Skip to Content
Last modified by Robert Schaub on 2025/12/24 20:33
From version 1.2
edited by Robert Schaub
on 2025/12/16 20:40
Change comment: Update document after refactoring.
To version 1.3
edited by Robert Schaub
on 2025/12/16 20:40
Change comment: Renamed back-links.

Summary

Details

Page properties
Content
... ... @@ -29,7 +29,6 @@
29 29  * Re-evaluation proposal generation
30 30  * Cross-node embedding exchange (Release 1.0+)
31 31  
32 -
33 33  == 2. Components ==
34 34  
35 35  * **AKEL Orchestrator** – central coordinator
... ... @@ -43,10 +43,10 @@
43 43  * **Embedding Handler** (Release 1.0+)
44 44  * **Federation Sync Adapter** (Release 1.0+)
45 45  
46 -
47 47  == 3. Inputs and Outputs ==
48 48  
49 49  === 3.1 Inputs ===
48 +
50 50  * User-submitted claims or evidence
51 51  * Uploaded documents
52 52  * URLs or citations
... ... @@ -54,6 +54,7 @@
54 54  * Embeddings (from local or federated peers)
55 55  
56 56  === 3.2 Outputs (publication mode varies by risk tier) ===
56 +
57 57  * ClaimVersion (draft or AI-generated)
58 58  * ScenarioVersion (draft or AI-generated)
59 59  * EvidenceVersion (summary + metadata, draft or AI-generated)
... ... @@ -63,7 +63,6 @@
63 63  * Re-evaluation proposals
64 64  * Updated embeddings
65 65  
66 -
67 67  == 4. Publication Modes ==
68 68  
69 69  AKEL content is published according to three modes:
... ... @@ -71,6 +71,7 @@
71 71  === 4.1 Mode 1: Draft-Only (Never Public) ===
72 72  
73 73  **Used for:**
73 +
74 74  * Failed quality gate checks
75 75  * Sensitive topics flagged for expert review
76 76  * Unclear scope or missing critical sources
... ... @@ -81,6 +81,7 @@
81 81  === 4.2 Mode 2: Published as AI-Generated (No Prior Human Review) ===
82 82  
83 83  **Requirements:**
84 +
84 84  * All automated quality gates passed (see below)
85 85  * Risk tier permits AI-draft publication (Tier B or C)
86 86  * Contradiction search completed successfully
... ... @@ -94,6 +94,7 @@
94 94  ```
95 95  
96 96  **User actions:**
98 +
97 97  * Browse and read content
98 98  * Request human review (escalates to review queue)
99 99  * Flag for expert attention
... ... @@ -101,6 +101,7 @@
101 101  === 4.3 Mode 3: Published as Human-Reviewed ===
102 102  
103 103  **Requirements:**
106 +
104 104  * Human reviewer or domain expert validated
105 105  * All quality gates passed
106 106  * Visible "Human-Reviewed" mark with reviewer role and timestamp
... ... @@ -122,12 +122,13 @@
122 122  **Domains:** Medical, legal, elections, safety/security, major reputational harm
123 123  
124 124  **Publication policy:**
128 +
125 125  * Human review REQUIRED before "Human-Reviewed" status
126 126  * AI-generated content MAY be published but:
127 - ** Clearly flagged as AI-draft with prominent disclaimer
128 - ** May have limited visibility
129 - ** Auto-escalated to expert review queue
130 - ** User warnings displayed
131 +** Clearly flagged as AI-draft with prominent disclaimer
132 +** May have limited visibility
133 +** Auto-escalated to expert review queue
134 +** User warnings displayed
131 131  
132 132  **Audit rate:** Recommendation: 30-50% of published AI-drafts sampled in first 6 months
133 133  
... ... @@ -136,6 +136,7 @@
136 136  **Domains:** Contested public policy, complex science, causality claims, significant financial impact
137 137  
138 138  **Publication policy:**
143 +
139 139  * AI-draft CAN publish immediately with clear labeling
140 140  * Sampling audits conducted (see Audit System below)
141 141  * High-engagement items auto-escalated to expert review
... ... @@ -148,6 +148,7 @@
148 148  **Domains:** Definitions, simple factual lookups with strong primary sources, historical facts, established scientific consensus
149 149  
150 150  **Publication policy:**
156 +
151 151  * AI-draft default publication mode
152 152  * Sampling audits sufficient
153 153  * Community flagging available
... ... @@ -161,6 +161,7 @@
161 161  All AI-generated content must pass these automated checks before Mode 2 publication:
162 162  
163 163  === 6.1 Gate 1: Source Quality ===
170 +
164 164  * Primary sources identified and accessible
165 165  * Source reliability scored against whitelist
166 166  * Citation completeness verified
... ... @@ -177,6 +177,7 @@
177 177  * **Bubble detection** – Conspiracy theories, echo chambers, ideologically isolated sources
178 178  
179 179  **Search coverage requirements:**
187 +
180 180  * Academic literature (BOTH supporting AND opposing views)
181 181  * Reputable media across diverse political/ideological perspectives
182 182  * Official contradictions (retractions, corrections, updates, amendments)
... ... @@ -184,6 +184,7 @@
184 184  * Cross-cultural and international perspectives
185 185  
186 186  **Search must actively avoid algorithmic bubbles:**
195 +
187 187  * Deliberately seek opposing viewpoints
188 188  * Check for echo chamber patterns in source clusters
189 189  * Identify tribal or ideological source clustering
... ... @@ -191,6 +191,7 @@
191 191  * Verify diversity of perspectives represented
192 192  
193 193  **Outcomes:**
203 +
194 194  * **Strong counter-evidence found** → Auto-escalate to Tier B or draft-only mode
195 195  * **Significant uncertainty detected** → Require uncertainty disclosure in verdict
196 196  * **Bubble indicators present** → Flag for expert review and human validation
... ... @@ -197,6 +197,7 @@
197 197  * **Limited perspective diversity** → Expand search or flag for human review
198 198  
199 199  === 6.3 Gate 3: Uncertainty Quantification ===
210 +
200 200  * Confidence scores calculated for all claims and verdicts
201 201  * Limitations explicitly stated
202 202  * Data gaps identified and disclosed
... ... @@ -204,6 +204,7 @@
204 204  * Alternative scenarios considered
205 205  
206 206  === 6.4 Gate 4: Structural Integrity ===
218 +
207 207  * No hallucinations detected (fact-checking against sources)
208 208  * Logic chain valid and traceable
209 209  * References accessible and verifiable
... ... @@ -211,12 +211,12 @@
211 211  * Premises clearly stated
212 212  
213 213  **If any gate fails:**
226 +
214 214  * Content remains in draft-only mode
215 215  * Failure reason logged
216 216  * Human review required before publication
217 217  * Failure patterns analyzed for system improvement
218 218  
219 -
220 220  == 7. Audit System (Sampling-Based Quality Assurance) ==
221 221  
222 222  Instead of reviewing ALL AI output, FactHarbor implements stratified sampling audits:
... ... @@ -224,6 +224,7 @@
224 224  === 7.1 Sampling Strategy ===
225 225  
226 226  Audits prioritize:
239 +
227 227  * **Risk tier** (higher tiers get more frequent audits)
228 228  * **AI confidence score** (low confidence → higher sampling rate)
229 229  * **Traffic and engagement** (high-visibility content audited more)
... ... @@ -235,11 +235,12 @@
235 235  1. System selects content for audit based on sampling strategy
236 236  2. Human auditor reviews AI-generated content against quality standards
237 237  3. Auditor validates or corrects:
238 - * Claim extraction accuracy
239 - * Scenario appropriateness
240 - * Evidence relevance and interpretation
241 - * Verdict reasoning
242 - * Contradiction search completeness
251 +
252 +* Claim extraction accuracy
253 +* Scenario appropriateness
254 +* Evidence relevance and interpretation
255 +* Verdict reasoning
256 +* Contradiction search completeness
243 243  4. Audit outcome recorded (pass/fail + detailed feedback)
244 244  5. Failed audits trigger immediate content review
245 245  6. Audit results feed back into system improvement
... ... @@ -247,6 +247,7 @@
247 247  === 7.3 Feedback Loop (Continuous Improvement) ===
248 248  
249 249  Audit outcomes systematically improve:
264 +
250 250  * **Query templates** – Refined based on missed evidence patterns
251 251  * **Retrieval source weights** – Adjusted for accuracy and reliability
252 252  * **Contradiction detection heuristics** – Enhanced to catch missed counter-evidence
... ... @@ -261,10 +261,9 @@
261 261  * System improvements documented
262 262  * Community can view aggregate audit performance
263 263  
264 -
265 265  == 8. Architecture Overview ==
266 266  
267 -{{include reference="Test.FactHarborV09.Specification.Diagrams.AKEL Architecture.WebHome"}}
281 +{{include reference="Test.FactHarbor V0\.9\.23 Lost Data.Specification.Diagrams.AKEL Architecture.WebHome"}}
268 268  
269 269  
270 270  == 9. AKEL and Federation ==
... ... @@ -319,4 +319,4 @@
319 319  * [[Requirements (Roles)>>Test.FactHarborV09.Specification.Requirements.WebHome]]
320 320  * [[Workflows>>Test.FactHarborV09.Specification.Workflows.WebHome]]
321 321  * [[Governance>>Test.FactHarborV09.Organisation.Governance]]
322 -
336 +{{/include}}