Skip to Content

Wiki source code of Workflows

Last modified by Robert Schaub on 2025/12/22 13:50
Hide last authors
Robert Schaub 1.1 1 = Workflows =
2
3 **Version:** 0.9.70
4 **Last Updated:** December 21, 2025
5 **Status:** CORRECTED - Automation Philosophy Consistent
6
7 This page describes FactHarbor's core workflows with the automation-first philosophy.
8
9
10 == 1. Core Workflow Principles ==
11
12 * **Automation First:** 90%+ content published automatically
13 * **No Approval Bottlenecks:** No centralized review queues
14 * **Quality Gates:** Automated validation before publication
15 * **Sampling Audits:** Pattern analysis for system improvement
16 * **Transparent Confidence:** All outputs labeled with confidence scores
17
18
19 == 2. Claim Submission Workflow ==
20
21 === 2.1 Claim Extraction ===
22
23 When users submit content (text, articles, web pages), FactHarbor first extracts individual verifiable claims:
24
25 **Input Types:**
26 * Single claim: "The Earth is flat"
27 * Text with multiple claims: "Climate change is accelerating. Sea levels rose 3mm in 2023. Arctic ice decreased 13% annually."
28 * URLs: Web pages analyzed for factual claims
29
30 **Extraction Process:**
31 * LLM analyzes submitted content
32 * Identifies distinct, verifiable factual claims
33 * Separates claims from opinions, questions, or commentary
34 * Each claim becomes independent for processing
35
36 **Output:**
37 * List of claims with context
38 * Each claim assigned unique ID
39 * Original context preserved for reference
40
41 This extraction ensures:
42 * Each claim receives focused analysis
43 * Multiple claims in one submission are all processed
44 * Claims are properly isolated for independent verification
45 * Context is preserved for accurate interpretation
46
47 **Flow:**
48 ```
49 User submits → Duplicate detection → Categorization → Processing queue → User receives ID
50 ```
51
52 **Timeline:** Seconds
53 **No approval needed:** Instant processing
54
55
56 == 3. Automated Analysis Workflow ==
57
58 **Complete Pipeline:**
59
60 ```
61 Claim from queue
62 ↓
63 Evidence gathering (AKEL)
64 ↓
65 Source evaluation (track record check)
66 ↓
67 Scenario generation
68 ↓
69 Verdict synthesis
70 ↓
71 Risk assessment
72 ↓
73 Quality gates validation
74 ↓
75 Decision: PUBLISH or BLOCK
76 ```
77
78 **Timeline:** 10-30 seconds
79 **Automation Rate:** 90%+ published automatically
80
81
82 === 3.1 Quality Gates Decision ==
83
84 **Gate Validation:**
85 1. Gate 1: Source Quality ✓
86 2. Gate 2: Contradiction Search ✓
87 3. Gate 3: Uncertainty Quantification ✓
88 4. Gate 4: Structural Integrity ✓
89
90 **If ALL gates PASS:**
91 → **Publish immediately** (Mode 2: AI-Generated)
92 → Apply appropriate risk tier label
93 → Display confidence score
94 → Make available for sampling audit
95
96 **If ANY gate FAILS:**
97 → **Block publication** (Mode 1: Draft-Only)
98 → Log failure reason
99 → Analyze failure pattern
100 → Queue system improvement task
101 → May re-process after improvements
102
103 **CRITICAL:** No human approval step - gates are automated.
104
105
106 == 4. Publication Workflow ==
107
108 **V0.9.70 CLARIFIED:** Risk tiers affect LABELS and AUDIT FREQUENCY, NOT approval requirements.
109
110 === Standard Flow (90%+) ===
111
112 ```
113 Pass quality gates
114 ↓
115 Determine risk tier (A/B/C)
116 ↓
117 Apply appropriate labels
118 ↓
119 PUBLISH IMMEDIATELY
120 ↓
121 Add to audit sampling pool
122 ```
123
124 **No delays, no approval queues**
125
126
127 === High-Risk Content (Tier A - <10%) ===
128
129 **V0.9.70 CORRECTION:**
130
131 ```
132 Pass quality gates
133 ↓
134 Identified as Tier A (medical/legal/safety)
135 ↓
136 PUBLISH IMMEDIATELY with prominent warnings
137 ↓
138 Higher sampling audit frequency (50%)
139 ```
140
141 **What changed from V0.9.69:**
142 - ❌ REMOVED: "Risk > 80% → Moderator review"
143 - ✅ ADDED: "Risk > 80% → Publish with WARNING labels"
144
145 **Philosophy:** Publish with strong warnings, monitor closely through sampling.
146
147 **Warning Labels for Tier A:**
148 ```
149 ⚠️ HIGH-IMPACT TOPIC
150 AI-Generated Analysis
151
152 This claim involves [medical/legal/financial/safety] topics.
153 - Confidence: [X]%
154 - Last Updated: [timestamp]
155 - This is NOT professional advice
156 - Consult qualified professionals for decisions
157
158 [View Evidence] [See Methodology] [Report Issue]
159 ```
160
161
162 === Low Quality Content (<10%) ===
163
164 ```
165 FAIL quality gates
166 ↓
167 Confidence < threshold OR structural issues
168 ↓
169 BLOCK (Mode 1: Draft-Only)
170 ↓
171 Log failure patterns
172 ↓
173 Queue for system improvement
174 ```
175
176 **NOT:** Send for human review
177 **IS:** Improve prompts/algorithms based on failure patterns
178
179
180 == 5. User Contribution Workflow ==
181
182 **Philosophy:** Wikipedia-style immediate application + audit trail
183
184 ```
185 Contributor edits published content
186 ↓
187 System validates (basic checks)
188 ↓
189 Applied IMMEDIATELY
190 ↓
191 Logged in version history
192 ↓
193 Reputation earned
194 ↓
195 May be selected for sampling audit
196 ```
197
198 **No approval required:** Changes apply instantly
199 **Quality control:** Through sampling audits and reputation system
200
201 **New contributors** (<50 reputation): Limited to minor edits
202
203
204 == 6. Sampling Audit Workflow ==
205
206 **Purpose:** Improve system quality through pattern analysis
207
208 === 6.1 Selection Process ===
209
210 ```
211 Published content
212 ↓
213 Stratified sampling (by risk tier, confidence, traffic)
214 ↓
215 Selected for audit (Tier A: 50%, B: 20%, C: 5%)
216 ↓
217 Added to audit queue
218 ```
219
220
221 === 6.2 Audit Execution ===
222
223 ```
224 Auditor receives sample
225 ↓
226 Reviews against quality standards
227 ↓
228 Identifies issues/patterns
229 ↓
230 Logs findings
231 ↓
232 System improvement tasks created
233 ```
234
235 **What auditors DO:**
236 * ✅ Analyze patterns across multiple outputs
237 * ✅ Identify systematic issues
238 * ✅ Recommend algorithm/prompt improvements
239 * ✅ Track accuracy trends
240
241 **What auditors DON'T DO:**
242 * ❌ Approve individual outputs before publication
243 * ❌ Manually fix individual outputs
244 * ❌ Act as gatekeepers
245 * ❌ Override quality gates
246
247
248 === 6.3 Improvement Loop ===
249
250 ```
251 Audit findings aggregated
252 ↓
253 Patterns identified
254 ↓
255 System improvements proposed
256 ↓
257 Implemented and tested
258 ↓
259 Deployed
260 ↓
261 Metrics monitored
262 ```
263
264 **Examples of Improvements:**
265 * Refine evidence search queries
266 * Adjust source reliability weights
267 * Enhance contradiction detection
268 * Improve claim extraction prompts
269 * Recalibrate risk tier thresholds
270
271
272 == 7. Flagging Workflow ==
273
274 **Two types of flags:**
275
276 === 7.1 Quality Issues ===
277
278 ```
279 User flags quality issue
280 ↓
281 Categorized automatically
282 ↓
283 Added to sampling audit pool (priority)
284 ↓
285 Pattern analysis
286 ↓
287 System improvement if pattern found
288 ```
289
290 **NOT:** Manual correction of individual claim
291 **IS:** Improve system to prevent similar issues
292
293
294 === 7.2 Abuse/Spam ===
295
296 ```
297 User flags abuse/spam
298 ↓
299 Automated pre-moderation check
300 ↓
301 Moderator review (if needed)
302 ↓
303 Action taken (hide/ban)
304 ```
305
306 **Moderator role:** Handle abuse/spam, NOT content quality
307
308
309 == 8. Moderation Workflow ==
310
311 **V0.9.70 CLARIFIED:** Moderators handle ABUSE, not content quality
312
313 === 8.1 Content Moderation (Abuse/Spam) ===
314
315 **Moderator Queue Contains:**
316 * Flagged abusive content
317 * Spam detection alerts
318 * Harassment reports
319 * Privacy violations
320 * Terms of service violations
321
322 **Moderator Actions:**
323 * Hide abusive content
324 * Ban repeat offenders
325 * Handle appeals
326 * Escalate to governing team
327
328 **Moderators DO NOT:**
329 * ❌ Approve content for publication
330 * ❌ Review content quality before publication
331 * ❌ Act as editorial gatekeepers
332 * ❌ Manually fix AI outputs
333
334
335 === 8.2 Appeal Process ===
336
337 ```
338 User disagrees with moderation
339 ↓
340 Appeals to different moderator
341 ↓
342 If still disagrees, escalates to Governing Team
343 ↓
344 Governing Team decision (final)
345 ```
346
347
348 == 9. Time Evolution Workflow ==
349
350 **Automatic Re-evaluation:**
351
352 ```
353 Published claim
354 ↓
355 Monitoring for triggers:
356 - New evidence published
357 - Source retractions
358 - Significant events
359 - Scheduled review (6-12 months)
360 ↓
361 Trigger detected
362 ↓
363 AKEL re-processes claim
364 ↓
365 Quality gates validate
366 ↓
367 If verdict changes: Correction workflow
368 ↓
369 If passes: Update published analysis
370 ```
371
372 **Correction Workflow (New in V0.9.70):**
373
374 ```
375 Verdict changed significantly
376 ↓
377 Generate correction notice
378 ↓
379 Publish correction banner (30 days)
380 ↓
381 Update corrections log
382 ↓
383 Notify users (email, RSS, API)
384 ↓
385 Update ClaimReview schema
386 ```
387
388
389 == 10. Contributor Journey ==
390
391 1. **Visitor** – Explores platform, reads documentation
392 2. **New Contributor** – Submits first improvements (typo fixes, clarifications)
393 3. **Contributor** – Contributes regularly, follows conventions
394 4. **Trusted Contributor** – Track record of quality work
395 5. **Reviewer** – Participates in sampling audits (pattern analysis)
396 6. **Moderator** – Handles abuse/spam (not content quality)
397 7. **Expert** (optional) – Provides domain expertise for contested claims
398
399 **All contributions apply immediately** - no approval workflow
400
401
402 == 11. Related Pages ==
403
404 * [[AKEL>>FactHarbor.Specification.AI Knowledge Extraction Layer (AKEL).WebHome]] - AI processing system
405 * [[Architecture>>FactHarbor.Specification.Architecture.WebHome]] - System architecture
406 * [[Requirements>>FactHarbor.Specification.Requirements.WebHome]] - Requirements and roles
407 * [[Decision Processes>>FactHarbor.Organisation.Decision-Processes.WebHome]] - Governance
408
409
410 **V0.9.70 CHANGES:**
411
412 **REMOVED:**
413 - ❌ "High Risk → Moderator review" (was approval workflow)
414 - ❌ "Review queue" language for publication
415 - ❌ Any implication that moderators approve content quality
416
417 **ADDED/CLARIFIED:**
418 - ✅ Risk tiers affect warnings and audit frequency, NOT approval
419 - ✅ High-risk content publishes immediately with prominent warnings
420 - ✅ Quality gate failures → Block + improve system (not human review)
421 - ✅ Clear distinction: Sampling audits (improvement) vs. Content moderation (abuse)
422 - ✅ Moderator role clarified: Abuse only, NOT content quality
423 - ✅ User contributions apply immediately (Wikipedia model)
424 - ✅ Correction workflow for significant verdict changes
425 - ✅ Time evolution and re-evaluation workflow
426