POC Requirements

1.1

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 ===

15

16

**Core Question:**

17

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

18

19

**What we're proving:**

20

* AI can identify factual claims from text

21

* AI can evaluate those claims with structured evidence

22

* Quality gates can filter unreliable outputs

23

* The core workflow is technically feasible

24

25

**What we're NOT proving:**

26

* Production-ready reliability (that's POC2)

27

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

28

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

29

30

=== 1.2 Requirements Mapping ===

31

32

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

33

34

**Scope Summary:**

35

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

36

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

37

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

38

39

40

== 2. Requirements Scope Matrix ==

41

42

43

**Authoritative Source:** See [[Requirements Roadmap Matrix>>Test.FactHarbor.Specification.Requirements-Roadmap-Matrix.WebHome]] for complete phase-to-requirement mapping across all phases.

44

45

46

**POC1 Scope Summary:**

47

48

POC1 implements the following requirements from the [[Main Requirements>>Test.FactHarbor.Specification.Requirements.WebHome]]:

49

50

**Full Implementation (8 requirements):**

51

* FR1: Claim Extraction

52

* FR2: Claim Context

53

* FR3: Multiple Scenarios

54

* FR4: Analysis Summary (Basic)

55

* FR5: Evidence Collection

56

* FR6: Source Quality Assessment

57

* FR7: Automated Verdicts (with quality gates)

58

* NFR11: AKEL Quality Assurance Framework (Basic - 4 quality gates)

59

60

**Partial Implementation (3 requirements):**

61

* NFR1: Explainability (Basic explanations only)

62

* NFR2: Performance (Functional but not optimized)

63

* NFR3: Transparency (Basic transparency)

64

65

**Deferred to Later Phases:**

66

* All other requirements (see Roadmap Matrix for phase assignments)

67

68

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

69

70

71

== 3. POC Simplifications ==

72

73

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

74

75

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

76

77

**POC Implementation:**

78

* ✅ AKEL extracts claims using LLM

79

* ✅ Each claim includes original text reference

80

* ✅ Claims are identified as factual/non-factual

81

* ❌ No advanced claim parsing (added in POC2)

82

83

**Acceptance Criteria:**

84

* Extracts 3-5 claims from typical article

85

* Identifies factual vs non-factual claims

86

* Quality Gate 1 validates extraction

87

88

89

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

90

91

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

92

93

**POC Implementation:**

94

* ✅ AKEL generates 2-3 scenarios per claim

95

* ✅ Scenarios capture different interpretations

96

* ✅ Each scenario is evaluated separately

97

* ✅ Verdict considers all scenarios

98

99

**Acceptance Criteria:**

100

* Generates 2+ scenarios for ambiguous claims

101

* Scenarios are meaningfully different

102

* All scenarios are evaluated

103

104

105

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

106

107

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

108

109

**POC Implementation:**

110

* ✅ Simple text summary generated

111

* ❌ No rich formatting (added in Beta 0)

112

* ❌ No visual elements (added in Beta 0)

113

* ❌ No interactive features (added in Beta 0)

**POC Format:**

```

Claim: [extracted claim]

118

Scenarios: [list of scenarios]

119

Evidence: [supporting/opposing evidence]

120

Verdict: [probability with uncertainty]

```

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

125

126

**Main Requirements:**

127

* FR5: Collect supporting and opposing evidence

128

* FR6: Evaluate evidence source reliability

129

130

**POC Implementation:**

131

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

132

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

133

* ✅ Source reliability scoring

134

* ❌ No evidence deduplication (added in POC2)

135

* ❌ No advanced source verification (added in POC2)

136

137

**Acceptance Criteria:**

138

* Finds 2+ supporting evidence items

139

* Finds 1+ opposing evidence (if exists)

140

* Sources scored for reliability

141

142

143

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

144

145

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

146

147

**POC Implementation:**

148

* ✅ Probabilistic verdicts (0-100% confidence)

149

* ✅ Uncertainty explicitly stated

150

* ✅ Reasoning chain provided

151

* ✅ Quality Gate 4 validates verdict confidence

**POC Output:**

```

Verdict: 70% likely true

156

Uncertainty: ±15% (moderate confidence)

157

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

158

Confidence Level: MEDIUM

159

```

160

161

**Acceptance Criteria:**

162

* Verdicts include probability (0-100%)

163

* Uncertainty explicitly quantified

164

* Reasoning chain explains verdict

165

166

167

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

168

169

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

170

171

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

172

173

**Quality Gate 1: Claim Validation**

174

* ✅ Validates claim is factual and verifiable

175

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

176

* ✅ Provides clear rejection reason

177

178

**Quality Gate 4: Verdict Confidence Assessment**

179

* ✅ Validates ≥2 sources found

180

* ✅ Validates quality score ≥0.6

181

* ✅ Blocks low-confidence verdicts

182

* ✅ Provides clear rejection reason

183

184

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

185

* ❌ Gate 2: Evidence Relevance

186

* ❌ Gate 3: Scenario Coherence

187

* ❌ Gate 5: Source Diversity

188

* ❌ Gate 6: Reasoning Validity

189

* ❌ Gate 7: Output Completeness

190

191

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

192

193

194

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

195

196

**Main Requirements:**

197

* NFR1: Response time < 30 seconds

198

* NFR2: Handle 1000+ concurrent users

199

* NFR3: 99.9% uptime

200

201

**POC Implementation:**

202

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

203

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

204

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

205

206

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

207

208

**POC Acceptance:**

209

* Analysis completes (no timeout requirement)

210

* Errors don't crash system

211

* Basic logging in place

212

213

214

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

215

216

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

217

218

219

**Deferred to Beta 0:**

**Out of Scope:**

* ❌ User accounts and authentication (FR8)

224

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

225

* ❌ Public publishing interface (FR10)

226

* ❌ Social sharing (FR11)

227

* ❌ Email notifications (FR12)

228

* ❌ API access (FR13)

229

230

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

231

232

233

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

234

235

**Out of Scope:**

236

* ❌ IFCN compliance (FR47)

237

* ❌ ClaimReview schema (FR48)

238

* ❌ Archive.org integration (FR49)

239

* ❌ OSINT toolkit (FR50)

240

* ❌ Video verification (FR51)

241

* ❌ Deepfake detection (FR52)

242

* ❌ Cross-org sharing (FR53)

243

244

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

245

246

247

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

248

249

**Out of Scope:**

250

* ❌ Security controls (NFR4, NFR12)

251

* ❌ Code maintainability (NFR5)

252

* ❌ System monitoring (NFR13)

253

* ❌ Evidence deduplication

254

* ❌ Advanced source verification

255

* ❌ Full 7-gate quality framework

256

257

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

258

259

260

== 5. POC Output Specification ==

261

262

=== 5.1 Required Output Elements ===

263

264

For each analyzed claim, POC must produce:

**1. Claim**

* Original text

* Classification (factual/non-factual/ambiguous)

269

* If non-factual: Clear reason why

270

271

**2. Scenarios** (if factual)

272

* 2-3 interpretation scenarios

273

* Each scenario clearly described

274

275

**3. Evidence** (if factual)

276

* Supporting evidence (2+ items)

277

* Opposing evidence (if exists)

278

* Source URLs and reliability scores

279

280

**4. Verdict** (if factual)

281

* Probability (0-100%)

282

* Uncertainty quantification

283

* Confidence level (LOW/MEDIUM/HIGH)

284

* Reasoning chain

285

286

**5. Quality Status**

287

* Which gates passed/failed

288

* If failed: Clear explanation why

289

290

291

=== 5.2 Example POC Output ===

{

"claim": {

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

297

"type": "factual",

298

"gate1_status": "PASS"

299

},

300

"scenarios": [

301

"Switzerland's overall life expectancy is highest",

302

"Switzerland ranks highest for specific age groups"

],

"evidence": {

"supporting": [

{

"source": "WHO Report 2023",

308

"reliability": 0.95,

309

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

}

],

"opposing": [

{

"source": "Eurostat 2024",

315

"reliability": 0.90,

316

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

}

]

},

"verdict": {

"probability": 0.65,

"uncertainty": 0.15,

"confidence": "MEDIUM",

324

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

325

"gate4_status": "PASS"

}

}

== 6. Success Criteria ==

332

333

334

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

335

336

337

=== 6.1 Functional Success ===

338

339

POC is successful if:

340

341

✅ **FR1-FR7 Requirements Met:**

342

1. Extracts 3-5 factual claims from test articles

343

2. Generates 2-3 scenarios per ambiguous claim

344

3. Finds supporting AND opposing evidence

345

4. Computes probabilistic verdicts with uncertainty

346

5. Provides clear reasoning chains

347

348

✅ **Quality Gates Work:**

349

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

350

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

351

3. Clear rejection reasons provided

352

353

✅ **NFR11 Met:**

354

1. Quality gates reduce hallucination rate

355

2. Blocked outputs have clear explanations

356

3. Quality metrics are logged

357

358

359

=== 6.2 Quality Thresholds ===

360

361

**Minimum Acceptable:**

362

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

363

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

364

* Gate 1 blocks 100% of non-factual claims

365

* Gate 4 blocks verdicts with <2 sources

366

367

**Target:**

368

* ≥80% claims correctly classified

369

* ≥75% verdicts are reasonable

370

* <10% false positives (blocking good claims)

371

372

373

=== 6.3 POC Decision Gate ===

374

375

**After POC1, we decide:**

376

377

**✅ PROCEED to POC2** if:

378

* Success criteria met

379

* Quality gates demonstrably improve output

380

* Core workflow is technically sound

381

* Clear path to production quality

382

383

**⚠️ ITERATE POC1** if:

384

* Success criteria partially met

385

* Gates work but need tuning

386

* Core issues identified but fixable

387

388

**❌ PIVOT APPROACH** if:

389

* Success criteria not met

390

* Fundamental AI limitations discovered

391

* Quality gates insufficient

392

* Alternative approach needed

== 7. Test Cases ==

=== 7.1 Happy Path ===

398

399

**Test 1: Simple Factual Claim**

400

* Input: "Paris is the capital of France"

401

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

402

403

**Test 2: Ambiguous Claim**

404

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

405

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

406

407

**Test 3: Statistical Claim**

408

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

409

* Expected: Factual, evidence with numbers, probabilistic verdict

410

411

412

=== 7.2 Edge Cases ===

413

414

**Test 4: Opinion**

415

* Input: "Paris is the best city"

416

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

417

418

**Test 5: Prediction**

419

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

420

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

421

422

**Test 6: Insufficient Evidence**

423

* Input: Obscure factual claim with no sources

424

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

425

426

427

=== 7.3 Quality Gate Tests ===

428

429

**Test 7: Gate 1 Effectiveness**

430

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

431

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

432

433

**Test 8: Gate 4 Effectiveness**

434

* Input: Claims with varying evidence availability

435

* Expected: Gate 4 blocks low-confidence verdicts

436

437

438

== 8. Technical Architecture (POC) ==

439

440

=== 8.1 Simplified Architecture ===

441

442

**POC Tech Stack:**

443

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

444

* **Backend:** Single API endpoint

445

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

446

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

447

* **Deployment:** Single server

448

449

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

450

451

452

=== 8.2 AKEL Implementation ===

453

454

**POC AKEL:**

455

* Single-threaded processing

456

* Synchronous API calls

457

* No caching

458

* Basic error handling

459

* Console logging

460

461

**Full AKEL (POC2+):**

462

* Multi-threaded processing

463

* Async API calls

464

* Evidence caching

465

* Advanced error handling with retry

466

* Structured logging + monitoring

467

468

469

== 9. POC Philosophy ==

470

471

472

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

473

474

475

=== 9.1 Core Principles ===

476

477

**1. Prove Concept, Not Production**

478

* POC validates AI can do the job

479

* Production quality comes in POC2 and Beta 0

480