POC Requirements (POC1 & POC2)

1

= POC Requirements =

2

3

**Status:** ✅ Approved for Development

4

**Version:** 3.0 (Aligned with Main Requirements)

5

**Goal:** Prove that AI can extract claims and determine verdicts automatically without human intervention

6

7

8

**Core Philosophy:** POC validates the [[Main Requirements>>FactHarbor.Specification.Requirements.WebHome]] through simplified implementation. All POC features map to formal FR/NFR requirements.

== 1. POC Overview ==

13

14

=== 1.1 What POC Tests ===

**Core Question:**

> Can AI automatically extract factual claims from articles and evaluate them with reasonable verdicts?

19

20

**What we're proving:**

21

22

* AI can identify factual claims from text

23

* AI can evaluate those claims with structured evidence

24

* Quality gates can filter unreliable outputs

25

* The core workflow is technically feasible

26

27

**What we're NOT proving:**

28

29

* Production-ready reliability (that's POC2)

30

* User-facing features (that's Beta 0)

31

* Full IFCN compliance (that's V1.0)

32

33

=== 1.2 Requirements Mapping ===

34

35

POC1 implements a **subset** of the full system requirements defined in [[Main Requirements>>FactHarbor.Specification.Requirements.WebHome]].

**Scope Summary:**

* **In Scope:** 8 requirements (7 FRs + 1 NFR)

40

* **Partial:** 3 NFRs (simplified versions)

41

* **Out of Scope:** 19 requirements (deferred to later phases)

== 2. POC1 Scope ==

**Authoritative Source for Phase Mapping:** [[Requirements Roadmap Matrix>>Test.FactHarbor V0\.9\.88.Roadmap.Requirements-Roadmap-Matrix.WebHome]]

47

48

The Roadmap Matrix is the single source of truth for which requirements are implemented in which phases. This page provides POC1-specific implementation details only.

49

50

51

**POC1 implements these formal requirements:**

52

53

|= Formal Req |= Implementation in POC1 |= Notes

54

| **FR4** | Analysis Summary | Basic format; quality metadata deferred to POC2

55

| **FR7** | Automated Verdicts | Full implementation with quality gates (NFR11)

56

| **NFR11** | Quality Assurance Framework | 4 quality gates implemented

57

58

**POC1 also implements these workflow components** (detailed as FR1-FR6 in implementation sections below)

59

60

{{info}}**Note:** FR11 (Audit Trail) and FR13 (In-Article Claim Highlighting) are deferred to Beta 0 for production readiness and user experience enhancement.{{/info}}:

61

62

* Claim extraction (FR1)

63

* Claim context (FR2)

64

* Multiple scenarios (FR3)

65

* Evidence collection (FR5)

66

* Source quality assessment (FR6)

67

* Time evolution tracking (FR8) - deferred to POC2

68

* Audit trail (FR11) - deferred to Beta 0

69

* In-article highlighting (FR13) - deferred to Beta 0

70

71

**Partial implementations:**

72

73

* NFR1 (Explainability) - Basic only

74

* NFR2 (Performance) - Functional but not optimized

75

* NFR3 (Transparency) - Basic only

76

77

**Detailed POC1 implementation specifications continue below...**

== 3. POC Simplifications ==

82

83

=== 3.1 FR1: Claim Extraction (Full Implementation) ===

84

85

**Main Requirement:** AI extracts factual claims from input text

86

87

**POC Implementation:**

88

89

* ✅ AKEL extracts claims using LLM

90

* ✅ Each claim includes original text reference

91

* ✅ Claims are identified as factual/non-factual

92

* ❌ No advanced claim parsing (added in POC2)

93

94

**Acceptance Criteria:**

95

96

* Extracts 3-5 claims from typical article

97

* Identifies factual vs non-factual claims

98

* Quality Gate 1 validates extraction

99

100

=== 3.2 FR3: Multiple Scenarios (Full Implementation) ===

101

102

**Main Requirement:** Generate multiple interpretation scenarios for ambiguous claims

103

104

**POC Implementation:**

105

106

* ✅ AKEL generates 2-3 scenarios per claim

107

* ✅ Scenarios capture different interpretations

108

* ✅ Each scenario is evaluated separately

109

* ✅ Verdict considers all scenarios

110

111

**Acceptance Criteria:**

112

113

* Generates 2+ scenarios for ambiguous claims

114

* Scenarios are meaningfully different

115

* All scenarios are evaluated

116

117

=== 3.3 FR4: Analysis Summary (Basic Implementation) ===

118

119

**Main Requirement:** Provide user-friendly summary of analysis

120

121

**POC Implementation:**

122

123

* ✅ Simple text summary generated

124

* ❌ No rich formatting (added in Beta 0)

125

* ❌ No visual elements (added in Beta 0)

126

* ❌ No interactive features (added in Beta 0)

**POC Format:**

```

Claim: [extracted claim]

131

Scenarios: [list of scenarios]

132

Evidence: [supporting/opposing evidence]

133

Verdict: [probability with uncertainty]

```

=== 3.4 FR5-FR6: Evidence Collection & Evaluation (Full Implementation) ===

138

139

**Main Requirements:**

140

141

* FR5: Collect supporting and opposing evidence

142

* FR6: Evaluate evidence source reliability

143

144

**POC Implementation:**

145

146

* ✅ AKEL searches for evidence (web/knowledge base)

147

* ✅ **Mandatory contradiction search** (finds opposing evidence)

148

* ✅ Source reliability scoring

149

* ❌ No evidence deduplication (added in POC2)

150

* ❌ No advanced source verification (added in POC2)

151

152

**Acceptance Criteria:**

153

154

* Finds 2+ supporting evidence items

155

* Finds 1+ opposing evidence (if exists)

156

* Sources scored for reliability

157

158

=== 3.5 FR7: Automated Verdicts (Full Implementation) ===

159

160

**Main Requirement:** AI computes verdicts with uncertainty quantification

161

162

**POC Implementation:**

163

164

* ✅ Probabilistic verdicts (0-100% confidence)

165

* ✅ Uncertainty explicitly stated

166

* ✅ Reasoning chain provided

167

* ✅ Quality Gate 4 validates verdict confidence

**POC Output:**

```

Verdict: 70% likely true

172

Uncertainty: ±15% (moderate confidence)

173

Reasoning: Based on 3 high-quality sources...

174

Confidence Level: MEDIUM

175

```

176

177

**Acceptance Criteria:**

178

179

* Verdicts include probability (0-100%)

180

* Uncertainty explicitly quantified

181

* Reasoning chain explains verdict

182

183

=== 3.6 NFR11: Quality Assurance Framework (LITE VERSION) ===

184

185

**Main Requirement:** Complete quality assurance with 7 quality gates

186

187

**POC Implementation:** **2 gates only**

188

189

**Quality Gate 1: Claim Validation**

190

191

* ✅ Validates claim is factual and verifiable

192

* ✅ Blocks non-factual claims (opinion/prediction/ambiguous)

193

* ✅ Provides clear rejection reason

194

195

**Quality Gate 4: Verdict Confidence Assessment**

196

197

* ✅ Validates ≥2 sources found

198

* ✅ Validates quality score ≥0.6

199

* ✅ Blocks low-confidence verdicts

200

* ✅ Provides clear rejection reason

201

202

**Out of Scope (POC2+):**

203

204

* ❌ Gate 2: Evidence Relevance

205

* ❌ Gate 3: Scenario Coherence

206

* ❌ Gate 5: Source Diversity

207

* ❌ Gate 6: Reasoning Validity

208

* ❌ Gate 7: Output Completeness

209

210

**Rationale:** Prove gate concept works. Add remaining gates in POC2 after validating approach.

211

212

213

=== 3.7 NFR1-3: Performance, Scalability, Reliability (Basic) ===

214

215

**Main Requirements:**

216

217

* NFR1: Response time < 30 seconds

218

* NFR2: Handle 1000+ concurrent users

219

* NFR3: 99.9% uptime

220

221

**POC Implementation:**

222

223

* ⚠️ **Response time monitored** (not optimized)

224

* ⚠️ **Single-threaded processing** (no concurrency)

225

* ⚠️ **Basic error handling** (no advanced retry logic)

226

227

**Rationale:** POC proves functionality. Performance optimization happens in POC2.

**POC Acceptance:**

* Analysis completes (no timeout requirement)

232

* Errors don't crash system

233

* Basic logging in place

234

235

== 4. What's NOT in POC Scope ==

236

237

=== 4.1 User-Facing Features (Beta 0+) ===

238

239

240

**Deferred to Beta 0:**

**Out of Scope:**

* ❌ User accounts and authentication (FR8)

246

* ❌ User corrections system (FR9, FR45-46)

247

* ❌ Public publishing interface (FR10)

248

* ❌ Social sharing (FR11)

249

* ❌ Email notifications (FR12)

250

* ❌ API access (FR13)

251

252

**Rationale:** POC validates AI capabilities. User features added in Beta 0.

253

254

255

=== 4.2 Advanced Features (V1.0+) ===

**Out of Scope:**

* ❌ IFCN compliance (FR47)

260

* ❌ ClaimReview schema (FR48)

261

* ❌ Archive.org integration (FR49)

262

* ❌ OSINT toolkit (FR50)

263

* ❌ Video verification (FR51)

264

* ❌ Deepfake detection (FR52)

265

* ❌ Cross-org sharing (FR53)

266

267

**Rationale:** Advanced features require proven platform. Added post-V1.0.

268

269

270

=== 4.3 Production Requirements (POC2, Beta 0) ===

**Out of Scope:**

* ❌ Security controls (NFR4, NFR12)

275

* ❌ Code maintainability (NFR5)

276

* ❌ System monitoring (NFR13)

277

* ❌ Evidence deduplication

278

* ❌ Advanced source verification

279

* ❌ Full 7-gate quality framework

280

281

**Rationale:** POC proves concept. Production hardening happens in POC2 and Beta 0.

282

283

284

== 5. POC Output Specification ==

285

286

=== 5.1 Required Output Elements ===

287

288

For each analyzed claim, POC must produce:

*

**

**1. Claim

* Original text

* Classification (factual/non-factual/ambiguous)

295

* If non-factual: Clear reason why

296

297

**2. Scenarios** (if factual)

298

299

* 2-3 interpretation scenarios

300

* Each scenario clearly described

301

302

**3. Evidence** (if factual)

303

304

* Supporting evidence (2+ items)

305

* Opposing evidence (if exists)

306

* Source URLs and reliability scores

307

308

**4. Verdict** (if factual)

309

310

* Probability (0-100%)

311

* Uncertainty quantification

312

* Confidence level (LOW/MEDIUM/HIGH)

313

* Reasoning chain

314

315

**5. Quality Status**

316

317

* Which gates passed/failed

318

* If failed: Clear explanation why

319

320

=== 5.2 Example POC Output ===

{

"claim": {

"text": "Switzerland has the highest life expectancy in Europe",

326

"type": "factual",

327

"gate1_status": "PASS"

328

},

329

"scenarios": [

330

"Switzerland's overall life expectancy is highest",

331

"Switzerland ranks highest for specific age groups"

],

"evidence": {

"supporting": [

{

"source": "WHO Report 2023",

337

"reliability": 0.95,

338

"excerpt": "Switzerland: 83.4 years average..."

}

],

"opposing": [

{

"source": "Eurostat 2024",

344

"reliability": 0.90,

345

"excerpt": "Spain leads at 83.5 years..."

}

]

},

"verdict": {

"probability": 0.65,

"uncertainty": 0.15,

"confidence": "MEDIUM",

353

"reasoning": "WHO and Eurostat show similar but conflicting data...",

354

"gate4_status": "PASS"

}

}

== 6. Success Criteria ==

361

362

363

**POC Success Definition:** POC validates that AI can extract claims, find balanced evidence, and compute reasonable verdicts with quality gates improving output quality.

364

365

366

=== 6.1 Functional Success ===

367

368

POC is successful if:

369

370

✅ **FR1-FR7 Requirements Met:**

371

372

1. Extracts 3-5 factual claims from test articles

373

2. Generates 2-3 scenarios per ambiguous claim

374

3. Finds supporting AND opposing evidence

375

4. Computes probabilistic verdicts with uncertainty

376

5. Provides clear reasoning chains

377

378

✅ **Quality Gates Work:**

379

380

1. Gate 1 blocks non-factual claims (100% block rate)

381

2. Gate 4 blocks low-quality verdicts (blocks if <2 sources or quality <0.6)

382

3. Clear rejection reasons provided

✅ **NFR11 Met:**

1. Quality gates reduce hallucination rate

387

2. Blocked outputs have clear explanations

388

3. Quality metrics are logged

389

390

=== 6.2 Quality Thresholds ===

391

392

**Minimum Acceptable:**

393

394

* ≥70% of test claims correctly classified (factual/non-factual)

395

* ≥60% of verdicts are reasonable (human evaluation)

396

* Gate 1 blocks 100% of non-factual claims

397

* Gate 4 blocks verdicts with <2 sources

**Target:**

* ≥80% claims correctly classified

402

* ≥75% verdicts are reasonable

403

* <10% false positives (blocking good claims)

404

405

=== 6.3 POC Decision Gate ===

406

407

**After POC1, we decide:**

408

409

**✅ PROCEED to POC2** if:

410

411

* Success criteria met

412

* Quality gates demonstrably improve output

413

* Core workflow is technically sound

414

* Clear path to production quality

415

416

**⚠️ ITERATE POC1** if:

417

418

* Success criteria partially met

419

* Gates work but need tuning

420

* Core issues identified but fixable

421

422

**❌ PIVOT APPROACH** if:

423

424

* Success criteria not met

425

* Fundamental AI limitations discovered

426

* Quality gates insufficient

427

* Alternative approach needed

== 7. Test Cases ==

=== 7.1 Happy Path ===

432

433

**Test 1: Simple Factual Claim**

434

435

* Input: "Paris is the capital of France"

436

* Expected: Factual, 1 scenario, verdict 95% true

437

438

**Test 2: Ambiguous Claim**

439

440

* Input: "Switzerland has the highest income in Europe"

441

* Expected: Factual, 2-3 scenarios, verdict with uncertainty

442

443

**Test 3: Statistical Claim**

444

445

* Input: "10% of people have condition X"

446

* Expected: Factual, evidence with numbers, probabilistic verdict

447

448

=== 7.2 Edge Cases ===

**Test 4: Opinion**

* Input: "Paris is the best city"

453

* Expected: Non-factual (opinion), blocked by Gate 1

454

455

**Test 5: Prediction**

456

457

* Input: "Bitcoin will reach $100,000 next year"

458

* Expected: Non-factual (prediction), blocked by Gate 1

459

460

**Test 6: Insufficient Evidence**

461

462

* Input: Obscure factual claim with no sources

463

* Expected: Blocked by Gate 4 (<2 sources)

464

465

=== 7.3 Quality Gate Tests ===

466

467

**Test 7: Gate 1 Effectiveness**

468

469

* Input: Mix of 10 factual + 10 non-factual claims

470

* Expected: Gate 1 blocks all 10 non-factual (100% precision)

471

472

**Test 8: Gate 4 Effectiveness**

473

474

* Input: Claims with varying evidence availability

475

* Expected: Gate 4 blocks low-confidence verdicts

476

477

== 8. Technical Architecture (POC) ==

478

479

=== 8.1 Simplified Architecture ===

**POC Tech Stack:**

* **Frontend:** Simple web interface (Next.js + TypeScript)

484

* **Backend:** Single API endpoint

485

* **AI:** Claude API (Sonnet 4.5)

486

* **Database:** Local JSON files (no database)

487

* **Deployment:** Single server

488

489

**Architecture Diagram:** See [[POC1 Specification>>FactHarbor.Specification.POC.Specification]]

490

491

492

=== 8.2 AKEL Implementation ===

**POC AKEL:**

* Single-threaded processing

497

* Synchronous API calls

498

* No caching

499

* Basic error handling

500

* Console logging

501

502

**Full AKEL (POC2+):**

503

504

* Multi-threaded processing

505

* Async API calls

506

* Evidence caching

507

* Advanced error handling with retry

508

* Structured logging + monitoring

509

510

== 9. POC Philosophy ==

511

512

513

**Important:** POC validates concept, not production readiness. Focus is on proving AI can do the job, with production quality coming in later phases.

514

515

516

=== 9.1 Core Principles ===

*

**

**1. Prove Concept, Not Production

521