Changes for page POC Requirements (POC1 & POC2)

Summary

• Page properties (2 modified, 0 added, 0 removed)

Details

Page properties

Parent

...	...	@@ -1,1 +1,1 @@
1		-Test.FactHarbor.Specification.POC.WebHome
	1	+WebHome

Content

...	...	@@ -18,9 +18,11 @@
18	18	=== 1.1 What POC Tests ===
19	19
20	20	**Core Question:**
	21	+
21	21	> Can AI automatically extract factual claims from articles and evaluate them with reasonable verdicts?
22	22
23	23	**What we're proving:**
	25	+
24	24	* AI can identify factual claims from text
25	25	* AI can evaluate those claims and produce verdicts
26	26	* Output is comprehensible and useful
...	...	@@ -27,6 +27,7 @@
27	27	* Fully automated approach is viable
28	28
29	29	**What we're NOT testing:**
	32	+
30	30	* Scenario generation (deferred to POC2)
31	31	* Evidence display (deferred to POC2)
32	32	* Production scalability
...	...	@@ -40,6 +40,7 @@
40	40	Scenarios are a core component of the full FactHarbor system (Claims → Scenarios → Evidence → Verdicts), but are **deliberately excluded from POC1**.
41	41
42	42	**Rationale:**
	46	+
43	43	* **POC1 tests:** Can AI extract claims and generate verdicts?
44	44	* **POC2 will add:** Scenario generation and management
45	45	* **Open questions remain:** Should scenarios be separate entities? How are they sequenced with evidence gathering? What's the optimal workflow?
...	...	@@ -51,6 +51,7 @@
51	51	**No Risk:**
52	52
53	53	Scenarios are additive complexity, not foundational. Deferring them to POC2 allows:
	58	+
54	54	* Faster POC1 validation
55	55	* Learning from POC1 to inform scenario design
56	56	* Iterative approach: fail fast if basic AI doesn't work
...	...	@@ -57,14 +57,10 @@
57	57	* Flexibility to adjust scenario architecture based on POC1 insights
58	58
59	59	**Full System Workflow (Future):**
60		-{{code}}
61		-Claims → Scenarios → Evidence → Verdicts
62		-{{/code}}
	65	+{{code}}Claims → Scenarios → Evidence → Verdicts{{/code}}
63	63
64	64	**POC1 Simplified Workflow:**
65		-{{code}}
66		-Claims → Verdicts (scenarios implicit in reasoning)
67		-{{/code}}
	68	+{{code}}Claims → Verdicts (scenarios implicit in reasoning){{/code}}
68	68
69	69	== 2. POC Output Specification ==
70	70
...	...	@@ -75,6 +75,7 @@
75	75	**Length:** 4-6 sentences
76	76
77	77	**Content (Required Elements):**
	79	+
78	78	1. **Article's main thesis/claim** - What is the article trying to argue or prove?
79	79	2. **Claim count and verdicts** - How many claims analyzed, distribution of verdicts
80	80	3. **Central vs. supporting claims** - Which claims are central to the article's argument?
...	...	@@ -84,30 +84,28 @@
84	84	**Critical Innovation:**
85	85
86	86	POC1 tests whether AI can understand that **article credibility ≠ simple average of claim verdicts**. An article might:
	89	+
87	87	* Make accurate supporting facts but draw unsupported conclusions
88	88	* Have one false central claim that invalidates the whole argument
89	89	* Misframe accurate information to mislead
90	90
91	91	**Good Example (Context-Aware):**
92		-{{code}}
93		-This article argues that coffee cures cancer based on its antioxidant
	95	+{{code}}This article argues that coffee cures cancer based on its antioxidant
94	94	content. We analyzed 3 factual claims: 2 about coffee's chemical
95	95	properties are well-supported, but the main causal claim is refuted
96	96	by current evidence. The article confuses correlation with causation.
97	97	Overall assessment: MISLEADING - makes an unsupported medical claim
98		-despite citing some accurate facts.
99		-{{/code}}
	100	+despite citing some accurate facts.{{/code}}
100	100
101	101	**Poor Example (Simple Aggregation - Don't Do This):**
102		-{{code}}
103		-This article makes 3 claims. 2 are well-supported and 1 is refuted.
104		-Overall assessment: mostly accurate (67% accurate).
105		-{{/code}}
	103	+{{code}}This article makes 3 claims. 2 are well-supported and 1 is refuted.
	104	+Overall assessment: mostly accurate (67% accurate).{{/code}}
106	106	↑ This misses that the refuted claim IS the article's main point!
107	107
108	108	**What POC1 Tests:**
109	109
110	110	Can AI identify and assess:
	110	+
111	111	* ✅ The article's main thesis/conclusion?
112	112	* ✅ Which claims are central vs. supporting?
113	113	* ✅ Whether the evidence supports the conclusion?
...	...	@@ -116,6 +116,7 @@
116	116	**If AI Cannot Do This:**
117	117
118	118	That's valuable to learn in POC1! We'll:
	119	+
119	119	* Note as limitation
120	120	* Fall back to simple aggregation with warning
121	121	* Design explicit article-level analysis for POC2
...	...	@@ -126,19 +126,18 @@
126	126	**Format:** Numbered list
127	127	**Quantity:** 3-5 claims
128	128	**Requirements:**
	130	+
129	129	* Factual claims only (not opinions/questions)
130	130	* Clearly stated
131	131	* Automatically extracted by AI
132	132
133	133	**Example:**
134		-{{code}}
135		-CLAIMS IDENTIFIED:
	136	+{{code}}CLAIMS IDENTIFIED:
136	136
137	137	[1] Coffee reduces diabetes risk by 30%
138	138	[2] Coffee improves heart health
139	139	[3] Decaf has same benefits as regular
140		-[4] Coffee prevents Alzheimer's completely
141		-{{/code}}
	141	+[4] Coffee prevents Alzheimer's completely{{/code}}
142	142
143	143	=== 2.3 Component 3: CLAIMS VERDICTS ===
144	144
...	...	@@ -146,6 +146,7 @@
146	146	**Format:** Per claim structure
147	147
148	148	**Required Elements:**
	149	+
149	149	* **Verdict Label:** WELL-SUPPORTED / PARTIALLY SUPPORTED / UNCERTAIN / REFUTED
150	150	* **Confidence Score:** 0-100%
151	151	* **Brief Reasoning:** 1-3 sentences explaining why
...	...	@@ -152,8 +152,7 @@
152	152	* **Risk Tier:** A (High) / B (Medium) / C (Low) - for demonstration
153	153
154	154	**Example:**
155		-{{code}}
156		-VERDICTS:
	156	+{{code}}VERDICTS:
157	157
158	158	[1] WELL-SUPPORTED (85%) [Risk: C]
159	159	Multiple studies confirm 25-30% risk reduction with regular consumption.
...	...	@@ -165,10 +165,10 @@
165	165	Some benefits overlap, but caffeine-related benefits are reduced in decaf.
166	166
167	167	[4] REFUTED (90%) [Risk: B]
168		-No evidence for complete prevention. Claim is significantly overstated.
169		-{{/code}}
	168	+No evidence for complete prevention. Claim is significantly overstated.{{/code}}
170	170
171	171	**Risk Tier Display:**
	171	+
172	172	* **Tier A (Red):** High Risk - Medical/Legal/Safety/Elections
173	173	* **Tier B (Yellow):** Medium Risk - Policy/Science/Causality
174	174	* **Tier C (Green):** Low Risk - Facts/Definitions/History
...	...	@@ -182,13 +182,11 @@
182	182	**Tone:** Neutral (article's position, not FactHarbor's analysis)
183	183
184	184	**Example:**
185		-{{code}}
186		-ARTICLE SUMMARY:
	185	+{{code}}ARTICLE SUMMARY:
187	187
188	188	Health News Today article discusses coffee benefits, citing studies
189	189	on diabetes and Alzheimer's. Author highlights research linking coffee
190		-to disease prevention. Recommends 2-3 cups daily for optimal health.
191		-{{/code}}
	189	+to disease prevention. Recommends 2-3 cups daily for optimal health.{{/code}}
192	192
193	193	=== 2.5 Component 5: USAGE STATISTICS (Cost Tracking) ===
194	194
...	...	@@ -195,6 +195,7 @@
195	195	**What:** LLM usage metrics for cost optimization and scaling decisions
196	196
197	197	**Purpose:**
	196	+
198	198	* Understand cost per analysis
199	199	* Identify optimization opportunities
200	200	* Project costs at scale
...	...	@@ -201,8 +201,7 @@
201	201	* Inform architecture decisions
202	202
203	203	**Display Format:**
204		-{{code}}
205		-USAGE STATISTICS:
	203	+{{code}}USAGE STATISTICS:
206	206	• Article: 2,450 words (12,300 characters)
207	207	• Input tokens: 15,234
208	208	• Output tokens: 892
...	...	@@ -210,17 +210,18 @@
210	210	• Estimated cost: $0.24 USD
211	211	• Response time: 8.3 seconds
212	212	• Cost per claim: $0.048
213		-• Model: claude-sonnet-4-20250514
214		-{{/code}}
	211	+• Model: claude-sonnet-4-20250514{{/code}}
215	215
216	216	**Why This Matters:**
217	217
218	218	At scale, LLM costs are critical:
	216	+
219	219	* 10,000 articles/month ≈ $200-500/month
220	220	* 100,000 articles/month ≈ $2,000-5,000/month
221	221	* Cost optimization can reduce expenses 30-50%
222	222
223	223	**What POC1 Learns:**
	222	+
224	224	* How cost scales with article length
225	225	* Prompt optimization opportunities (caching, compression)
226	226	* Output verbosity tradeoffs
...	...	@@ -228,6 +228,7 @@
228	228	* Article length limits (if needed)
229	229
230	230	**Implementation:**
	230	+
231	231	* Claude API already returns usage data
232	232	* No extra API calls needed
233	233	* Display to user + log for aggregate analysis
...	...	@@ -237,7 +237,8 @@
237	237
238	238	=== 2.6 Total Output Size ===
239	239
240		-**Combined:** ~220-350 words
	240	+**Combined:** 220-350 words
	241	+
241	241	* Analysis Summary (Context-Aware): 60-90 words (4-6 sentences)
242	242	* Claims Identification: 30-50 words
243	243	* Claims Verdicts: 100-150 words
...	...	@@ -252,6 +252,7 @@
252	252	The following are **explicitly excluded** from POC:
253	253
254	254	**Content Features:**
	256	+
255	255	* ❌ Scenarios (deferred to POC2)
256	256	* ❌ Evidence display (supporting/opposing lists)
257	257	* ❌ Source links (clickable references)
...	...	@@ -261,6 +261,7 @@
261	261	* ❌ Risk assessment (shown but not workflow-integrated)
262	262
263	263	**Platform Features:**
	266	+
264	264	* ❌ User accounts / authentication
265	265	* ❌ Saved history
266	266	* ❌ Search functionality
...	...	@@ -270,6 +270,7 @@
270	270	* ❌ Social sharing
271	271
272	272	**Technical Features:**
	276	+
273	273	* ❌ Browser extensions
274	274	* ❌ Mobile apps
275	275	* ❌ API endpoints
...	...	@@ -277,6 +277,7 @@
277	277	* ❌ Export features (PDF, CSV)
278	278
279	279	**Quality Features:**
	284	+
280	280	* ❌ Accessibility (WCAG compliance)
281	281	* ❌ Multilingual support
282	282	* ❌ Mobile optimization
...	...	@@ -283,6 +283,7 @@
283	283	* ❌ Media verification (images/videos)
284	284
285	285	**Production Features:**
	291	+
286	286	* ❌ Security hardening
287	287	* ❌ Privacy compliance (GDPR)
288	288	* ❌ Terms of service
...	...	@@ -296,17 +296,13 @@
296	296	=== 4.1 Architecture Comparison ===
297	297
298	298	**POC Architecture (Simplified):**
299		-{{code}}
300		-User Input → Single AKEL Call → Output Display
301		- (all processing)
302		-{{/code}}
	305	+{{code}}User Input → Single AKEL Call → Output Display
	306	+ (all processing){{/code}}
303	303
304	304	**Full System Architecture:**
305		-{{code}}
306		-User Input → Claim Extractor → Claim Classifier → Scenario Generator
	309	+{{code}}User Input → Claim Extractor → Claim Classifier → Scenario Generator
307	307	→ Evidence Summarizer → Contradiction Detector → Verdict Generator
308		-→ Quality Gates → Publication → Output Display
309		-{{/code}}
	311	+→ Quality Gates → Publication → Output Display{{/code}}
310	310
311	311	**Key Differences:**
312	312
...	...	@@ -322,12 +322,14 @@
322	322	=== 4.2 Workflow Comparison ===
323	323
324	324	**POC1 Workflow:**
	327	+
325	325	1. User submits text/URL
326	326	2. Single AKEL call (all processing in one prompt)
327	327	3. Display results
328		-**Total: 3 steps, ~10-18 seconds**
	331	+**Total: 3 steps, 10-18 seconds**
329	329
330	330	**Full System Workflow:**
	334	+
331	331	1. **Claim Submission** (extraction, normalization, clustering)
332	332	2. **Scenario Building** (definitions, assumptions, boundaries)
333	333	3. **Evidence Handling** (retrieval, assessment, linking)
...	...	@@ -334,7 +334,7 @@
334	334	4. **Verdict Creation** (synthesis, reasoning, approval)
335	335	5. **Public Presentation** (summaries, landscapes, deep dives)
336	336	6. **Time Evolution** (versioning, re-evaluation triggers)
337		-**Total: 6 phases with quality gates, ~10-30 seconds**
	341	+**Total: 6 phases with quality gates, 10-30 seconds**
338	338
339	339	=== 4.3 Why POC is Simplified ===
340	340
...	...	@@ -357,6 +357,7 @@
357	357	=== 4.4 Gap Between POC1 and POC2/Beta ===
358	358
359	359	**What needs to be built for POC2:**
	364	+
360	360	* Scenario generation component
361	361	* Evidence Model structure (full)
362	362	* Scenario-evidence linking
...	...	@@ -364,6 +364,7 @@
364	364	* Truth landscape visualization
365	365
366	366	**What needs to be built for Beta:**
	372	+
367	367	* Multi-component AKEL pipeline
368	368	* Quality gate infrastructure
369	369	* Review workflow system
...	...	@@ -380,6 +380,7 @@
380	380	**Mode:** Mode 2 (AI-Generated, No Prior Human Review)
381	381
382	382	Per FactHarbor Specification Section 11 "POC v1 Behavior":
	389	+
383	383	* Produces public AI-generated output
384	384	* No human approval gate
385	385	* Clear AI-Generated labeling
...	...	@@ -389,8 +389,7 @@
389	389	=== 5.2 User-Facing Labels ===
390	390
391	391	**Primary Label (top of analysis):**
392		-{{code}}
393		-╔════════════════════════════════════════════════════════════╗
	399	+{{code}}╔════════════════════════════════════════════════════════════╗
394	394	║ [AI-GENERATED - POC/DEMO] ║
395	395	║ ║
396	396	║ This analysis was produced entirely by AI and has not ║
...	...	@@ -400,10 +400,10 @@
400	400	║ Review Status: Not Reviewed (Proof-of-Concept) ║
401	401	║ Quality Gates: 4/4 Passed (Simplified) ║
402	402	║ Last Updated: [timestamp] ║
403		-╚════════════════════════════════════════════════════════════╝
404		-{{/code}}
	409	+╚════════════════════════════════════════════════════════════╝{{/code}}
405	405
406	406	**Per-Claim Risk Labels:**
	412	+
407	407	* **[Risk: A]** 🔴 High Risk (Medical/Legal/Safety)
408	408	* **[Risk: B]** 🟡 Medium Risk (Policy/Science)
409	409	* **[Risk: C]** 🟢 Low Risk (Facts/Definitions)
...	...	@@ -411,6 +411,7 @@
411	411	=== 5.3 Display Requirements ===
412	412
413	413	**Must Show:**
	420	+
414	414	* AI-Generated status (prominent)
415	415	* POC/Demo disclaimer
416	416	* Risk tier per claim
...	...	@@ -419,6 +419,7 @@
419	419	* Timestamp
420	420
421	421	**Must NOT Claim:**
	429	+
422	422	* Human review
423	423	* Production quality
424	424	* Medical/legal advice
...	...	@@ -442,6 +442,7 @@
442	442	Per FactHarbor Specification Section 6, all AI-generated content must pass quality gates before publication. POC implements **simplified versions** of the 4 mandatory gates.
443	443
444	444	**Full System Has 4 Gates:**
	453	+
445	445	1. Source Quality
446	446	2. Contradiction Search (MANDATORY)
447	447	3. Uncertainty Quantification
...	...	@@ -448,6 +448,7 @@
448	448	4. Structural Integrity
449	449
450	450	**POC Implements Simplified Versions:**
	460	+
451	451	* Focus on demonstrating concept
452	452	* Basic implementations sufficient
453	453	* Failures displayed to user (not blocking)
...	...	@@ -456,6 +456,7 @@
456	456	=== 6.2 Gate 1: Source Quality (Basic) ===
457	457
458	458	**Full System Requirements:**
	469	+
459	459	* Primary sources identified and accessible
460	460	* Source reliability scored against whitelist
461	461	* Citation completeness verified
...	...	@@ -463,6 +463,7 @@
463	463	* Author credentials validated
464	464
465	465	**POC Implementation:**
	477	+
466	466	* ✅ At least 2 sources found
467	467	* ✅ Sources accessible (URLs valid)
468	468	* ❌ No whitelist checking
...	...	@@ -476,6 +476,7 @@
476	476	=== 6.3 Gate 2: Contradiction Search (Basic) ===
477	477
478	478	**Full System Requirements:**
	491	+
479	479	* Counter-evidence actively searched
480	480	* Reservations and limitations identified
481	481	* Alternative interpretations explored
...	...	@@ -484,6 +484,7 @@
484	484	* Academic literature (supporting AND opposing)
485	485
486	486	**POC Implementation:**
	500	+
487	487	* ✅ Basic search for counter-evidence
488	488	* ✅ Identify obvious contradictions
489	489	* ❌ No comprehensive academic search
...	...	@@ -498,6 +498,7 @@
498	498	=== 6.4 Gate 3: Uncertainty Quantification (Basic) ===
499	499
500	500	**Full System Requirements:**
	515	+
501	501	* Confidence scores calculated for all claims/verdicts
502	502	* Limitations explicitly stated
503	503	* Data gaps identified and disclosed
...	...	@@ -505,6 +505,7 @@
505	505	* Alternative scenarios considered
506	506
507	507	**POC Implementation:**
	523	+
508	508	* ✅ Confidence scores (0-100%)
509	509	* ✅ Basic uncertainty acknowledgment
510	510	* ❌ No detailed limitation disclosure
...	...	@@ -518,6 +518,7 @@
518	518	=== 6.5 Gate 4: Structural Integrity (Basic) ===
519	519
520	520	**Full System Requirements:**
	537	+
521	521	* No hallucinations detected (fact-checking against sources)
522	522	* Logic chain valid and traceable
523	523	* References accessible and verifiable
...	...	@@ -525,6 +525,7 @@
525	525	* Premises clearly stated
526	526
527	527	**POC Implementation:**
	545	+
528	528	* ✅ Basic coherence check
529	529	* ✅ References accessible
530	530	* ❌ No comprehensive hallucination detection
...	...	@@ -538,24 +538,20 @@
538	538	=== 6.6 Quality Gate Display ===
539	539
540	540	**POC shows simplified status:**
541		-{{code}}
542		-Quality Gates: 4/4 Passed (Simplified)
	559	+{{code}}Quality Gates: 4/4 Passed (Simplified)
543	543	✓ Source Quality: 3 sources found
544	544	✓ Contradiction Search: Basic search completed
545	545	✓ Uncertainty: Confidence scores assigned
546		-✓ Structural Integrity: Output coherent
547		-{{/code}}
	563	+✓ Structural Integrity: Output coherent{{/code}}
548	548
549	549	**If any gate fails:**
550		-{{code}}
551		-Quality Gates: 3/4 Passed (Simplified)
	566	+{{code}}Quality Gates: 3/4 Passed (Simplified)
552	552	✓ Source Quality: 3 sources found
553	553	✗ Contradiction Search: Search failed - limited evidence
554	554	✓ Uncertainty: Confidence scores assigned
555	555	✓ Structural Integrity: Output coherent
556	556
557		-Note: This analysis has limited evidence. Use with caution.
558		-{{/code}}
	572	+Note: This analysis has limited evidence. Use with caution.{{/code}}
559	559
560	560	=== 6.7 Simplified vs. Full System ===
561	561
...	...	@@ -572,6 +572,7 @@
572	572	=== 7.1 POC AKEL (Simplified) ===
573	573
574	574	**Implementation:**
	589	+
575	575	* Single Claude API call (Sonnet 4.5)
576	576	* One comprehensive prompt
577	577	* All processing in single request
...	...	@@ -579,8 +579,7 @@
579	579	* No orchestration layer
580	580
581	581	**Prompt Structure:**
582		-{{code}}
583		-Task: Analyze this article and provide:
	597	+{{code}}Task: Analyze this article and provide:
584	584
585	585	1. Extract 3-5 factual claims
586	586	2. For each claim:
...	...	@@ -592,8 +592,7 @@
592	592	4. Generate article summary (3-5 sentences)
593	593	5. Run basic quality checks
594	594
595		-Return as structured JSON.
596		-{{/code}}
	609	+Return as structured JSON.{{/code}}
597	597
598	598	**Processing Time:** 10-18 seconds (estimate)
599	599
...	...	@@ -600,8 +600,7 @@
600	600	=== 7.2 Full System AKEL (Production) ===
601	601
602	602	**Architecture:**
603		-{{code}}
604		-AKEL Orchestrator
	616	+{{code}}AKEL Orchestrator
605	605	├── Claim Extractor
606	606	├── Claim Classifier (with risk tier assignment)
607	607	├── Scenario Generator
...	...	@@ -609,10 +609,10 @@
609	609	├── Contradiction Detector
610	610	├── Quality Gate Validator
611	611	├── Audit Sampling Scheduler
612		-└── Federation Sync Adapter (Release 1.0+)
613		-{{/code}}
	624	+└── Federation Sync Adapter (Release 1.0+){{/code}}
614	614
615	615	**Processing:**
	627	+
616	616	* Parallel processing where possible
617	617	* Separate component calls
618	618	* Quality gates between phases
...	...	@@ -624,6 +624,7 @@
624	624	=== 7.3 Why POC Uses Single Call ===
625	625
626	626	**Advantages:**
	639	+
627	627	* ✅ Simpler to implement
628	628	* ✅ Faster POC development
629	629	* ✅ Easier to debug
...	...	@@ -631,6 +631,7 @@
631	631	* ✅ Good enough for concept validation
632	632
633	633	**Limitations:**
	647	+
634	634	* ❌ No component reusability
635	635	* ❌ No parallel processing
636	636	* ❌ All-or-nothing (can't partially succeed)
...	...	@@ -657,6 +657,7 @@
657	657	**Requirement:** User can submit article for analysis
658	658
659	659	**Functionality:**
	674	+
660	660	* Text input field (paste article text, up to 5000 characters)
661	661	* URL input field (paste article URL)
662	662	* "Analyze" button to trigger processing
...	...	@@ -663,6 +663,7 @@
663	663	* Loading indicator during analysis
664	664
665	665	**Excluded:**
	681	+
666	666	* No user authentication
667	667	* No claim history
668	668	* No search functionality
...	...	@@ -669,6 +669,7 @@
669	669	* No saved templates
670	670
671	671	**Acceptance Criteria:**
	688	+
672	672	* User can paste text from article
673	673	* User can paste URL of article
674	674	* System accepts input and triggers analysis
...	...	@@ -678,6 +678,7 @@
678	678	**Requirement:** AI automatically extracts 3-5 factual claims
679	679
680	680	**Functionality:**
	698	+
681	681	* AI reads article text
682	682	* AI identifies factual claims (not opinions/questions)
683	683	* AI extracts 3-5 most important claims
...	...	@@ -684,6 +684,7 @@
684	684	* System displays numbered list
685	685
686	686	**Critical:** NO MANUAL EDITING ALLOWED
	705	+
687	687	* AI selects which claims to extract
688	688	* AI identifies factual vs. non-factual
689	689	* System processes claims as extracted
...	...	@@ -690,11 +690,13 @@
690	690	* No human curation or correction
691	691
692	692	**Error Handling:**
	712	+
693	693	* If extraction fails: Display error message
694	694	* User can retry with different input
695	695	* No manual intervention to fix extraction
696	696
697	697	**Acceptance Criteria:**
	718	+
698	698	* AI extracts 3-5 claims automatically
699	699	* Claims are factual (not opinions)
700	700	* Claims are clearly stated
...	...	@@ -705,15 +705,17 @@
705	705	**Requirement:** AI automatically generates verdict for each claim
706	706
707	707	**Functionality:**
	729	+
708	708	* For each claim, AI:
709		- * Evaluates claim based on available evidence/knowledge
710		- * Determines verdict: WELL-SUPPORTED / PARTIALLY SUPPORTED / UNCERTAIN / REFUTED
711		- * Assigns confidence score (0-100%)
712		- * Assigns risk tier (A/B/C)
713		- * Writes brief reasoning (1-3 sentences)
	731	+* Evaluates claim based on available evidence/knowledge
	732	+* Determines verdict: WELL-SUPPORTED / PARTIALLY SUPPORTED / UNCERTAIN / REFUTED
	733	+* Assigns confidence score (0-100%)
	734	+* Assigns risk tier (A/B/C)
	735	+* Writes brief reasoning (1-3 sentences)
714	714	* System displays verdict for each claim
715	715
716	716	**Critical:** NO MANUAL EDITING ALLOWED
	739	+
717	717	* AI computes verdicts based on evidence
718	718	* AI generates confidence scores
719	719	* AI writes reasoning
...	...	@@ -720,11 +720,13 @@
720	720	* No human review or adjustment
721	721
722	722	**Error Handling:**
	746	+
723	723	* If verdict generation fails: Display error message
724	724	* User can retry
725	725	* No manual intervention to adjust verdicts
726	726
727	727	**Acceptance Criteria:**
	752	+
728	728	* Each claim has a verdict
729	729	* Confidence score is displayed (0-100%)
730	730	* Risk tier is displayed (A/B/C)
...	...	@@ -737,15 +737,17 @@
737	737	**Requirement:** AI generates brief summary of analysis
738	738
739	739	**Functionality:**
	765	+
740	740	* AI summarizes findings in 3-5 sentences:
741		- * How many claims found
742		- * Distribution of verdicts
743		- * Overall assessment
	767	+* How many claims found
	768	+* Distribution of verdicts
	769	+* Overall assessment
744	744	* System displays at top of results
745	745
746	746	**Critical:** NO MANUAL EDITING ALLOWED
747	747
748	748	**Acceptance Criteria:**
	775	+
749	749	* Summary is coherent
750	750	* Accurately reflects analysis
751	751	* 3-5 sentences
...	...	@@ -756,6 +756,7 @@
756	756	**Requirement:** AI generates brief summary of original article
757	757
758	758	**Functionality:**
	786	+
759	759	* AI summarizes article content (not FactHarbor's analysis)
760	760	* 3-5 sentences
761	761	* System displays
...	...	@@ -765,6 +765,7 @@
765	765	**Critical:** NO MANUAL EDITING ALLOWED
766	766
767	767	**Acceptance Criteria:**
	796	+
768	768	* Summary is neutral (article's position)
769	769	* Accurately reflects article content
770	770	* 3-5 sentences
...	...	@@ -775,6 +775,7 @@
775	775	**Requirement:** Clear labeling of AI-generated content
776	776
777	777	**Functionality:**
	807	+
778	778	* Display Mode 2 publication label
779	779	* Show POC/Demo disclaimer
780	780	* Display risk tiers per claim
...	...	@@ -782,6 +782,7 @@
782	782	* Display timestamp
783	783
784	784	**Acceptance Criteria:**
	815	+
785	785	* Label is prominent and clear
786	786	* User understands this is AI-generated POC output
787	787	* Risk tiers are color-coded
...	...	@@ -792,6 +792,7 @@
792	792	**Requirement:** Execute simplified quality gates
793	793
794	794	**Functionality:**
	826	+
795	795	* Check source quality (basic)
796	796	* Attempt contradiction search (basic)
797	797	* Calculate confidence scores
...	...	@@ -799,6 +799,7 @@
799	799	* Display gate results
800	800
801	801	**Acceptance Criteria:**
	834	+
802	802	* All 4 gates attempted
803	803	* Pass/fail status displayed
804	804	* Failures explained to user
...	...	@@ -813,6 +813,7 @@
813	813	**Critical Rule:** NO MANUAL EDITING AT ANY STAGE
814	814
815	815	**What this means:**
	849	+
816	816	* Claims: AI selects (no human curation)
817	817	* Scenarios: N/A (deferred to POC2)
818	818	* Evidence: AI evaluates (no human selection)
...	...	@@ -820,13 +820,12 @@
820	820	* Summaries: AI writes (no human editing)
821	821
822	822	**Pipeline:**
823		-{{code}}
824		-User Input → AKEL Processing → Output Display
	857	+{{code}}User Input → AKEL Processing → Output Display
825	825	↓
826		- ZERO human editing
827		-{{/code}}
	859	+ ZERO human editing{{/code}}
828	828
829	829	**If AI output is poor:**
	862	+
830	830	* ❌ Do NOT manually fix it
831	831	* ✅ Document the failure
832	832	* ✅ Improve prompts and retry
...	...	@@ -833,6 +833,7 @@
833	833	* ✅ Accept that POC might fail
834	834
835	835	**Why this matters:**
	869	+
836	836	* Tests whether AI can do this without humans
837	837	* Validates scalability (humans can't review every analysis)
838	838	* Honest test of technical feasibility
...	...	@@ -842,16 +842,19 @@
842	842	**Requirement:** Analysis completes in reasonable time
843	843
844	844	**Acceptable Performance:**
	879	+
845	845	* Processing time: 1-5 minutes (acceptable for POC)
846	846	* Display loading indicator to user
847	847	* Show progress if possible ("Extracting claims...", "Generating verdicts...")
848	848
849	849	**Not Required:**
	885	+
850	850	* Production-level speed (< 30 seconds)
851	851	* Optimization for scale
852	852	* Caching
853	853
854	854	**Acceptance Criteria:**
	891	+
855	855	* Analysis completes within 5 minutes
856	856	* User sees loading indicator
857	857	* No timeout errors
...	...	@@ -861,16 +861,19 @@
861	861	**Requirement:** System works for manual testing sessions
862	862
863	863	**Acceptable:**
	901	+
864	864	* Occasional errors (< 20% failure rate)
865	865	* Manual restart if needed
866	866	* Display error messages clearly
867	867
868	868	**Not Required:**
	907	+
869	869	* 99.9% uptime
870	870	* Automatic error recovery
871	871	* Production monitoring
872	872
873	873	**Acceptance Criteria:**
	913	+
874	874	* System works for test demonstrations
875	875	* Errors are handled gracefully
876	876	* User receives clear error messages
...	...	@@ -880,6 +880,7 @@
880	880	**Requirement:** Runs on simple infrastructure
881	881
882	882	**Acceptable:**
	923	+
883	883	* Single machine or simple cloud setup
884	884	* No distributed architecture
885	885	* No load balancing
...	...	@@ -887,6 +887,7 @@
887	887	* Local development environment viable
888	888
889	889	**Not Required:**
	931	+
890	890	* Production infrastructure
891	891	* Multi-region deployment
892	892	* Auto-scaling
...	...	@@ -897,6 +897,7 @@
897	897	**Requirement:** Track and display LLM usage metrics to inform optimization decisions
898	898
899	899	**Must Track:**
	942	+
900	900	* Input tokens (article + prompt)
901	901	* Output tokens (generated analysis)
902	902	* Total tokens
...	...	@@ -905,16 +905,19 @@
905	905	* Article length (words/characters)
906	906
907	907	**Must Display:**
	951	+
908	908	* Usage statistics in UI (Component 5)
909	909	* Cost per analysis
910	910	* Cost per claim extracted
911	911
912	912	**Must Log:**
	957	+
913	913	* Aggregate metrics for analysis
914	914	* Cost distribution by article length
915	915	* Token efficiency trends
916	916
917	917	**Purpose:**
	963	+
918	918	* Understand unit economics
919	919	* Identify optimization opportunities
920	920	* Project costs at scale
...	...	@@ -921,6 +921,7 @@
921	921	* Inform architecture decisions (caching, model selection, etc.)
922	922
923	923	**Acceptance Criteria:**
	970	+
924	924	* ✅ Usage data displayed after each analysis
925	925	* ✅ Metrics logged for aggregate analysis
926	926	* ✅ Cost calculated accurately (Claude API pricing)
...	...	@@ -928,6 +928,7 @@
928	928	* ✅ POC1 report includes cost analysis section
929	929
930	930	**Success Target:**
	978	+
931	931	* Average cost per analysis < $0.05 USD
932	932	* Cost scaling behavior understood (linear/exponential)
933	933	* 2+ optimization opportunities identified
...	...	@@ -939,11 +939,13 @@
939	939	=== 10.1 System Components ===
940	940
941	941	**Frontend:**
	990	+
942	942	* Simple HTML form (text input + URL input + button)
943	943	* Loading indicator
944	944	* Results display page (single page, no tabs/navigation)
945	945
946	946	**Backend:**
	996	+
947	947	* Single API endpoint
948	948	* Calls Claude API (Sonnet 4.5 or latest)
949	949	* Parses response
...	...	@@ -950,10 +950,12 @@
950	950	* Returns JSON to frontend
951	951
952	952	**Data Storage:**
	1003	+
953	953	* None required (stateless POC)
954	954	* Optional: Simple file storage or SQLite for demo examples
955	955
956	956	**External Services:**
	1008	+
957	957	* Claude API (Anthropic) - required
958	958	* Optional: URL fetch service for article text extraction
959	959
...	...	@@ -986,8 +986,7 @@
986	986	=== 10.3 AI Prompt Strategy ===
987	987
988	988	**Single Comprehensive Prompt:**
989		-{{code}}
990		-Task: Analyze this article and provide:
	1041	+{{code}}Task: Analyze this article and provide:
991	991
992	992	1. Identify the article's main thesis/conclusion
993	993	- What is the article trying to argue or prove?
...	...	@@ -1023,8 +1023,7 @@
1023	1023
1024	1024	7. Write article summary (3-5 sentences: neutral summary of article content)
1025	1025
1026		-Return as structured JSON with quality gate results.
1027		-{{/code}}
	1077	+Return as structured JSON with quality gate results.{{/code}}
1028	1028
1029	1029	**One prompt generates everything.**
1030	1030
...	...	@@ -1035,25 +1035,30 @@
1035	1035	=== 10.4 Technology Stack Suggestions ===
1036	1036
1037	1037	**Frontend:**
	1088	+
1038	1038	* HTML + CSS + JavaScript (minimal framework)
1039	1039	* OR: Next.js (if team prefers)
1040	1040	* Hosted: Local machine OR Vercel/Netlify free tier
1041	1041
1042	1042	**Backend:**
	1094	+
1043	1043	* Python Flask/FastAPI (simple REST API)
1044	1044	* OR: Next.js API routes (if using Next.js)
1045	1045	* Hosted: Local machine OR Railway/Render free tier
1046	1046
1047	1047	**AKEL Integration:**
	1100	+
1048	1048	* Claude API via Anthropic SDK
1049	1049	* Model: Claude Sonnet 4.5 or latest available
1050	1050
1051	1051	**Database:**
	1105	+
1052	1052	* None (stateless acceptable)
1053	1053	* OR: SQLite if want to store demo examples
1054	1054	* OR: JSON files on disk
1055	1055
1056	1056	**Deployment:**
	1111	+
1057	1057	* Local development environment sufficient for POC
1058	1058	* Optional: Deploy to cloud for remote demos
1059	1059
...	...	@@ -1062,6 +1062,7 @@
1062	1062	=== 11.1 Minimum Success (POC Passes) ===
1063	1063
1064	1064	**Required for GO decision:**
	1120	+
1065	1065	* ✅ AI extracts 3-5 factual claims automatically
1066	1066	* ✅ AI provides verdict for each claim automatically
1067	1067	* ✅ Verdicts are reasonable (≥70% make logical sense)
...	...	@@ -1075,6 +1075,7 @@
1075	1075	* ✅ **Optimization opportunities identified** (≥2 potential improvements documented)
1076	1076
1077	1077	**Quality Definition:**
	1134	+
1078	1078	* "Reasonable verdict" = Defensible given general knowledge
1079	1079	* "Coherent summary" = Logically structured, grammatically correct
1080	1080	* "Comprehensible" = Reviewers understand what analysis means
...	...	@@ -1082,6 +1082,7 @@
1082	1082	=== 11.2 POC Fails If ===
1083	1083
1084	1084	**Automatic NO-GO if any of these:**
	1142	+
1085	1085	* ❌ Claim extraction poor (< 60% accuracy - extracts non-claims or misses obvious ones)
1086	1086	* ❌ Verdicts nonsensical (< 60% reasonable - contradictory or random)
1087	1087	* ❌ Output incomprehensible (reviewers can't understand analysis)
...	...	@@ -1093,14 +1093,15 @@
1093	1093	**POC quality expectations:**
1094	1094
1095	1095	|=Component|=Quality Threshold|=Definition
1096		-|Claim Extraction|(% class="success" %)≥70% accuracy(%%) |Identifies obvious factual claims, may miss some edge cases
1097		-|Verdict Logic|(% class="success" %)≥70% defensible(%%) |Verdicts are logical given reasoning provided
1098		-|Reasoning Clarity|(% class="success" %)≥70% clear(%%) |1-3 sentences are understandable and relevant
1099		-|Overall Analysis|(% class="success" %)≥70% useful(%%) |Output helps user understand article claims
	1154	+|Claim Extraction|(% class="success" %)≥70% accuracy |Identifies obvious factual claims, may miss some edge cases
	1155	+|Verdict Logic|(% class="success" %)≥70% defensible |Verdicts are logical given reasoning provided
	1156	+|Reasoning Clarity|(% class="success" %)≥70% clear |1-3 sentences are understandable and relevant
	1157	+|Overall Analysis|(% class="success" %)≥70% useful |Output helps user understand article claims
1100	1100
1101	1101	**Analogy:** "B student" quality (70-80%), not "A+" perfection yet
1102	1102
1103	1103	**Not expecting:**
	1162	+
1104	1104	* 100% accuracy
1105	1105	* Perfect claim coverage
1106	1106	* Comprehensive evidence gathering
...	...	@@ -1108,6 +1108,7 @@
1108	1108	* Production polish
1109	1109
1110	1110	**Expecting:**
	1170	+
1111	1111	* Reasonable claim extraction
1112	1112	* Defensible verdicts
1113	1113	* Understandable reasoning
...	...	@@ -1120,6 +1120,7 @@
1120	1120	**Input:** "Coffee reduces the risk of type 2 diabetes by 30%"
1121	1121
1122	1122	**Expected Output:**
	1183	+
1123	1123	* Extract claim correctly
1124	1124	* Provide verdict: WELL-SUPPORTED or PARTIALLY SUPPORTED
1125	1125	* Confidence: 70-90%
...	...	@@ -1133,6 +1133,7 @@
1133	1133	**Input:** News article URL with multiple claims about politics/health/science
1134	1134
1135	1135	**Expected Output:**
	1197	+
1136	1136	* Extract 3-5 key claims
1137	1137	* Verdict for each (may vary: some supported, some uncertain, some refuted)
1138	1138	* Coherent analysis summary
...	...	@@ -1146,6 +1146,7 @@
1146	1146	**Input:** Article on contested political or scientific topic
1147	1147
1148	1148	**Expected Output:**
	1211	+
1149	1149	* Balanced analysis
1150	1150	* Acknowledges uncertainty where appropriate
1151	1151	* Doesn't overstate confidence
...	...	@@ -1158,6 +1158,7 @@
1158	1158	**Input:** Article with obviously false claim (e.g., "The Earth is flat")
1159	1159
1160	1160	**Expected Output:**
	1224	+
1161	1161	* Extract claim
1162	1162	* Verdict: REFUTED
1163	1163	* High confidence (> 90%)
...	...	@@ -1171,6 +1171,7 @@
1171	1171	**Input:** Article with claim where evidence is genuinely mixed
1172	1172
1173	1173	**Expected Output:**
	1238	+
1174	1174	* Extract claim
1175	1175	* Verdict: UNCERTAIN
1176	1176	* Moderate confidence (40-60%)
...	...	@@ -1183,6 +1183,7 @@
1183	1183	**Input:** Article making medical claims
1184	1184
1185	1185	**Expected Output:**
	1251	+
1186	1186	* Extract claim
1187	1187	* Verdict: [appropriate based on evidence]
1188	1188	* Risk tier: A (High - medical)
...	...	@@ -1200,6 +1200,7 @@
1200	1200	**Option A: GO (Proceed to POC2)**
1201	1201
1202	1202	**Conditions:**
	1269	+
1203	1203	* AI quality ≥70% without manual editing
1204	1204	* Basic claim → verdict pipeline validated
1205	1205	* Internal + advisor feedback positive
...	...	@@ -1208,6 +1208,7 @@
1208	1208	* Clear path to improving AI quality to ≥90%
1209	1209
1210	1210	**Next Steps:**
	1278	+
1211	1211	* Plan POC2 development (add scenarios)
1212	1212	* Design scenario architecture
1213	1213	* Expand to Evidence Model structure
...	...	@@ -1216,6 +1216,7 @@
1216	1216	**Option B: NO-GO (Pivot or Stop)**
1217	1217
1218	1218	**Conditions:**
	1287	+
1219	1219	* AI quality < 60%
1220	1220	* Requires manual editing for most analyses (> 50%)
1221	1221	* Feedback indicates fundamental flaws
...	...	@@ -1223,6 +1223,7 @@
1223	1223	* No clear path to improvement
1224	1224
1225	1225	**Next Steps:**
	1295	+
1226	1226	* **Pivot:** Change to hybrid human-AI approach (accept manual review required)
1227	1227	* **Stop:** Conclude approach not viable, revisit later
1228	1228
...	...	@@ -1229,6 +1229,7 @@
1229	1229	**Option C: ITERATE (Improve POC)**
1230	1230
1231	1231	**Conditions:**
	1302	+
1232	1232	* Concept has merit but execution needs work
1233	1233	* Specific improvements identified
1234	1234	* Addressable with better prompts/approach
...	...	@@ -1235,6 +1235,7 @@
1235	1235	* AI quality between 60-70%
1236	1236
1237	1237	**Next Steps:**
	1309	+
1238	1238	* Improve AI prompts
1239	1239	* Test different approaches
1240	1240	* Re-run POC with improvements
...	...	@@ -1256,6 +1256,7 @@
1256	1256	**Impact:** POC fails
1257	1257
1258	1258	**Mitigation:**
	1331	+
1259	1259	* Extensive prompt engineering and testing
1260	1260	* Use best available AI models (Sonnet 4.5)
1261	1261	* Test with diverse article types
...	...	@@ -1269,6 +1269,7 @@
1269	1269	**Impact:** Works sometimes, fails other times
1270	1270
1271	1271	**Mitigation:**
	1345	+
1272	1272	* Test with 10+ diverse articles
1273	1273	* Measure success rate honestly
1274	1274	* Improve prompts to increase consistency
...	...	@@ -1281,6 +1281,7 @@
1281	1281	**Impact:** Users can't understand analysis
1282	1282
1283	1283	**Mitigation:**
	1358	+
1284	1284	* Create clear explainer document
1285	1285	* Iterate on output format
1286	1286	* Test with non-technical reviewers
...	...	@@ -1294,6 +1294,7 @@
1294	1294	**Impact:** System slow or expensive
1295	1295
1296	1296	**Mitigation:**
	1372	+
1297	1297	* Monitor API usage
1298	1298	* Implement retry logic
1299	1299	* Estimate costs before scaling
...	...	@@ -1306,6 +1306,7 @@
1306	1306	**Impact:** POC becomes too complex
1307	1307
1308	1308	**Mitigation:**
	1385	+
1309	1309	* Strict scope discipline
1310	1310	* Say NO to feature additions
1311	1311	* Keep focus on core question
...	...	@@ -1316,12 +1316,15 @@
1316	1316
1317	1317	=== 15.1 Core Principles ===
1318	1318
1319		-**1. Build Less, Learn More**
	1396	+*
	1397	+**
	1398	+**1. Build Less, Learn More
1320	1320	* Minimum features to test hypothesis
1321	1321	* Don't build unvalidated features
1322	1322	* Focus on core question only
1323	1323
1324	1324	**2. Fail Fast**
	1404	+
1325	1325	* Quick test of hardest part (AI capability)
1326	1326	* Accept that POC might fail
1327	1327	* Better to discover issues early
...	...	@@ -1328,16 +1328,19 @@
1328	1328	* Honest assessment over optimistic hope
1329	1329
1330	1330	**3. Test First, Build Second**
	1411	+
1331	1331	* Validate AI can do this before building platform
1332	1332	* Don't assume it will work
1333	1333	* Let results guide decisions
1334	1334
1335	1335	**4. Automation First**
	1417	+
1336	1336	* No manual editing allowed
1337	1337	* Tests scalability, not just feasibility
1338	1338	* Proves approach can work at scale
1339	1339
1340	1340	**5. Honest Assessment**
	1423	+
1341	1341	* Don't cherry-pick examples
1342	1342	* Don't manually fix bad outputs
1343	1343	* Document failures openly
...	...	@@ -1358,9 +1358,12 @@
1358	1358	❌ Perfectly accurate analysis
1359	1359	❌ Polished user experience
1360	1360
1361		-== 16. Success = Clear Path Forward ==
	1444	+== 16. Success ==
1362	1362
	1446	+ Clear Path Forward ==
	1447	+
1363	1363	**If POC succeeds (≥70% AI quality):**
	1449	+
1364	1364	* ✅ Approach validated
1365	1365	* ✅ Proceed to POC2 (add scenarios)
1366	1366	* ✅ Design full Evidence Model structure
...	...	@@ -1368,6 +1368,7 @@
1368	1368	* ✅ Focus on improving AI quality from 70% → 90%
1369	1369
1370	1370	**If POC fails (< 60% AI quality):**
	1457	+
1371	1371	* ✅ Learn what doesn't work
1372	1372	* ✅ Pivot to different approach
1373	1373	* ✅ OR wait for better AI technology
...	...	@@ -1394,16 +1394,16 @@
1394	1394	**POC1 Implementation:**
1395	1395
1396	1396	* **Primary Provider:** Anthropic Claude API
1397		- * Stage 1: Claude Haiku 4
1398		- * Stage 2: Claude Sonnet 3.5 (cached)
1399		- * Stage 3: Claude Sonnet 3.5
	1484	+* Stage 1: Claude Haiku 4
	1485	+* Stage 2: Claude Sonnet 3.5 (cached)
	1486	+* Stage 3: Claude Sonnet 3.5
1400	1400
1401	1401	* **Provider Interface:** Abstract LLMProvider interface implemented
1402	1402
1403	1403	* **Configuration:** Environment variables for provider selection
1404		- * {{code}}LLM_PRIMARY_PROVIDER=anthropic{{/code}}
1405		- * {{code}}LLM_STAGE1_MODEL=claude-haiku-4{{/code}}
1406		- * {{code}}LLM_STAGE2_MODEL=claude-sonnet-3-5{{/code}}
	1491	+* {{code}}LLM_PRIMARY_PROVIDER=anthropic{{/code}}
	1492	+* {{code}}LLM_STAGE1_MODEL=claude-haiku-4{{/code}}
	1493	+* {{code}}LLM_STAGE2_MODEL=claude-sonnet-3-5{{/code}}
1407	1407
1408	1408	* **Failover:** Basic error handling with cache fallback for Stage 2
1409	1409
...	...	@@ -1423,9 +1423,10 @@
1423	1423	* Cost tracking includes provider name in logs
1424	1424	* Stage 2 falls back to cache on provider failure
1425	1425
1426		-**Implementation:** See [[POC1 API & Schemas Specification>>Test.FactHarbor.Specification.POC.API-and-Schemas.WebHome]] Section 6
	1513	+**Implementation:** See [[POC1 API & Schemas Specification>>Test.FactHarbor V0\.9\.105.Specification.POC.API-and-Schemas.WebHome]] Section 6
1427	1427
1428	1428	**Dependencies:**
	1516	+
1429	1429	* NFR-14 (Main Requirements)
1430	1430	* Design Decision 9
1431	1431	* Architecture Section 2.2
...	...	@@ -1433,5 +1433,3 @@
1433	1433	**Priority:** HIGH (P1)
1434	1434
1435	1435	**Rationale:** Even though POC1 uses single provider, abstraction must be in place from start to avoid costly refactoring later.
1436		-
1437		-