Wiki source code of POC2: Robust Quality & Reliability
Version 1.1 by Robert Schaub on 2025/12/22 14:22
Show last authors
| author | version | line-number | content |
|---|---|---|---|
| 1 | = POC2: Robust Quality & Reliability = **Phase Goal:** Prove AKEL produces high-quality outputs consistently at scale **Success Metric:** <5% hallucination rate, all 4 quality gates operational == 1. Overview == POC2 extends POC1 by implementing the full quality assurance framework (all 4 gates), adding evidence deduplication, and processing significantly more test articles to validate system reliability at scale. **Key Innovation:** Complete quality validation pipeline catches all categories of errors **What We're Proving:** | ||
| 2 | * All 4 quality gates work together effectively | ||
| 3 | * Evidence deduplication prevents artificial inflation | ||
| 4 | * System maintains quality at larger scale | ||
| 5 | * Quality metrics dashboard provides actionable insights == 2. New Requirements == === 2.1 NFR11: Complete Quality Assurance Framework === **Add Gates 2 & 3** (POC1 had only Gates 1 & 4) ==== Gate 2: Evidence Relevance Validation ==== **Purpose:** Ensure AI-linked evidence actually relates to the claim **Validation Checks:** | ||
| 6 | 1. **Semantic Similarity:** Cosine similarity between claim and evidence embeddings ≥ 0.6 | ||
| 7 | 2. **Entity Overlap:** At least 1 shared named entity between claim and evidence | ||
| 8 | 3. **Topic Relevance:** Evidence discusses the claim's subject matter (score ≥ 0.5) **Action if Failed:** | ||
| 9 | * Discard irrelevant evidence (don't count it) | ||
| 10 | * If <2 relevant evidence items remain → "Insufficient Evidence" verdict | ||
| 11 | * Log discarded evidence for quality review **Target:** 0% of evidence cited is off-topic ==== Gate 3: Scenario Coherence Check ==== **Purpose:** Validate scenarios are logical, complete, and meaningfully different **Validation Checks:** | ||
| 12 | 1. **Completeness:** All required fields populated (assumptions, scope, evidence context) | ||
| 13 | 2. **Internal Consistency:** Assumptions don't contradict each other (score <0.3) | ||
| 14 | 3. **Distinctiveness:** Scenarios are meaningfully different (similarity <0.8) | ||
| 15 | 4. **Minimum Detail:** At least 1 specific assumption per scenario **Action if Failed:** | ||
| 16 | * Merge duplicate scenarios | ||
| 17 | * Flag contradictory assumptions for review | ||
| 18 | * Reduce confidence score by 20% | ||
| 19 | * Do not publish if <2 distinct scenarios **Target:** 0% duplicate scenarios, all scenarios internally consistent === 2.2 FR54: Evidence Deduplication (NEW) === **Priority:** HIGH **Fulfills:** Accurate evidence counting, prevents artificial inflation **Purpose:** Prevent counting the same evidence multiple times when cited by different sources **Problem:** | ||
| 20 | * Wire services (AP, Reuters) redistribute same content | ||
| 21 | * Different sites cite the same original study | ||
| 22 | * Aggregators copy primary sources | ||
| 23 | * AKEL might count this as "5 sources" when it's really 1 **Solution: Content Fingerprinting** | ||
| 24 | * Generate SHA-256 hash of normalized text | ||
| 25 | * Detect near-duplicates (≥85% similarity) using fuzzy matching | ||
| 26 | * Track which sources cited each unique piece of evidence | ||
| 27 | * Display provenance chain to user **Target:** Duplicate detection >95% accurate, evidence counts reflect reality === 2.3 NFR13: Quality Metrics Dashboard (Internal) === **Priority:** HIGH **Fulfills:** Real-time quality monitoring during development **Dashboard Metrics:** | ||
| 28 | * Claim processing statistics | ||
| 29 | * Gate performance (pass/fail rates for each gate) | ||
| 30 | * Evidence quality metrics | ||
| 31 | * Hallucination rate tracking | ||
| 32 | * Processing performance **Target:** Dashboard functional, all metrics tracked, exportable == 3. Success Criteria == **✅ Quality:** | ||
| 33 | * Hallucination rate <5% (target: <3%) | ||
| 34 | * Average quality rating ≥8.0/10 | ||
| 35 | * 0 critical failures (publishable falsities) | ||
| 36 | * Gates correctly identify >95% of low-quality outputs **✅ All 4 Gates Operational:** | ||
| 37 | * Gate 1: Claim validation working | ||
| 38 | * Gate 2: Evidence relevance filtering working | ||
| 39 | * Gate 3: Scenario coherence checking working | ||
| 40 | * Gate 4: Verdict confidence assessment working **✅ Evidence Deduplication:** | ||
| 41 | * Duplicate detection >95% accurate | ||
| 42 | * Evidence counts reflect reality | ||
| 43 | * Provenance tracked correctly **✅ Metrics Dashboard:** | ||
| 44 | * All metrics implemented and tracking | ||
| 45 | * Dashboard functional and useful | ||
| 46 | * Alerts trigger appropriately == 4. Architecture Notes == **POC2 Enhanced Architecture:** {{code}} | ||
| 47 | Input → AKEL Processing → All 4 Quality Gates → Display (claims + scenarios (1: Claim validation + evidence linking 2: Evidence relevance + verdicts) 3: Scenario coherence 4: Verdict confidence) | ||
| 48 | {{/code}} **Key Additions from POC1:** | ||
| 49 | * Scenario generation component | ||
| 50 | * Evidence deduplication system | ||
| 51 | * Gates 2 & 3 implementation | ||
| 52 | * Quality metrics collection **Still Simplified vs. Full System:** | ||
| 53 | * Single AKEL orchestration (not multi-component pipeline) | ||
| 54 | * No review queue | ||
| 55 | * No federation architecture **See:** [[Architecture>>Test.FactHarbor.Specification.Architecture.WebHome]] for details == Related Pages == * [[POC1>>Test.FactHarbor.Roadmap.POC1.WebHome]] - Previous phase | ||
| 56 | * [[Beta 0>>Test.FactHarbor.Roadmap.Beta0.WebHome]] - Next phase | ||
| 57 | * [[Roadmap Overview>>Test.FactHarbor.Roadmap.WebHome]] | ||
| 58 | * [[Architecture>>Test.FactHarbor.Specification.Architecture.WebHome]] **Document Status:** ✅ POC2 Specification Complete - Waiting for POC1 Completion **Version:** V0.9.70 |