Skip to Content
Last modified by Robert Schaub on 2026/02/08 21:23
From version 1.2
edited by Robert Schaub
on 2026/01/20 20:20
Change comment: Renamed back-links.
To version 1.1
edited by Robert Schaub
on 2025/12/18 12:03
Change comment: Imported from XAR

Summary

Details

Page properties
Content
... ... @@ -1,18 +1,13 @@
1 1  = AKEL — AI Knowledge Extraction Layer =
2 -
3 3  AKEL is FactHarbor's automated intelligence subsystem.
4 4  Its purpose is to reduce human workload, enhance consistency, and enable scalable knowledge processing — **without ever replacing human judgment**.
5 5  AKEL outputs are marked with **AuthorType = AI** and published according to risk-based review policies (see Publication Modes below).
6 6  AKEL operates in two modes:
7 -
8 8  * **Single-node mode** (POC & Beta 0)
9 9  * **Federated multi-node mode** (Release 1.0+)
10 -
11 11  == 1. Purpose and Role ==
12 -
13 13  AKEL transforms unstructured inputs into structured, publication-ready content.
14 14  Core responsibilities:
15 -
16 16  * Claim extraction from arbitrary text
17 17  * Claim classification (domain, type, evaluability, safety, **risk tier**)
18 18  * Scenario generation (definitions, boundaries, assumptions, methodology)
... ... @@ -22,9 +22,7 @@
22 22  * **Bubble detection** (echo chambers, conspiracy theories, isolated sources)
23 23  * Re-evaluation proposal generation
24 24  * Cross-node embedding exchange (Release 1.0+)
25 -
26 26  == 2. Components ==
27 -
28 28  * **AKEL Orchestrator** – central coordinator
29 29  * **Claim Extractor**
30 30  * **Claim Classifier** (with risk tier assignment)
... ... @@ -35,19 +35,14 @@
35 35  * **Audit Sampling Scheduler**
36 36  * **Embedding Handler** (Release 1.0+)
37 37  * **Federation Sync Adapter** (Release 1.0+)
38 -
39 39  == 3. Inputs and Outputs ==
40 -
41 41  === 3.1 Inputs ===
42 -
43 43  * User-submitted claims or evidence
44 44  * Uploaded documents
45 45  * URLs or citations
46 46  * External LLM API (optional)
47 47  * Embeddings (from local or federated peers)
48 -
49 49  === 3.2 Outputs (publication mode varies by risk tier) ===
50 -
51 51  * ClaimVersion (draft or AI-generated)
52 52  * ScenarioVersion (draft or AI-generated)
53 53  * EvidenceVersion (summary + metadata, draft or AI-generated)
... ... @@ -56,25 +56,17 @@
56 56  * Reservation and limitation notices
57 57  * Re-evaluation proposals
58 58  * Updated embeddings
59 -
60 60  == 4. Publication Modes ==
61 -
62 62  AKEL content is published according to three modes:
63 -
64 64  === 4.1 Mode 1: Draft-Only (Never Public) ===
65 -
66 66  **Used for:**
67 -
68 68  * Failed quality gate checks
69 69  * Sensitive topics flagged for expert review
70 70  * Unclear scope or missing critical sources
71 71  * High reputational risk content
72 72  **Visibility:** Internal review queue only
73 -
74 74  === 4.2 Mode 2: Published as AI-Generated (No Prior Human Review) ===
75 -
76 76  **Requirements:**
77 -
78 78  * All automated quality gates passed (see below)
79 79  * Risk tier permits AI-draft publication (Tier B or C)
80 80  * Contradiction search completed successfully
... ... @@ -89,62 +89,44 @@
89 89  * Browse and read content
90 90  * Request human review (escalates to review queue)
91 91  * Flag for expert attention
92 -
93 93  == 5. Risk tiers ==
94 -
95 95  AKEL assigns risk tiers to all content to determine appropriate review requirements:
96 -
97 97  === 5.1 Tier A — High Risk / High Impact ===
98 -
99 99  **Domains:** Medical, legal, elections, safety/security, major reputational harm
100 100  **Publication policy:**
101 -
102 102  * Human review REQUIRED before "AKEL-Generated" status
103 103  * AI-generated content MAY be published but:
104 -** Clearly flagged as AI-draft with prominent disclaimer
105 -** May have limited visibility
106 -** Auto-escalated to expert review queue
107 -** User warnings displayed
79 + ** Clearly flagged as AI-draft with prominent disclaimer
80 + ** May have limited visibility
81 + ** Auto-escalated to expert review queue
82 + ** User warnings displayed
108 108  **Audit rate:** Recommendation: 30-50% of published AI-drafts sampled in first 6 months
109 -
110 110  === 5.2 Tier B — Medium Risk ===
111 -
112 112  **Domains:** Contested public policy, complex science, causality claims, significant financial impact
113 113  **Publication policy:**
114 -
115 115  * AI-draft CAN publish immediately with clear labeling
116 116  * Sampling audits conducted (see Audit System below)
117 117  * High-engagement items auto-escalated to expert review
118 118  * Users can report issue for moderator review
119 119  **Audit rate:** Recommendation: 10-20% of published AI-drafts sampled
120 -
121 121  === 5.3 Tier C — Low Risk ===
122 -
123 123  **Domains:** Definitions, simple factual lookups with strong primary sources, historical facts, established scientific consensus
124 124  **Publication policy:**
125 -
126 126  * AI-draft default publication mode
127 127  * Sampling audits sufficient
128 128  * Community flagging available
129 129  * Human review on request
130 130  **Audit rate:** Recommendation: 5-10% of published AI-drafts sampled
131 -
132 132  == 6. Quality Gates (Mandatory Before AI-Draft Publication) ==
133 -
134 134  All AI-generated content must pass these automated checks before Mode 2 publication:
135 -
136 136  === 6.1 Gate 1: Source Quality ===
137 -
138 138  * Primary sources identified and accessible
139 139  * Source reliability scored against whitelist
140 140  * Citation completeness verified
141 141  * Publication dates checked
142 142  * Author credentials validated (where applicable)
143 -
144 144  === 6.2 Gate 2: Contradiction Search (MANDATORY) ===
145 -
146 146  **The system MUST actively search for:**
147 -
148 148  * **Counter-evidence** – Rebuttals, conflicting results, contradictory studies
149 149  * **Reservations** – Caveats, limitations, boundary conditions, applicability constraints
150 150  * **Alternative interpretations** – Different framings, definitions, contextual variations
... ... @@ -166,17 +166,13 @@
166 166  * **Significant uncertainty detected** → Require uncertainty disclosure in verdict
167 167  * **Bubble indicators present** → Flag for expert review and human validation
168 168  * **Limited perspective diversity** → Expand search or flag for human review
169 -
170 170  === 6.3 Gate 3: Uncertainty Quantification ===
171 -
172 172  * Confidence scores calculated for all claims and verdicts
173 173  * Limitations explicitly stated
174 174  * Data gaps identified and disclosed
175 175  * Strength of evidence assessed
176 176  * Alternative scenarios considered
177 -
178 178  === 6.4 Gate 4: Structural Integrity ===
179 -
180 180  * No hallucinations detected (fact-checking against sources)
181 181  * Logic chain valid and traceable
182 182  * References accessible and verifiable
... ... @@ -187,40 +187,29 @@
187 187  * Failure reason logged
188 188  * Human review required before publication
189 189  * Failure patterns analyzed for system improvement
190 -
191 191  == 7. Audit System (Sampling-Based Quality Assurance) ==
192 -
193 193  Instead of reviewing ALL AI output, FactHarbor implements stratified sampling audits:
194 -
195 195  === 7.1 Sampling Strategy ===
196 -
197 197  Audits prioritize:
198 -
199 199  * **Risk tier** (higher tiers get more frequent audits)
200 200  * **AI confidence score** (low confidence → higher sampling rate)
201 201  * **Traffic and engagement** (high-visibility content audited more)
202 202  * **Novelty** (new claim types, new domains, emerging topics)
203 203  * **Disagreement signals** (user flags, contradiction alerts, community reports)
204 -
205 205  === 7.2 Audit Process ===
206 -
207 207  1. System selects content for audit based on sampling strategy
208 208  2. Human auditor reviews AI-generated content against quality standards
209 209  3. Moderator validates or corrects:
210 -
211 -* Claim extraction accuracy
212 -* Scenario appropriateness
213 -* Evidence relevance and interpretation
214 -* Verdict reasoning
215 -* Contradiction search completeness
161 + * Claim extraction accuracy
162 + * Scenario appropriateness
163 + * Evidence relevance and interpretation
164 + * Verdict reasoning
165 + * Contradiction search completeness
216 216  4. Audit outcome recorded (pass/fail + detailed feedback)
217 217  5. Failed audits trigger immediate content review
218 218  6. Audit results feed back into system improvement
219 -
220 220  === 7.3 Feedback Loop (Continuous Improvement) ===
221 -
222 222  Audit outcomes systematically improve:
223 -
224 224  * **Query templates** – Refined based on missed evidence patterns
225 225  * **Retrieval source weights** – Adjusted for accuracy and reliability
226 226  * **Contradiction detection heuristics** – Enhanced to catch missed counter-evidence
... ... @@ -227,22 +227,15 @@
227 227  * **Model prompts and extraction rules** – Tuned for better claim extraction
228 228  * **Risk tier assignments** – Recalibrated based on error patterns
229 229  * **Bubble detection algorithms** – Improved to identify echo chambers
230 -
231 231  === 7.4 Audit Transparency ===
232 -
233 233  * Audit statistics published regularly
234 234  * Accuracy rates by risk tier tracked and reported
235 235  * System improvements documented
236 236  * Community can view aggregate audit performance
237 -
238 238  == 8. Architecture Overview ==
239 -
240 240  {{include reference="FactHarbor.Specification.Diagrams.AKEL Architecture.WebHome"/}}
241 -
242 242  == 9. AKEL and Federation ==
243 -
244 244  In Release 1.0+, AKEL participates in cross-node knowledge alignment:
245 -
246 246  * Shares embeddings
247 247  * Exchanges canonicalized claim forms
248 248  * Exchanges scenario templates
... ... @@ -254,11 +254,8 @@
254 254  * Trusted nodes: auto-merge embeddings + templates
255 255  * Neutral nodes: require additional verification
256 256  * Untrusted nodes: fully manual import
257 -
258 258  == 10. Human Review Workflow (Mode 3 Publication) ==
259 -
260 260  For content requiring human validation before "AKEL-Generated" status:
261 -
262 262  1. AKEL generates content and publishes as AI-draft (Mode 2) or keeps as draft (Mode 1)
263 263  2. Contributors inspect content in review queue
264 264  3. Contributors validate quality gates were correctly applied
... ... @@ -266,11 +266,8 @@
266 266  5. Moderators finalize "AKEL-Generated" publication
267 267  6. Version numbers increment, full history preserved
268 268  **Note:** Most AI-generated content (Tier B and C) can remain in Mode 2 (AI-Generated) indefinitely. Human review is optional for these tiers unless users or audits flag issues.
269 -
270 270  == 11. POC v1 Behavior ==
271 -
272 272  The POC explicitly demonstrates AI-generated content publication:
273 -
274 274  * Produces public AI-generated output (Mode 2)
275 275  * No human data sources required
276 276  * No human approval gate
... ... @@ -278,10 +278,8 @@
278 278  * All quality gates active (including contradiction search)
279 279  * Users understand this demonstrates AI reasoning capabilities
280 280  * Risk tier classification shown (demo purposes)
281 -
282 282  == 12. Related Pages ==
283 -
284 284  * [[Automation>>FactHarbor.Specification.Automation.WebHome]]
285 285  * [[Requirements (Roles)>>FactHarbor.Specification.Requirements.WebHome]]
286 286  * [[Workflows>>FactHarbor.Specification.Workflows.WebHome]]
287 -* [[Governance>>Archive.FactHarbor.Organisation.Governance.WebHome]]
219 +* [[Governance>>FactHarbor.Organisation.Governance.WebHome]]