datasynth-eval 3.1.1

Evaluation framework for synthetic financial data quality and coherence
Documentation 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62

# datasynth-eval

Evaluation framework for synthetic financial data quality and coherence.

## Overview

`datasynth-eval` provides automated quality assessment for generated data:

- **Statistical Evaluation**: Benford's Law compliance, distribution analysis
- **Coherence Checking**: Balance verification, document chain integrity
- **Intercompany Validation**: IC matching and elimination verification
- **Uniqueness Analysis**: Duplicate detection across datasets

## Evaluation Categories

| Category | Description |
|----------|-------------|
| Statistical | Benford's Law, amount distributions, temporal patterns |
| Coherence | Trial balance, subledger reconciliation, FX consistency |
| Intercompany | IC matching rates, elimination completeness |
| Uniqueness | Document ID collisions, duplicate transaction detection |
| Banking/AML | 10 dedicated analyzers covering KYC, typologies, velocity, networks, lifecycle |

### Banking Analyzers (`banking/`)

| Analyzer | What it validates |
|----------|-------------------|
| `KycCompletenessAnalyzer` | Core KYC field coverage (name, DOB, ID, risk rating, beneficial owners) |
| `AmlDetectabilityAnalyzer` | Typology coverage + scenario case_id coherence |
| `CrossLayerCoherenceAnalyzer` | Payment↔BankTransaction referential integrity, fraud propagation rate |
| `VelocityQualityAnalyzer` | Rolling-window ordering invariants (1h≤24h≤7d≤30d), z-score calibration |
| `FalsePositiveAnalyzer` | FP rate bounds, label mutual exclusivity, reason coverage |
| `DeviceFingerprintAnalyzer` | Power-law device distribution, single-device dominance, trust calibration |
| `SanctionsScreeningAnalyzer` | Low-risk Clear rate, high-risk match rate, PEP name variations |
| `SophisticationAnalyzer` | Sophistication level diversity, context-appropriate skew |
| `LifecycleAnalyzer` | Phase diversity, progression rate, event-driven transition rate |
| `NetworkStructureAnalyzer` | Power-law topology (hub ratio ≥2.5× avg), role diversity |

## Usage

```rust
use datasynth_eval::{Evaluator, EvaluationConfig};

let evaluator = Evaluator::new(EvaluationConfig::default());
let report = evaluator.evaluate(&generated_data)?;

println!("Benford compliance: {:.2}%", report.benford_score * 100.0);
println!("Balance coherence: {}", report.balance_check.passed);
```

## Evaluation Report

The evaluation produces a comprehensive report including:

- **Pass/Fail Status**: Overall and per-category
- **Scores**: Numerical scores for statistical measures
- **Warnings**: Potential issues that don't fail validation
- **Details**: Specific findings and recommendations

## License

Apache-2.0 - See [LICENSE](../../LICENSE) for details.