Technical forensic specification explaining how BotRefund analyzes browser environments and network invariants to distinguish novel retail browsers from server-side automation.
Human neuromuscular motor coordination produces continuous chaotic micro-tremors, non-zero curvature jerk, and variable velocity arcs. Interaction is governed by cognitive reading pauses (80ms–150ms per keystroke; 2,000ms–8,000ms form dwell times) and natural visual scanning.
Automated scraping rings route clicks through commercial datacenter servers or residential proxy tunnels. Tunnel encapsulation (WireGuard/SOCKS5) clamps TCP SYN MSS to <= 1420 bytes, leaks residential IPs via WebRTC, or routes DNS queries through discordant overseas resolvers.
Synthetic Mouse Trajectory Replay Cluster ($O(1)$ SimHash Cross-Session Correlation): Arc-length normalized mouse movement generated an identical 45-bit spatial angle SimHash that perfectly matches trajectories recorded in prior sessions across distinct visitor tokens or residential IPs. Biological neuromuscular human motor control is subject to chaotic micro-tremors and variable velocity arcs, making identical normalized trajectories statistically impossible ($p < 10^{-9}$). This match proves cross-session automated script execution using hardcoded waypoint paths or synthetic Bezier curve generators.
Ad Platform Fraud Classification: Under Google Ads Invalid Traffic guidelines (Automated Click Activity) and Meta Advertising Integrity standards, sessions exhibiting Trajectory Replay Cluster violate fundamental consumer hardware invariants. They fail deterministic silicon rasterization, protocol compliance, or neuromuscular biometrics.
Why This Proves Non-Human Origin: Genuine consumer browsers on retail operating systems execute this API natively with verified hardware backing. Server software (headless cloud containers, unpatched automation harnesses, and proxy tunnels) cannot replicate this execution environment without incurring prohibitive server hardware costs.
Evidence Dossier Invariants: Each disputed click in our export packet matches an immutable server-side timestamp, landing URL, GCLID/fbclid, and observed IP address with zero client-side tamperability.
Human Baseline & Exception Scenario: Absolute zero for genuine humans across distinct sessions. Resampling to 16 points along the arc-length normalizes for screen resolution and display DPI while preserving spatial angular geometry.
BotRefund operates a multi-vector ensemble model. This signal adds objective forensic weight to the session dossier but never acts in isolation. Positive verdicts require corroboration across network routing, hardware silicon, and cursor kinematics to guarantee 99% precision.
This signal registers an objective, mathematically verifiable anomaly on the session ledger.
BotRefund correlates L4 packet timing, GPU silicon, and cursor jerk before finalizing risk.
Compiled directly into Google Ads and Meta refund packets with verified click IDs and server logs.