Skip to Content
Last modified by Robert Schaub on 2026/02/08 21:26
From version 1.3
edited by Robert Schaub
on 2026/01/20 20:21
Change comment: Renamed back-links.
To version 1.2
edited by Robert Schaub
on 2025/12/22 13:49
Change comment: Update document after refactoring.

Summary

Details

Page properties
Content
... ... @@ -10,10 +10,10 @@
10 10  AKEL outputs are marked with **AuthorType = AI** and published according to risk-based policies (see Publication Modes below).
11 11  
12 12  AKEL operates in two modes:
13 -
14 14  * **Single-node mode** (POC & Beta 0)
15 15  * **Federated multi-node mode** (Release 1.0+)
16 16  
16 +
17 17  == 1. Core Philosophy: Automation First ==
18 18  
19 19  **V0.9.50+ Philosophy Shift:**
... ... @@ -41,7 +41,6 @@
41 41  **Status:** Not visible to public
42 42  
43 43  **When Used:**
44 -
45 45  * Quality gates failed
46 46  * Confidence below threshold
47 47  * Structural integrity issues
... ... @@ -48,7 +48,6 @@
48 48  * Insufficient evidence
49 49  
50 50  **What Happens:**
51 -
52 52  * Content remains private
53 53  * System logs failure reasons
54 54  * Prompts/algorithms improved based on patterns
... ... @@ -62,7 +62,6 @@
62 62  **Status:** Published and visible to all users
63 63  
64 64  **When Used:**
65 -
66 66  * Quality gates passed
67 67  * Confidence ≥ threshold
68 68  * Meets structural requirements
... ... @@ -69,7 +69,6 @@
69 69  * Sufficient evidence found
70 70  
71 71  **Includes:**
72 -
73 73  * Confidence score displayed (0-100%)
74 74  * Risk tier badge (A/B/C)
75 75  * Quality indicators
... ... @@ -77,17 +77,16 @@
77 77  * Sampling audit status
78 78  
79 79  **Labels by Risk Tier:**
80 -
81 81  * **Tier A (High Risk):** "⚠️ AI-Generated - High Impact Topic - Seek Professional Advice"
82 82  * **Tier B (Medium Risk):** "🤖 AI-Generated - May Contain Errors"
83 83  * **Tier C (Low Risk):** "🤖 AI-Generated"
84 84  
80 +
85 85  === REMOVED: "Mode 3: Human-Reviewed" ===
86 86  
87 87  **V0.9.50 Decision:** No centralized approval workflow.
88 88  
89 89  **Rationale:**
90 -
91 91  * Defeats automation purpose
92 92  * Creates bottleneck
93 93  * Inconsistent quality
... ... @@ -94,12 +94,12 @@
94 94  * Not scalable
95 95  
96 96  **What Replaced It:**
97 -
98 98  * Better quality gates
99 99  * Sampling audits for system improvement
100 100  * Transparent confidence scoring
101 101  * Risk-based warnings
102 102  
97 +
103 103  == 3. Risk Tiers (A/B/C) ==
104 104  
105 105  Risk classification determines WARNING LABELS and AUDIT FREQUENCY, NOT approval requirements.
... ... @@ -109,7 +109,6 @@
109 109  **Examples:** Medical advice, legal interpretations, financial recommendations, safety information
110 110  
111 111  **Impact:**
112 -
113 113  * ✅ Publish immediately (if passes gates)
114 114  * ✅ Prominent warning labels
115 115  * ✅ Higher sampling audit frequency (50% audited)
... ... @@ -124,22 +124,22 @@
124 124  **Examples:** Political claims, controversial topics, scientific debates
125 125  
126 126  **Impact:**
127 -
128 128  * ✅ Publish immediately (if passes gates)
129 129  * ✅ Standard warning labels
130 130  * ✅ Medium sampling audit frequency (20% audited)
131 131  * ❌ NOT held for moderator approval
132 132  
126 +
133 133  === Tier C: Low-Stakes Claims ===
134 134  
135 135  **Examples:** Entertainment facts, sports statistics, general knowledge
136 136  
137 137  **Impact:**
138 -
139 139  * ✅ Publish immediately (if passes gates)
140 140  * ✅ Minimal warning labels
141 141  * ✅ Low sampling audit frequency (5% audited)
142 142  
136 +
143 143  == 4. Quality Gates (Automated, Not Human) ==
144 144  
145 145  All AI-generated content must pass these **AUTOMATED checks** before publication:
... ... @@ -147,7 +147,6 @@
147 147  === Gate 1: Source Quality ===
148 148  
149 149  **Automated Checks:**
150 -
151 151  * Primary sources identified and accessible
152 152  * Source reliability scored against database
153 153  * Citation completeness verified
... ... @@ -167,7 +167,6 @@
167 167  * **Bubble detection** – Echo chambers, ideologically isolated sources
168 168  
169 169  **Search Coverage Requirements:**
170 -
171 171  * Academic literature (BOTH supporting AND opposing views)
172 172  * Diverse media across political/ideological perspectives
173 173  * Official contradictions (retractions, corrections, amendments)
... ... @@ -174,7 +174,6 @@
174 174  * Cross-cultural and international perspectives
175 175  
176 176  **Search Must Avoid Algorithmic Bubbles:**
177 -
178 178  * Deliberately seek opposing viewpoints
179 179  * Check for echo chamber patterns
180 180  * Identify tribal source clustering
... ... @@ -182,7 +182,6 @@
182 182  * Verify diversity of perspectives
183 183  
184 184  **Outcomes:**
185 -
186 186  * Strong counter-evidence → Auto-escalate to Tier B or draft-only
187 187  * Significant uncertainty → Require uncertainty disclosure in verdict
188 188  * Bubble indicators → Flag for sampling audit
... ... @@ -194,7 +194,6 @@
194 194  === Gate 3: Uncertainty Quantification ===
195 195  
196 196  **Automated Checks:**
197 -
198 198  * Confidence scores calculated for all claims and verdicts
199 199  * Limitations explicitly stated
200 200  * Data gaps identified and disclosed
... ... @@ -207,7 +207,6 @@
207 207  === Gate 4: Structural Integrity ===
208 208  
209 209  **Automated Checks:**
210 -
211 211  * No hallucinations detected (fact-checking against sources)
212 212  * Logic chain valid and traceable
213 213  * References accessible and verifiable
... ... @@ -218,7 +218,6 @@
218 218  
219 219  
220 220  **CRITICAL:** If any gate fails:
221 -
222 222  * ✅ Content remains in draft-only mode
223 223  * ✅ Failure reason logged
224 224  * ✅ Failure patterns analyzed for system improvement
... ... @@ -237,7 +237,6 @@
237 237  **Stratified Sampling Strategy:**
238 238  
239 239  Audits prioritize:
240 -
241 241  * **Risk tier** (Tier A: 50%, Tier B: 20%, Tier C: 5%)
242 242  * **AI confidence score** (low confidence → higher sampling rate)
243 243  * **Traffic and engagement** (high-visibility content audited more)
... ... @@ -253,17 +253,16 @@
253 253  1. **System selects** content for audit based on sampling strategy
254 254  2. **Human auditor** reviews AI-generated content against quality standards
255 255  3. **Auditor validates or identifies issues:**
256 -
257 -* Claim extraction accuracy
258 -* Scenario appropriateness
259 -* Evidence relevance and interpretation
260 -* Verdict reasoning
261 -* Contradiction search completeness
242 + * Claim extraction accuracy
243 + * Scenario appropriateness
244 + * Evidence relevance and interpretation
245 + * Verdict reasoning
246 + * Contradiction search completeness
262 262  4. **Audit outcome recorded** (pass/fail + detailed feedback)
263 263  5. **Failed audits trigger:**
264 -* Analysis of failure pattern
265 -* System improvement tasks
266 -* Algorithm/prompt adjustments
249 + * Analysis of failure pattern
250 + * System improvement tasks
251 + * Algorithm/prompt adjustments
267 267  6. **Audit results feed back** into system improvement
268 268  
269 269  **CRITICAL:** Auditors analyze PATTERNS, not fix individual outputs.
... ... @@ -286,7 +286,6 @@
286 286  === 5.4 Audit Transparency ===
287 287  
288 288  **Publicly Published:**
289 -
290 290  * Audit statistics (monthly)
291 291  * Accuracy rates by risk tier
292 292  * System improvements made
... ... @@ -293,11 +293,11 @@
293 293  * Aggregate audit performance
294 294  
295 295  **Enables:**
296 -
297 297  * Public accountability
298 298  * System trust
299 299  * Continuous improvement visibility
300 300  
284 +
301 301  == 6. Human Intervention Criteria ==
302 302  
303 303  **From Organisation.Decision-Processes:**
... ... @@ -339,7 +339,6 @@
339 339  ```
340 340  
341 341  **Components in Single Call:**
342 -
343 343  1. Extract 3-5 factual claims
344 344  2. For each claim: verdict + confidence + risk tier + reasoning
345 345  3. Generate analysis summary
... ... @@ -369,7 +369,6 @@
369 369  ```
370 370  
371 371  **Processing:**
372 -
373 373  * Parallel processing where possible
374 374  * Separate component calls
375 375  * Quality gates between phases
... ... @@ -400,11 +400,11 @@
400 400  * Never overrides local governance
401 401  
402 402  Nodes may choose trust levels for AKEL-related data:
403 -
404 404  * Trusted nodes: auto-merge embeddings + templates
405 405  * Neutral nodes: require additional verification
406 406  * Untrusted nodes: fully manual import
407 407  
389 +
408 408  == 9. POC Behavior ==
409 409  
410 410  The POC explicitly demonstrates AI-generated content publication:
... ... @@ -425,9 +425,10 @@
425 425  * [[Automation>>FactHarbor.Specification.Automation.WebHome]]
426 426  * [[Requirements (Roles)>>FactHarbor.Specification.Requirements.WebHome]]
427 427  * [[Workflows>>FactHarbor.Specification.Workflows.WebHome]]
428 -* [[Governance>>Archive.FactHarbor.Organisation.Governance.WebHome]]
410 +* [[Governance>>FactHarbor.Organisation.Governance.WebHome]]
429 429  * [[Decision Processes>>FactHarbor.Organisation.Decision-Processes.WebHome]]
430 430  
413 +
431 431  **V0.9.70 CHANGES:**
432 432  - ❌ REMOVED: Section "Human Review Workflow (Mode 3 Publication)"
433 433  - ❌ REMOVED: All references to "Mode 3"
... ... @@ -439,3 +439,4 @@
439 439  - ✅ ENHANCED: Gate 2 (Contradiction Search) specification
440 440  - ✅ ADDED: Clear human intervention criteria
441 441  - ✅ ADDED: Detailed audit system explanation
425 +