Detection Engine v17

13 Layers Between Your Budget and Bots

PureGuard is server-side bot detection built for pop, push, and native media buyers. 13 detection layers. Under 15ms decision speed. No JS tags. No pixel fires. No false positives.

13
Detection Layers
<15ms
Decision Speed
0%
False Positives

Detection Philosophy

“Can I PROVE it’s a bot?” — not “is it suspicious?”

Evidence-Based, Not Paranoia-Based

A single weak signal never blocks traffic. PureGuard requires multiple converging pieces of evidence before making a block decision. One missing header is a data point, not a death sentence.

Trust Scoring, Not Binary

Every click gets a 0–10 trust score built from positive and negative signals. Borderline traffic is analyzed carefully, not killed on sight. The threshold is tunable per source.

Zone Intelligence, Not Just Bot Killing

The goal is finding gold zones vs trash zones. Bot detection is the means — zone intelligence is the end. No competitor offers zone-level quality scoring like this.

Server-Side Only

No JavaScript tags to load. No client-side latency. No pixel fires. PureGuard reads HTTP headers and fingerprints server-side, before the redirect even happens.

Built for Cheap Traffic

Competitors charge $3,000–$50,000/month. PureGuard is built for media buyers running $0.50–$1.00 CPM pop traffic. Enterprise-grade detection at a fraction of the cost.

Guard-First Architecture

Zone verdicts come from real Guard data: accept_rate, average trust score, hit volume. No statistical pattern matching without hard evidence to back it up.

13 Detection Layers

Every click passes through all 13 layers in sequence. Hard kills exit immediately. Trust scoring layers accumulate evidence.

LAYER 01
Instant Kill

FireHOL Threat Intelligence

Cross-references every IP against FireHOL Level 1 — a curated list of 4,500+ known malicious IPs from botnets, scanners, and attack infrastructure. Updated every 6 hours via kernel ipset for zero-latency lookups.

LAYER 02
Instant Kill

CrowdSec Community Blocklist

22,000+ IPs flagged by the CrowdSec community intelligence network. Real-time threat sharing from thousands of servers worldwide. Kernel ipset integration for sub-millisecond matching.

LAYER 03
Instant Kill

Bot User-Agent Detection

Matches against 30+ known bot signatures: Puppeteer, Selenium, PhantomJS, curl, WebDriver, MauiBot, DataForSeoBot, and headless browser fingerprints. Pattern-based with zero false positives.

LAYER 04
Instant Kill

Fake Chrome Version

Chrome version numbers above 160 do not exist. Any UA claiming Chrome/161+ is a lazy bot that didn’t check the current release. Instant kill, zero ambiguity.

LAYER 05
Instant Kill

Chrome UA Reduction Analysis

W3C-spec aware. Chrome 110+ legitimately sends .0.0.0 build numbers (privacy feature). But Chrome <110 with .0.0.0 is a spoofed user-agent — real old Chrome always sent full build numbers.

LAYER 06
Instant Kill

Burst Rate Limiter

Detects rapid-fire clicks from the same IP within a configurable time window. Uses APCu in-memory cache for microsecond lookups. No database overhead, no disk I/O.

LAYER 07
Instant Kill

Ad Fraud Network Signatures

Identifies crawlers from ad verification companies: Pixalate, DoubleVerify, Snobi, Criteo, and other ad fraud detection bots that inflate your click counts without converting.

LAYER 08
Trust Scoring

Sec-Fetch Header Validation

Sec-Fetch-Site, Sec-Fetch-Mode, and Sec-Fetch-Dest are browser-enforced headers that cannot be forged by simple HTTP clients. Missing: −2.0. Valid navigation: +1.0. Invalid combination: −0.5.

LAYER 09
Trust Scoring

Datacenter ASN Detection

Checks IP against 50+ known cloud/datacenter ASNs (AWS, GCP, Azure, DigitalOcean, OVH, Hetzner, Vultr, Linode). Uses MaxMind GeoIP2 ASN database. Datacenter: −1.5. Residential: +0.3.

LAYER 10
Trust Scoring

Header Consistency Analysis

Chrome 89+ must send Sec-CH-UA Client Hints. Missing Accept-Encoding is abnormal. Connection: close is a bot fingerprint. Each inconsistency adds negative trust weight.

LAYER 11
Trust Scoring

Referer Chain Validation

Source-aware referer pattern checking. Each traffic source has expected referer behavior. Empty referer from RTB sources gets −1.5 penalty. Context-dependent, not blanket rules.

LAYER 12
Trust Scoring

HTTP Version Fingerprinting

Chrome 80+ uses HTTP/2 or HTTP/3 exclusively. A request claiming Chrome 131 but arriving via HTTP/1.0 is a clear bot fingerprint. Penalty: −3.0 (strong signal).

LAYER 13
Trust Scoring

Hosting ASN + Residential Proxy

MaxMind ASN lookup identifies hosting providers. Additionally detects 12 known residential proxy ASNs (Luminati, Oxylabs, Smartproxy, etc.) that bots use to appear residential. Hosting: −2.0. Resi proxy: −1.5.

Observation Layers (data collection, no scoring impact): JA3/JA4 TLS fingerprinting, Accept-Language vs geo consistency, HTTP header order CRC32 fingerprinting. These layers collect intelligence that feeds into zone reputation over time.

Trust Scoring Engine

Every click starts at base 5.0 and accumulates positive and negative signals. The final score determines the verdict.

BASE TRUST SCORE 5.0 / 10.0 ─── Positive Signals ─────────────────────────── + 1.0 Sec-Fetch headers valid (navigate/document) + 0.5 Real Chrome build number detected + 0.3 Residential ISP (not datacenter) + 0.3 Mobile carrier detected + 0.3 Accept-Language matches geo ─── Negative Signals ─────────────────────────── - 3.0 HTTP/1.0 with Chrome 80+ UA - 2.0 Missing Sec-Fetch headers entirely - 2.0 Chrome 89+ without Sec-CH-UA Client Hints - 1.5 Datacenter/cloud ASN - 1.5 Known residential proxy ASN - 1.5 Empty referer (source-dependent) - 1.0 Missing Accept-Encoding header - 1.0 Connection: close (bot fingerprint) ─── Decision ─────────────────────────────────── TRUST >= 5.5 → ACCEPT → Redirect to money URL TRUST < 5.5 → BLOCK → Redirect to dump URL

Real-World Example

Same User-Agent, different verdict:
Chrome 131 on a mobile ISP in Thailand with valid Sec-Fetch headers → trust ~6.8 ACCEPT
Same Chrome 131 UA from an AWS IP, no Sec-Fetch headers, HTTP/1.0 → trust ~0.5 BLOCK

The UA string is identical. The surrounding evidence is completely different. That’s why trust scoring works and binary UA matching doesn’t.

Zone Intelligence Engine

Bot detection is the means. Zone intelligence is the end.

Finding Gold in Trash

Every traffic source has hundreds or thousands of zones (publisher IDs). Some zones deliver 95% real humans. Others deliver 95% bots. The difference between profit and loss is knowing which is which.

PureGuard aggregates Guard verdicts per zone — accept rate, average trust score, hit volume, bot evidence patterns, device fingerprint diversity, conversion data — and produces a quality verdict for every zone, updated every 10 minutes.

CONFIRMED_BOT
Multiple hard kill signals. Guard score 40+. Zero human evidence. Auto-blocked.
DEAD_ZONE
Zero accepts out of 3+ hits. Every click failed Guard checks. Wasting budget.
SUSPECT
Mixed signals. Some bot evidence but not conclusive. Monitored, not blocked.
CLEAN
High accept rate, strong trust scores, real device diversity. Proven human traffic.

Result: A ready-to-paste blocklist and whitelist for every traffic source you run. Export as CSV, flat text, or use the API. Updated every 10 minutes. No manual analysis needed.

Traffic Flow

Where PureGuard sits in your traffic chain.

Traffic Source
PopAds, RollerAds, HilltopAds...
PureGuard
13 layers, <15ms
ACCEPT
Redirect to offer/landing
BLOCK
Redirect to dump URL

Shadow Mode

Log all Guard decisions without blocking anything. See exactly what PureGuard would block before you flip the switch. Perfect for validating detection accuracy on your specific traffic sources.

  • Full event logging with trust scores
  • Zone intelligence builds in background
  • Zero impact on live traffic flow

Protect Mode

Active blocking. Bots get redirected to dump URLs. Clean traffic reaches your money URLs. Zone blocklists auto-generate and update every 10 minutes.

  • Real-time bot blocking (<15ms)
  • Auto-generated zone blocklists
  • Configurable trust threshold per source

RTB Pre-Bid Filtering

Score traffic before you pay for it. Not after.

Without PureGuard

The traditional RTB flow burns money on bots:

  • Bid request arrives → you bid
  • You win the auction → you pay
  • Click arrives → it’s a bot
  • Money gone. No conversion. No recourse.

With PureGuard RTB

Guard scores the bid request before you commit:

  • Bid request arrives → scored in 3ms
  • 36 signals analyzed (UA, geo, ASN, domain, device)
  • Bot detected → NOBID (you pay nothing)
  • Clean traffic → BID (money well spent)
36 RTB Scoring Signals:
Bot UA patterns · CrowdSec blocklist · FireHOL CIDR · Zone blocklist · Internal domains · Empty UA · Short UA · Device type · Bid floor analysis · Geo presence · IPv6 detection · Datacenter IP prefix · Chrome version · Domain quality · Burst rate · OS/device mismatch · Geo mismatch · Chrome/Windows mismatch · Linux Chrome · IP-as-domain · Made-for-ads domain · TLD validation · Zero bidfloor · Site info presence · IP presence · MaxMind geo verification · MaxMind hosting ASN · Tier-1 geo bonus · Browser detection · Mobile bonus · Desktop bonus · Page URL bonus · Ad fraud UA patterns · Domain entropy · Device type scoring · Win-rate correlation

Competitor Comparison

Built different. Built for media buyers, not enterprise marketing teams.

Capability PureGuard ClickCease
$150/mo
Lunio
$500/mo
HUMAN
$50K/mo
Pop/push traffic support
Zone-level intelligence
RTB pre-bid filtering
Server-side (no JS tags)
Auto zone blocklist
Auto zone whitelist
Detection layers 13 3–5 5–8 20+
Decision latency <15ms 100ms+ 50ms+ <10ms
Threat intelligence feeds
ML behavioral analysis
Free tier 100K checks

Why competitors don’t work for media buyers: ClickCease and Lunio are built for Google/Facebook advertisers — they use JS tags and client-side detection that doesn’t work with pop/push traffic. HUMAN (formerly White Ops) is enterprise-only at $50K+/month and doesn’t offer zone-level intelligence. None of them understand the media buying workflow: zones, blocklists, source-specific tuning, or cheap traffic economics.

Real Detection Data

From production traffic across pop, push, and RTB sources.

79.4%
Clean Traffic Passed
20.6%
Bots Caught & Blocked
0
False Positives
0.9%
Accept Rate on Blocked Zones
Missing Sec-Fetch Headers 34%
Datacenter / Cloud IPs 22%
Header Inconsistency 18%
Known Threat IPs 8%
Bot User Agents 7%
Other Signals 11%

Detection Roadmap

What’s coming next to the detection engine.

See It Work on Your Traffic

Connect your traffic source, enable Shadow Mode, and watch PureGuard score every click in real time. No code changes. No JS tags. Takes 2 minutes.

Start Free Read the Docs
100,000 free checks · Full dashboard · Zone intelligence included