Skip to Content
Last modified by Robert Schaub on 2026/02/08 08:31
Hide last authors
Robert Schaub 1.1 1 {{info}}
2 **Current Implementation (v2.6.33)** - Triple-Path Pipeline Architecture. Three pipeline variants share common modules for scope detection, aggregation, and claim processing.
3 {{/info}}
4
5 = Triple-Path Pipeline Architecture =
6
7 {{mermaid}}
8
9 graph TB
10 subgraph Input[User Input]
11 URL[URL Input]
12 TEXT[Text Input]
13 end
14
15 subgraph Shared[Shared Modules]
16 SCOPES[scopes.ts Scope Detection]
17 AGG[aggregation.ts Verdict Aggregation]
18 CLAIM_D[claim-decomposition.ts]
19 end
20
21 subgraph Dispatch[Pipeline Dispatch]
22 SELECT{Select Pipeline}
23 end
24
25 subgraph Pipelines[Pipeline Implementations]
26 ORCH[Orchestrated Pipeline]
27 CANON[Monolithic Canonical]
28 DYN[Monolithic Dynamic]
29 end
30
31 subgraph LLM[LLM Layer]
32 PROVIDER[AI SDK Provider]
33 end
34
35 subgraph Output[Result]
36 RESULT[AnalysisResult JSON]
37 REPORT[Markdown Report]
38 end
39
40 URL --> SELECT
41 TEXT --> SELECT
42 SELECT -->|orchestrated| ORCH
43 SELECT -->|monolithic_canonical| CANON
44 SELECT -->|monolithic_dynamic| DYN
45 SCOPES --> ORCH
46 SCOPES --> CANON
47 AGG --> ORCH
48 AGG --> CANON
49 CLAIM_D --> ORCH
50 CLAIM_D --> CANON
51 ORCH --> PROVIDER
52 CANON --> PROVIDER
53 DYN --> PROVIDER
54 ORCH --> RESULT
55 CANON --> RESULT
56 DYN --> RESULT
57 RESULT --> REPORT
58
59 {{/mermaid}}
60
61 == Pipeline Variants ==
62
63 |= Variant |= File |= Approach |= Output Schema
64 | **Orchestrated** | ##orchestrated.ts## | Multi-step workflow with explicit stages | Canonical (structured)
65 | **Monolithic Canonical** | ##monolithic-canonical.ts## | Single LLM tool-loop call | Canonical (structured)
66 | **Monolithic Dynamic** | ##monolithic-dynamic.ts## | Single LLM tool-loop call | Dynamic (flexible)
67
68 == Shared Modules ==
69
70 |= Module |= Purpose
71 | ##scopes.ts## | Heuristic scope pre-detection before LLM
72 | ##aggregation.ts## | Verdict weighting, contestation validation
73 | ##claim-decomposition.ts## | Claim text parsing and normalization
74
75 == Orchestrated Pipeline Steps ==
76
77 1. **Understand** - Detect input type, extract claims, identify dependencies
78 2. **Research** (iterative) - Generate queries, fetch sources, extract facts
79 3. **Verdict Generation** - Generate claim and article verdicts
80 4. **Summary** - Build two-panel summary
81 5. **Report** - Generate markdown report
Robert Schaub 2.1 82
83 == Detailed Pipeline Diagrams ==
84
85 For internal implementation details of each pipeline variant:
Robert Schaub 2.3 86
Robert Schaub 2.5 87 * [[Orchestrated Pipeline Internal>>Archive.FactHarbor 2026\.02\.08.Specification.Diagrams.Orchestrated Pipeline Internal.WebHome]] - 5-step staged workflow (9000 lines)
Robert Schaub 2.3 88 * [[Monolithic Canonical Internal>>Archive.FactHarbor 2026\.02\.08.Specification.Diagrams.Monolithic Canonical Pipeline Internal.WebHome]] - Single-context canonical output (1100 lines)
Robert Schaub 2.4 89 * [[Monolithic Dynamic Internal>>Archive.FactHarbor 2026\.02\.08.Specification.Diagrams.Monolithic Dynamic Pipeline Internal.WebHome]] - Flexible experimental output (550 lines)