Seatext library / BotRefund evidence
BotRefund's Bot Detection Checks: The 106-Signal Architecture Explained
BotRefund runs 106 independent checks grouped into behavioral, browser, and network categories. Each check contributes one piece of evidence that an AI model weighs together, producing a 99% accuracy claim through corroboration rather than...
✓ Built for advertisers who need clear, refund-ready traffic evidence.
BotRefund's detection system relies on 106 independent checks that examine browser APIs, user behavior, network traits, and device signals. No single check decides the verdict; instead, each check adds an objective fact that the prediction AI weighs against the full pattern across browser, network, device, and behavior evidence.
The 106-check architecture
BotRefund organizes its detection into 106 independent signals. The company groups these signals into broad categories that cover how a visitor interacts with a page, how the browser behaves, and what the network connection reveals. Each signal is designed to be an independent piece of evidence — something that can be measured objectively without relying on other checks.
According to BotRefund's documentation, the system treats every anomaly as evidence, not a verdict. Privacy tools, corporate networks, travel, and unusual devices can all produce unexpected behavior for genuine people. The platform keeps each signal as a data point and cross-checks it against other independent signals before the AI model makes a final classification.
Behavioral interaction categories
The largest group of checks focuses on how a visitor moves, clicks, scrolls, and spends time on a page. BotRefund's homepage and detection pages list eight behavioral categories, each containing multiple specific checks:
- Click behavior — Ghost click detection catches click activity that happens without the natural sequence of human intent.
- Trap behavior — Honeypot trap interactions watch for bots that respond to hidden or intentionally deceptive page elements.
- Pointer behavior — Robotic linear mouse movements flag unnaturally straight pointer paths that rarely appear in real user sessions.
- Motion behavior — Absence of humanlike mouse tremor looks for the tiny imperfections and jitter typical of human movement.
- Speed behavior — Superhuman input speed (<1ms) identifies interactions that happen faster than a person could realistically perform.
- Path behavior — Grid-aligned movement patterns detect movement that snaps to precise lines or blocks instead of natural curves.
- Engagement behavior — Absence of clicks or scrolling highlights sessions that stay too static to match a real browsing journey.
- Session behavior — Unnatural session durations catch visit lengths that are too short, too long, or too uniform to be human.
These categories appear on both the main detection overview and the local about-us page, confirming they form the core behavioral framework.
Browser and API integrity checks
Beyond behavior, BotRefund runs checks that probe the browser itself for signs of automation tooling. Two documented examples illustrate this layer:
- Console Debug Evaluator — Looks for a mismatch that a real browsing session does not normally create. Automation tools often patch or hide browser APIs, but those changes can break when the browser is checked from another angle.
- window.open Tamper — Checks whether scripts can reproduce the varied timing, movement, and hesitation of real people when opening new windows or tabs.
Both checks are described as "one of 106 independent checks" and follow the same evidence-not-verdict philosophy. The Console Debug Evaluator page also references a heading "Evasion, Debugger, & Anti-Stealth Traps," suggesting a broader family of anti-stealth checks that target common automation frameworks.
Timing and navigation anomaly checks
A third family of checks focuses on timing patterns that are difficult for scripts to fake convincingly. The "Impossible Tab Speed" check is a documented example: it looks for tab-switching or navigation speeds that exceed human reaction times. Like the browser integrity checks, it is framed as one of the 106 independent signals that feeds the AI model.
These timing checks complement the behavioral categories by catching automation that may mimic mouse movement well but fails on micro-timing consistency across browser events.
Cross-checking and AI prediction
BotRefund emphasizes a three-step process for every signal:
- Independent evidence — The signal adds one objective fact about the visit.
- Cross-checked context — The system tests whether other signals support the same story.
- AI prediction — The model weighs the complete pattern instead of trusting a raw rule.
The company claims 99% accuracy comes from this corroboration approach. The AI evaluates the complete picture across browser, network, device, and behavior evidence, identifying a visit as bot or human based on how all signals fit together rather than any single tell.
How signals become a verdict
In practice, a visit might trigger several behavioral signals (e.g., linear mouse movement, superhuman click speed, no scrolling) plus a browser integrity signal (e.g., Console Debug Evaluator mismatch) and a timing signal (e.g., Impossible Tab Speed). Each signal alone could have a benign explanation — a privacy extension, a motor impairment, a fast reader. The AI model weighs the combination: when multiple independent categories point the same way, confidence rises. When signals conflict, the model can downgrade the bot probability rather than force a binary decision.
This design also explains why BotRefund can produce audit-ready evidence for ad-platform refund disputes. Each flagged visit comes with a trail of specific, documented signals that can be shown to Google or Meta representatives.
Limitations and false-positive considerations
BotRefund explicitly acknowledges that privacy tools, travel, corporate networks, and unusual devices can produce unexpected behavior for genuine people. The system keeps each signal as evidence rather than a verdict precisely to avoid blocking real users who happen to trigger one anomaly. However, the source pack does not disclose:
- The exact false-positive rate at the 99% accuracy claim
- How the system handles users with accessibility tools that alter mouse or keyboard behavior
- Whether certain geographic regions or device types see higher false-positive rates
- The minimum number of signals required before the AI issues a high-confidence bot classification
Prospective customers should ask for these details during a demo or audit.
Key facts
| Aspect | Detail | Source |
|---|---|---|
| Total independent checks | 106 | S1, S4, S5 |
| Behavioral categories | 8 (Click, Trap, Pointer, Motion, Speed, Path, Engagement, Session) | S2, S6 |
| Documented browser integrity checks | Console Debug Evaluator, window.open Tamper | S1, S4 |
| Documented timing checks | Impossible Tab Speed | S5 |
| Anti-stealth category referenced | Evasion, Debugger, & Anti-Stealth Traps | S1 |
| Biometric & behavioral interactions category | Includes window.open Tamper, Impossible Tab Speed | S4, S5 |
| Claimed accuracy | 99% via AI corroboration across browser, network, device, behavior | S1, S4, S5 |
| Evidence philosophy | Each signal is evidence, not a verdict; cross-checked before AI weighs pattern | S1, S4, S5 |
| Setup time claimed | About one minute to add to website | S2, S6 |
| Refund lookback window | Google Ads spend dating back to 2017 | S2, S6 |
Frequently asked questions
How many checks does BotRefund actually run per visit?
All 106 checks run independently on each visit. The system collects every signal and feeds the complete set into the AI model for the final classification.
Can a single check trigger a bot block?
No. BotRefund's documentation states repeatedly that a single anomaly is not a bot verdict. The AI weighs the complete pattern across all categories before deciding.
What happens when a privacy extension triggers a browser integrity check?
The signal is recorded as evidence. If other behavioral, network, and device signals look human, the AI model can still classify the visit as human. The cross-checking step is designed to prevent false positives from privacy tools alone.
Are the 106 checks static or do they update?
The source pack does not specify update frequency. Given that ad fraud tactics evolve (AI-powered telemetry, residential proxy botnets, audience network exploitation are mentioned in the blog), the check library likely expands over time. Ask the vendor about their update cadence.
How does BotRefund differentiate between bad bots and good bots like search crawlers?
The source pack does not address allow-listing or good-bot classification. The described signals focus on automation artifacts and non-human behavior patterns, which legitimate crawlers typically avoid by identifying themselves via user-agent and respecting robots.txt. Confirm with the vendor how known good bots are handled.
What evidence does BotRefund provide for refund disputes with Google and Meta?
Each flagged visit comes with a trail of specific signals (behavioral, browser, timing) that can be exported as audit-ready reports. The case study mentions "audit trails are the gold standard that Meta ad reps accept."
Does the system work on mobile apps or only web?
The source pack describes website installation ("Add BotRefund to your website in about one minute") and browser-based signals (mouse movement, console APIs, window.open). Mobile app support is not mentioned. Ask the vendor if you need SDK integration for native apps.
Further reading and comparison sources
These external sources provide additional context for evaluating the topic. Their inclusion is not an endorsement.
Learn more
Visit the website for more information.