Skip to Content
Last modified by Robert Schaub on 2025/12/24 20:33
From version 6.3
edited by Robert Schaub
on 2025/12/16 20:26
Change comment: Update document after refactoring.
To version 6.9
edited by Robert Schaub
on 2025/12/16 21:50
Change comment: Renamed back-links.

Summary

Details

Page properties
Content
... ... @@ -50,6 +50,7 @@
50 50  == Inputs and Outputs ==
51 51  
52 52  === Inputs ===
53 +
53 53  * User-submitted claims or evidence
54 54  * Uploaded documents
55 55  * URLs or citations
... ... @@ -57,6 +57,7 @@
57 57  * Embeddings (from local or federated peers)
58 58  
59 59  === Outputs (publication mode varies by risk tier) ===
61 +
60 60  * ClaimVersion (draft or AI-generated)
61 61  * ScenarioVersion (draft or AI-generated)
62 62  * EvidenceVersion (summary + metadata, draft or AI-generated)
... ... @@ -75,6 +75,7 @@
75 75  === Mode 1: Draft-Only (Never Public) ===
76 76  
77 77  **Used for:**
80 +
78 78  * Failed quality gate checks
79 79  * Sensitive topics flagged for expert review
80 80  * Unclear scope or missing critical sources
... ... @@ -85,6 +85,7 @@
85 85  === Mode 2: Published as AI-Generated (No Prior Human Review) ===
86 86  
87 87  **Requirements:**
91 +
88 88  * All automated quality gates passed (see below)
89 89  * Risk tier permits AI-draft publication (Tier B or C)
90 90  * Contradiction search completed successfully
... ... @@ -98,6 +98,7 @@
98 98  ```
99 99  
100 100  **User actions:**
105 +
101 101  * Browse and read content
102 102  * Request human review (escalates to review queue)
103 103  * Flag for expert attention
... ... @@ -105,6 +105,7 @@
105 105  === Mode 3: Published as Human-Reviewed ===
106 106  
107 107  **Requirements:**
113 +
108 108  * Human reviewer or domain expert validated
109 109  * All quality gates passed
110 110  * Visible "Human-Reviewed" mark with reviewer role and timestamp
... ... @@ -127,12 +127,13 @@
127 127  **Domains:** Medical, legal, elections, safety/security, major reputational harm
128 128  
129 129  **Publication policy:**
136 +
130 130  * Human review REQUIRED before "Human-Reviewed" status
131 131  * AI-generated content MAY be published but:
132 - ** Clearly flagged as AI-draft with prominent disclaimer
133 - ** May have limited visibility
134 - ** Auto-escalated to expert review queue
135 - ** User warnings displayed
139 +** Clearly flagged as AI-draft with prominent disclaimer
140 +** May have limited visibility
141 +** Auto-escalated to expert review queue
142 +** User warnings displayed
136 136  
137 137  **Audit rate:** Recommendation: 30-50% of published AI-drafts sampled in first 6 months
138 138  
... ... @@ -141,6 +141,7 @@
141 141  **Domains:** Contested public policy, complex science, causality claims, significant financial impact
142 142  
143 143  **Publication policy:**
151 +
144 144  * AI-draft CAN publish immediately with clear labeling
145 145  * Sampling audits conducted (see Audit System below)
146 146  * High-engagement items auto-escalated to expert review
... ... @@ -153,6 +153,7 @@
153 153  **Domains:** Definitions, simple factual lookups with strong primary sources, historical facts, established scientific consensus
154 154  
155 155  **Publication policy:**
164 +
156 156  * AI-draft default publication mode
157 157  * Sampling audits sufficient
158 158  * Community flagging available
... ... @@ -167,6 +167,7 @@
167 167  All AI-generated content must pass these automated checks before Mode 2 publication:
168 168  
169 169  === Gate 1: Source Quality ===
179 +
170 170  * Primary sources identified and accessible
171 171  * Source reliability scored against whitelist
172 172  * Citation completeness verified
... ... @@ -183,6 +183,7 @@
183 183  * **Bubble detection** – Conspiracy theories, echo chambers, ideologically isolated sources
184 184  
185 185  **Search coverage requirements:**
196 +
186 186  * Academic literature (BOTH supporting AND opposing views)
187 187  * Reputable media across diverse political/ideological perspectives
188 188  * Official contradictions (retractions, corrections, updates, amendments)
... ... @@ -190,6 +190,7 @@
190 190  * Cross-cultural and international perspectives
191 191  
192 192  **Search must actively avoid algorithmic bubbles:**
204 +
193 193  * Deliberately seek opposing viewpoints
194 194  * Check for echo chamber patterns in source clusters
195 195  * Identify tribal or ideological source clustering
... ... @@ -197,6 +197,7 @@
197 197  * Verify diversity of perspectives represented
198 198  
199 199  **Outcomes:**
212 +
200 200  * **Strong counter-evidence found** → Auto-escalate to Tier B or draft-only mode
201 201  * **Significant uncertainty detected** → Require uncertainty disclosure in verdict
202 202  * **Bubble indicators present** → Flag for expert review and human validation
... ... @@ -203,6 +203,7 @@
203 203  * **Limited perspective diversity** → Expand search or flag for human review
204 204  
205 205  === Gate 3: Uncertainty Quantification ===
219 +
206 206  * Confidence scores calculated for all claims and verdicts
207 207  * Limitations explicitly stated
208 208  * Data gaps identified and disclosed
... ... @@ -210,6 +210,7 @@
210 210  * Alternative scenarios considered
211 211  
212 212  === Gate 4: Structural Integrity ===
227 +
213 213  * No hallucinations detected (fact-checking against sources)
214 214  * Logic chain valid and traceable
215 215  * References accessible and verifiable
... ... @@ -217,6 +217,7 @@
217 217  * Premises clearly stated
218 218  
219 219  **If any gate fails:**
235 +
220 220  * Content remains in draft-only mode
221 221  * Failure reason logged
222 222  * Human review required before publication
... ... @@ -231,6 +231,7 @@
231 231  === Sampling Strategy ===
232 232  
233 233  Audits prioritize:
250 +
234 234  * **Risk tier** (higher tiers get more frequent audits)
235 235  * **AI confidence score** (low confidence → higher sampling rate)
236 236  * **Traffic and engagement** (high-visibility content audited more)
... ... @@ -242,11 +242,12 @@
242 242  1. System selects content for audit based on sampling strategy
243 243  2. Human auditor reviews AI-generated content against quality standards
244 244  3. Auditor validates or corrects:
245 - * Claim extraction accuracy
246 - * Scenario appropriateness
247 - * Evidence relevance and interpretation
248 - * Verdict reasoning
249 - * Contradiction search completeness
262 +
263 +* Claim extraction accuracy
264 +* Scenario appropriateness
265 +* Evidence relevance and interpretation
266 +* Verdict reasoning
267 +* Contradiction search completeness
250 250  4. Audit outcome recorded (pass/fail + detailed feedback)
251 251  5. Failed audits trigger immediate content review
252 252  6. Audit results feed back into system improvement
... ... @@ -254,6 +254,7 @@
254 254  === Feedback Loop (Continuous Improvement) ===
255 255  
256 256  Audit outcomes systematically improve:
275 +
257 257  * **Query templates** – Refined based on missed evidence patterns
258 258  * **Retrieval source weights** – Adjusted for accuracy and reliability
259 259  * **Contradiction detection heuristics** – Enhanced to catch missed counter-evidence
... ... @@ -272,7 +272,7 @@
272 272  
273 273  == Architecture Overview ==
274 274  
275 -{{include reference="Test.FactHarborV09.Specification.Diagrams.AKEL Architecture.WebHome"/}}
294 +{{include reference="FactHarbor.Archive.FactHarbor V0\.9\.23 Lost Data.Specification.Diagrams.AKEL Architecture.WebHome"/}}
276 276  
277 277  ----
278 278  
... ... @@ -327,8 +327,7 @@
327 327  
328 328  == Related Pages ==
329 329  
330 -* [[Automation>>FactHarbor.Specification.Automation.WebHome]]
331 -* [[Requirements (Roles)>>FactHarbor.Specification.Requirements.WebHome]]
332 -* [[Workflows>>FactHarbor.Specification.Workflows.WebHome]]
349 +* [[Automation>>FactHarbor.Archive.FactHarbor V0\.9\.18 copy.Specification.Automation.WebHome]]
350 +* [[Requirements (Roles)>>FactHarbor.Archive.FactHarbor V0\.9\.18 copy.Specification.Requirements.WebHome]]
351 +* [[Workflows>>FactHarbor.Specification V0\.9\.18.Workflows.WebHome]]
333 333  * [[Governance>>FactHarbor.Organisation.Governance]]
334 -