims are notified of updates | ||
| 206 | * Can compare old vs new verdicts | ||
| 207 | * Can see which evidence changed conclusions | ||
| 208 | |||
| 209 | ---- | ||
| 210 | |||
| 211 | == Who can submit claims to FactHarbor? == | ||
| 212 | |||
| 213 | **Anyone** - even without login: | ||
| 214 | |||
| 215 | **Readers** (no login required): | ||
| 216 | * Browse and search all published content | ||
| 217 | * Submit text for analysis | ||
| 218 | * New claims added automatically unless duplicates exist | ||
| 219 | * System deduplicates and normalizes | ||
| 220 | |||
| 221 | **Contributors** (logged in): | ||
| 222 | * Everything Readers can do | ||
| 223 | * Submit evidence sources | ||
| 224 | * Suggest scenarios | ||
| 225 | * Participate in discussions | ||
| 226 | |||
| 227 | **Workflow**: | ||
| 228 | 1. User submits text (as Reader or Contributor) | ||
| 229 | 2. AKEL extracts claims | ||
| 230 | 3. Checks for existing duplicates | ||
| 231 | 4. Normalizes claim text | ||
| 232 | 5. Assigns risk tier | ||
| 233 | 6. Generates scenarios (draft) | ||
| 234 | 7. Runs quality gates | ||
| 235 | 8. Publishes as AI-Generated (Mode 2) if passes | ||
| 236 | |||
| 237 | ---- | ||
| 238 | |||
| 239 | == What are "risk tiers" and why do they matter? == | ||
| 240 | |||
| 241 | Risk tiers determine review requirements and publication workflow: | ||
| 242 | |||
| 243 | **Tier A (High Risk)**: | ||
| 244 | * **Domains**: Medical, legal, elections, safety, security, major financial | ||
| 245 | * **Publication**: AI can publish with warnings, expert review required for "Human-Reviewed" status | ||
| 246 | * **Audit rate**: Recommendation 30-50% | ||
| 247 | * **Why**: Potential for significant harm if wrong | ||
| 248 | |||
| 249 | **Tier B (Medium Risk)**: | ||
| 250 | * **Domains**: Complex policy, science causality, contested issues | ||
| 251 | * **Publication**: AI can publish immediately with clear labeling | ||
| 252 | * **Audit rate**: Recommendation 10-20% | ||
| 253 | * **Why**: Nuanced but lower immediate harm risk | ||
| 254 | |||
| 255 | **Tier C (Low Risk)**: | ||
| 256 | * **Domains**: Definitions, established facts, historical data | ||
| 257 | * **Publication**: AI publication default | ||
| 258 | * **Audit rate**: Recommendation 5-10% | ||
| 259 | * **Why**: Well-established, low controversy | ||
| 260 | |||
| 261 | **Assignment**: | ||
| 262 | * AKEL suggests tier based on domain, keywords, impact | ||
| 263 | * Moderators and Experts can override | ||
| 264 | * Risk tiers reviewed based on audit outcomes | ||
| 265 | |||
| 266 | ---- | ||
| 267 | |||
| 268 | == How does federation work and why is it important? == | ||
| 269 | |||
| 270 | **Federation Model**: | ||
| 271 | * Multiple independent FactHarbor nodes | ||
| 272 | * Each node has own database, AKEL, governance | ||
| 273 | * Nodes exchange claims, scenarios, evidence, verdicts | ||
| 274 | * No central authority | ||
| 275 | |||
| 276 | **Why Federation Matters**: | ||
| 277 | * **Resilience**: No single point of failure or censorship | ||
| 278 | * **Autonomy**: Communities govern themselves | ||
| 279 | * **Scalability**: Add nodes to handle more users | ||
| 280 | * **Specialization**: Domain-focused nodes (health, energy, etc.) | ||
| 281 | * **Trust diversity**: Multiple perspectives, not single truth source | ||
| 282 | |||
| 283 | **How Nodes Exchange Data**: | ||
| 284 | 1. Local node creates versions | ||
| 285 | 2. Builds signed bundle | ||
| 286 | 3. Pushes to trusted neighbor nodes | ||
| 287 | 4. Remote nodes validate signatures and lineage | ||
| 288 | 5. Accept or branch versions | ||
| 289 | 6. Local re-evaluation if needed | ||
| 290 | |||
| 291 | **Trust Model**: | ||
| 292 | * Trusted nodes → auto-import | ||
| 293 | * Neutral nodes → import with review | ||
| 294 | * Untrusted nodes → manual only | ||
| 295 | |||
| 296 | ---- | ||
| 297 | |||
| 298 | == Can experts disagree in FactHarbor? == | ||
| 299 | |||
| 300 | **Yes - and that's a feature, not a bug**: | ||
| 301 | |||
| 302 | **Multiple Scenarios**: | ||
| 303 | * Experts can create different scenarios with different assumptions | ||
| 304 | * Each scenario gets its own verdict | ||
| 305 | * Users see *why* experts disagree (different definitions, boundaries, evidence weighting) | ||
| 306 | |||
| 307 | **Parallel Verdicts**: | ||
| 308 | * Same scenario, different expert interpretations | ||
| 309 | * Both verdicts visible with expert attribution | ||
| 310 | * No forced consensus | ||
| 311 | |||
| 312 | **Transparency**: | ||
| 313 | * Expert reasoning documented | ||
| 314 | * Assumptions stated explicitly | ||
| 315 | * Evidence chains traceable | ||
| 316 | * Users can evaluate competing expert opinions | ||
| 317 | |||
| 318 | **Federation**: | ||
| 319 | * Different nodes can have different expert conclusions | ||
| 320 | * Cross-node branching allowed | ||
| 321 | * Users can see how conclusions vary across nodes | ||
| 322 | |||
| 323 | ---- | ||
| 324 | |||
| 325 | == What prevents AI from hallucinating or making up facts? == | ||
| 326 | |||
| 327 | **Multiple Safeguards**: | ||
| 328 | |||
| 329 | **Quality Gate 4: Structural Integrity**: | ||
| 330 | * Fact-checking against sources | ||
| 331 | * No hallucinations allowed | ||
| 332 | * Logic chain must be valid and traceable | ||
| 333 | * References must be accessible and verifiable | ||
| 334 | |||
| 335 | **Evidence Requirements**: | ||
| 336 | * Primary sources required | ||
| 337 | * Citations must be complete | ||
| 338 | * Sources must be accessible | ||
| 339 | * Reliability scored | ||
| 340 | |||
| 341 | **Audit System**: | ||
| 342 | * Human auditors check AI-generated content | ||
| 343 | * Hallucinations caught and fed back into training | ||
| 344 | * Patterns of errors trigger system improvements | ||
| 345 | |||
| 346 | **Transparency**: | ||
| 347 | * All reasoning chains visible | ||
| 348 | * Sources linked | ||
| 349 | * Users can verify claims against sources | ||
| 350 | * AKEL outputs clearly labeled | ||
| 351 | |||
| 352 | **Human Oversight**: | ||
| 353 | * Tier A requires expert review for "Human-Reviewed" status | ||
| 354 | * Audit sampling catches errors | ||
| 355 | * Community can flag issues | ||
| 356 | |||
| 357 | ---- | ||
| 358 | |||
| 359 | == How does FactHarbor make money / is it sustainable? == | ||
| 360 | |||
| 361 | [ToDo: Business model and sustainability to be defined] | ||
| 362 | |||
| 363 | Potential models under consideration: | ||
| 364 | * Non-profit foundation with grants and donations | ||
| 365 | * Institutional subscriptions (universities, research organizations, media) | ||
| 366 | * API access for third-party integrations | ||
| 367 | * Premium features for power users | ||
| 368 | * Federated node hosting services | ||
| 369 | |||
| 370 | Core principle: **Public benefit** mission takes priority over profit. | ||
| 371 | |||
| 372 | ---- | ||
| 373 | |||
| 374 | == Related Pages == | ||
| 375 | |||
| 376 | * [[Requirements (Roles)>>FactHarbor.Specification.Requirements.WebHome]] | ||
| 377 | * [[AKEL (AI Knowledge Extraction Layer)>>FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]] | ||
| 378 | * [[Automation>>FactHarbor.Specification.Automation.WebHome]] | ||
| 379 | * [[Federation & Decentralization>>FactHarbor.Specification.Federation & Decentralization.WebHome]] | ||
| 380 | * [[Mission & Purpose>>FactHarbor.Organisation.Core Problems FactHarbor Solves.WebHome]] |