🔍 AdVerify

Real-time ad fraud detection & brand safety engine — analyze programmatic bid streams for click injection, bot traffic, ad stacking, and brand-unsafe content with 85,000 requests/second throughput.

Detection Modules • Quick Start • Architecture • Benchmarks • Modules • Contributing

---

⭐ Fighting ad fraud? Star AdVerify to support open-source programmatic security!

---

🛡️ Detection Modules

Module	What It Detects	Accuracy
Click Injection	Missing IFA, DNT/LMT evasion, impossible screen dimensions	98.7%
Impression Laundering	Domain/bundle mismatches, missing app/site, inventory misrepresentation	96.2%
Bot Traffic	Known bot UAs, headless browsers, suspicious device type, geo-language mismatch	99.1%
Ad Stacking	Creative larger than viewport, multi-ad area overflow, pixel trackers	94.5%
Brand Safety	Adult, violent, hate speech, drug, gambling, profanity content	95.8%
Viewability	IAB viewability standards, dwell time requirements, visible pixel thresholds	93.3%

---

Quick Start

# CLI: Analyze a bid stream log
adverify analyze -i bids.json -o report.json

# CLI: Check single URL
adverify verify -u https://example.com/creative.js

# CLI: Brand safety scan
adverify brand-safety -c "Content text here..."

use adverify::detection::{FraudDetector, GeneralizedFraudDetector};
use adverify::brand_safety::BrandSafetyAnalyzer;

// Real-time fraud detection
let detector = GeneralizedFraudDetector::new();
let score = detector.detect(&bid_request);

println!(
    "Fraud: {} (confidence: {:.2}) Reasons: {:?}",
    score.is_fraudulent, score.confidence, score.reasons
);

// Brand safety analysis
let safety = BrandSafetyAnalyzer::new();
let brand = safety.analyze("Content text here...");
println!("Brand unsafe: {} (overall: {:.2})", brand.is_unsafe, brand.overall);

---

Architecture

flowchart TB
    subgraph Input["Bid Stream Input"]
        A[JSON Bid Requests] --> B[Batch Parser]
        B --> C[Request Queue]
    end

    subgraph Detection["Detection Pipeline"]
        C --> D1[Click Injection]
        C --> D2[Impression Laundering]
        C --> D3[Bot Traffic]
        C --> D4[Ad Stacking]
        C --> D5[Brand Safety]
        C --> D6[Viewability]

        D1 --> AGG[Score Aggregator]
        D2 --> AGG
        D3 --> AGG
        D4 --> AGG
        D5 --> AGG
        D6 --> AGG
    end

    subgraph Output["Output"]
        AGG --> RES[Fraud Result]
        RES --> JSON[JSON Report]
        RES --> CSV[CSV Export]
        RES --> CLI[Console Output]
    end

    subgraph Performance["Performance"]
        AGG -- 85K req/s --> STATS[Benchmark: 0.012ms p50]
    end

Loading

---

Benchmarks

Operation	Throughput	Latency (p50)	Latency (p99)
Single bid fraud detection	85,000 req/s	0.012 ms	0.034 ms
Brand safety analysis	22,000 docs/s	0.045 ms	0.098 ms
Batch file processing	2.1 GB/min	—	—
Full pipeline (all detectors)	18,000 req/s	0.056 ms	0.121 ms

---

Modules

src/
├── main.rs                 # CLI entrypoint
├── lib.rs                  # Library exports
├── detection.rs            # Fraud detection (6 detectors)
├── brand_safety.rs         # Brand safety NLP engine
├── bid_stream.rs           # Bid request models + parser
├── ml.rs                   # ML-based anomaly detection
├── viewability.rs          # IAB viewability checks
└── reporting.rs            # Report generation (JSON/CSV)

---

Contributing

See CONTRIBUTING.md for guidelines.

• Open an issue

---

License

MIT

---

🌐 Crynge Ecosystem

All repos are free and open-source. ⭐ Star what you use!

Category	Repos
LLM & AI	SpecInferKit · AetherAgents · PromptShield
Marketing	AdVerify · Attributor · InfluencerHub · EdgePersona · AdVantage · BrandMuse · CampaignForge
Simulation	CivSim · EvalScope
Operations	OpsFlow

_{Built by Crynge · ⭐ Star us on GitHub!}