POC Requirements (POC1 & POC2)

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)

== 2. POC1 Scope ==

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

44

45

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.

46

47

48

**POC1 implements these formal requirements:**

49

50

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

51

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

52

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

53

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

54

55

**POC1 also implements these workflow components** (detailed as FR1-FR6, FR8, FR11, FR13 in implementation sections below):

56

* Claim extraction (FR1)

57

* Claim context (FR2)

58

* Multiple scenarios (FR3)

59

* Evidence collection (FR5)

60

* Source quality assessment (FR6)

61

* Time evolution tracking (FR8) - deferred to POC2

62

* Audit trail (FR11) - deferred to Beta 0

63

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

64

65

**Partial implementations:**

66

* NFR1 (Explainability) - Basic only

67

* NFR2 (Performance) - Functional but not optimized

68

* NFR3 (Transparency) - Basic only

69

70

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

== 3. POC Simplifications ==

75

76

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

77

78

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

79

80

**POC Implementation:**

81

* ✅ AKEL extracts claims using LLM

82

* ✅ Each claim includes original text reference

83

* ✅ Claims are identified as factual/non-factual

84

* ❌ No advanced claim parsing (added in POC2)

85

86

**Acceptance Criteria:**

87

* Extracts 3-5 claims from typical article

88

* Identifies factual vs non-factual claims

89

* Quality Gate 1 validates extraction

90

91

92

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

93

94

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

95

96

**POC Implementation:**

97

* ✅ AKEL generates 2-3 scenarios per claim

98

* ✅ Scenarios capture different interpretations

99

* ✅ Each scenario is evaluated separately

100

* ✅ Verdict considers all scenarios

101

102

**Acceptance Criteria:**

103

* Generates 2+ scenarios for ambiguous claims

104

* Scenarios are meaningfully different

105

* All scenarios are evaluated

106

107

108

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

109

110

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

111

112

**POC Implementation:**

113

* ✅ Simple text summary generated

114

* ❌ No rich formatting (added in Beta 0)

115

* ❌ No visual elements (added in Beta 0)

116

* ❌ No interactive features (added in Beta 0)

**POC Format:**

```

Claim: [extracted claim]

121

Scenarios: [list of scenarios]

122

Evidence: [supporting/opposing evidence]

123

Verdict: [probability with uncertainty]

```

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

128

129

**Main Requirements:**

130

* FR5: Collect supporting and opposing evidence

131

* FR6: Evaluate evidence source reliability

132

133

**POC Implementation:**

134

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

135

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

136

* ✅ Source reliability scoring

137

* ❌ No evidence deduplication (added in POC2)

138

* ❌ No advanced source verification (added in POC2)

139

140

**Acceptance Criteria:**

141

* Finds 2+ supporting evidence items

142

* Finds 1+ opposing evidence (if exists)

143

* Sources scored for reliability

144

145

146

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

147

148

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

149

150

**POC Implementation:**

151

* ✅ Probabilistic verdicts (0-100% confidence)

152

* ✅ Uncertainty explicitly stated

153

* ✅ Reasoning chain provided

154

* ✅ Quality Gate 4 validates verdict confidence

**POC Output:**

```

Verdict: 70% likely true

159

Uncertainty: ±15% (moderate confidence)

160

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

161

Confidence Level: MEDIUM

162

```

163

164

**Acceptance Criteria:**

165

* Verdicts include probability (0-100%)

166

* Uncertainty explicitly quantified

167

* Reasoning chain explains verdict

168

169

170

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

171

172

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

173

174

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

175

176

**Quality Gate 1: Claim Validation**

177

* ✅ Validates claim is factual and verifiable

178

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

179

* ✅ Provides clear rejection reason

180

181

**Quality Gate 4: Verdict Confidence Assessment**

182

* ✅ Validates ≥2 sources found

183

* ✅ Validates quality score ≥0.6

184

* ✅ Blocks low-confidence verdicts

185

* ✅ Provides clear rejection reason

186

187

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

188

* ❌ Gate 2: Evidence Relevance

189

* ❌ Gate 3: Scenario Coherence

190

* ❌ Gate 5: Source Diversity

191

* ❌ Gate 6: Reasoning Validity

192

* ❌ Gate 7: Output Completeness

193

194

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

195

196

197

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

198

199

**Main Requirements:**

200

* NFR1: Response time < 30 seconds

201

* NFR2: Handle 1000+ concurrent users

202

* NFR3: 99.9% uptime

203

204

**POC Implementation:**

205

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

206

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

207

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

208

209

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

210

211

**POC Acceptance:**

212

* Analysis completes (no timeout requirement)

213

* Errors don't crash system

214

* Basic logging in place

215

216

217

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

218

219

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

220

221

222

**Deferred to Beta 0:**

**Out of Scope:**

* ❌ User accounts and authentication (FR8)

227

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

228

* ❌ Public publishing interface (FR10)

229

* ❌ Social sharing (FR11)

230

* ❌ Email notifications (FR12)

231

* ❌ API access (FR13)

232

233

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

234

235

236

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

237

238

**Out of Scope:**

239

* ❌ IFCN compliance (FR47)

240

* ❌ ClaimReview schema (FR48)

241

* ❌ Archive.org integration (FR49)

242

* ❌ OSINT toolkit (FR50)

243

* ❌ Video verification (FR51)

244

* ❌ Deepfake detection (FR52)

245

* ❌ Cross-org sharing (FR53)

246

247

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

248

249

250

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

251

252

**Out of Scope:**

253

* ❌ Security controls (NFR4, NFR12)

254

* ❌ Code maintainability (NFR5)

255

* ❌ System monitoring (NFR13)

256

* ❌ Evidence deduplication

257

* ❌ Advanced source verification

258

* ❌ Full 7-gate quality framework

259

260

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

261

262

263

== 5. POC Output Specification ==

264

265

=== 5.1 Required Output Elements ===

266

267

For each analyzed claim, POC must produce:

**1. Claim**

* Original text

* Classification (factual/non-factual/ambiguous)

272

* If non-factual: Clear reason why

273

274

**2. Scenarios** (if factual)

275

* 2-3 interpretation scenarios

276

* Each scenario clearly described

277

278

**3. Evidence** (if factual)

279

* Supporting evidence (2+ items)

280

* Opposing evidence (if exists)

281

* Source URLs and reliability scores

282

283

**4. Verdict** (if factual)

284

* Probability (0-100%)

285

* Uncertainty quantification

286

* Confidence level (LOW/MEDIUM/HIGH)

287

* Reasoning chain

288

289

**5. Quality Status**

290

* Which gates passed/failed

291

* If failed: Clear explanation why

292

293

294

=== 5.2 Example POC Output ===

{

"claim": {

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

300

"type": "factual",

301

"gate1_status": "PASS"

302

},

303

"scenarios": [

304

"Switzerland's overall life expectancy is highest",

305

"Switzerland ranks highest for specific age groups"

],

"evidence": {

"supporting": [

{

"source": "WHO Report 2023",

311

"reliability": 0.95,

312

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

}

],

"opposing": [

{

"source": "Eurostat 2024",

318

"reliability": 0.90,

319

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

}

]

},

"verdict": {

"probability": 0.65,

"uncertainty": 0.15,

"confidence": "MEDIUM",

327

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

328

"gate4_status": "PASS"

}

}

== 6. Success Criteria ==

335

336

337

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

338

339

340

=== 6.1 Functional Success ===

341

342

POC is successful if:

343

344

✅ **FR1-FR7 Requirements Met:**

345

1. Extracts 3-5 factual claims from test articles

346

2. Generates 2-3 scenarios per ambiguous claim

347

3. Finds supporting AND opposing evidence

348

4. Computes probabilistic verdicts with uncertainty

349

5. Provides clear reasoning chains

350

351

✅ **Quality Gates Work:**

352

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

353

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

354

3. Clear rejection reasons provided

355

356

✅ **NFR11 Met:**

357

1. Quality gates reduce hallucination rate

358

2. Blocked outputs have clear explanations

359

3. Quality metrics are logged

360

361

362

=== 6.2 Quality Thresholds ===

363

364

**Minimum Acceptable:**

365

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

366

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

367

* Gate 1 blocks 100% of non-factual claims

368

* Gate 4 blocks verdicts with <2 sources

369

370

**Target:**

371

* ≥80% claims correctly classified

372

* ≥75% verdicts are reasonable

373

* <10% false positives (blocking good claims)

374

375

376

=== 6.3 POC Decision Gate ===

377

378

**After POC1, we decide:**

379

380

**✅ PROCEED to POC2** if:

381

* Success criteria met

382

* Quality gates demonstrably improve output

383

* Core workflow is technically sound

384

* Clear path to production quality

385

386

**⚠️ ITERATE POC1** if:

387

* Success criteria partially met

388

* Gates work but need tuning

389

* Core issues identified but fixable

390

391

**❌ PIVOT APPROACH** if:

392

* Success criteria not met

393

* Fundamental AI limitations discovered

394

* Quality gates insufficient

395

* Alternative approach needed

== 7. Test Cases ==

=== 7.1 Happy Path ===

401

402

**Test 1: Simple Factual Claim**

403

* Input: "Paris is the capital of France"

404

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

405

406

**Test 2: Ambiguous Claim**

407

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

408

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

409

410

**Test 3: Statistical Claim**

411

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

412

* Expected: Factual, evidence with numbers, probabilistic verdict

413

414

415

=== 7.2 Edge Cases ===

416

417

**Test 4: Opinion**

418

* Input: "Paris is the best city"

419

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

420

421

**Test 5: Prediction**

422

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

423

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

424

425

**Test 6: Insufficient Evidence**

426

* Input: Obscure factual claim with no sources

427

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

428

429

430

=== 7.3 Quality Gate Tests ===

431

432

**Test 7: Gate 1 Effectiveness**

433

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

434

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

435

436

**Test 8: Gate 4 Effectiveness**

437

* Input: Claims with varying evidence availability

438

* Expected: Gate 4 blocks low-confidence verdicts

439

440

441

== 8. Technical Architecture (POC) ==

442

443

=== 8.1 Simplified Architecture ===

444

445

**POC Tech Stack:**

446

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

447

* **Backend:** Single API endpoint

448

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

449

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

450

* **Deployment:** Single server

451

452

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

453

454

455

=== 8.2 AKEL Implementation ===

456

457

**POC AKEL:**

458

* Single-threaded processing

459

* Synchronous API calls

460

* No caching

461

* Basic error handling

462

* Console logging

463

464

**Full AKEL (POC2+):**

465

* Multi-threaded processing

466

* Async API calls

467

* Evidence caching

468

* Advanced error handling with retry

469

* Structured logging + monitoring

470

471

472

== 9. POC Philosophy ==

473

474

475

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

476

477

478

=== 9.1 Core Principles ===

479

480

**1. Prove Concept, Not Production**

481

* POC validates AI can do the job

482

* Production quality comes in POC2 and Beta 0

483