The background noise of the coffee shop hummed, but inside the terminal of laptop 0x4F53, a silent, invisible theft was taking place.

Elena sat two tables away, looking at nothing more than a standard, beautifully designed web portfolio she had deployed an hour earlier. On the target's machine, a user had clicked her link. There were no permission pop-ups, no security warnings, and no antivirus alerts. The sandbox was holding perfectly.

Or so the browser thought.

Inside the malicious tab, Elena’s script initiated Project Hypersurface. It quietly tapped into the Origin Private File System (OPFS), generating a massive, multi-gigabyte phantom file. The script began hammering this file with high-frequency, random read requests.

Beneath the software layer, the host operating system panicked. To handle the sudden, massive influx of data blocks, it began aggressively flushing its lightning-fast RAM Page Cache. Within seconds, the system’s digital shield dropped. The laptop was forced out of its protected, cached memory state and dragged straight down into raw, physical hardware lookups.

Simultaneously, because Elena’s server had deployed strict cross-origin isolation headers, the browser rewarded her tab with an un-fuzzed, ultra-precise microsecond clock. The trap was sprung. The pipeline pointed its precise timer directly at the un-isolated SSD controller queue.

On the surface, the laptop's user opened a new, unrelated tab to check their banking portal, then launched Discord to message a colleague.

Each independent application fired its own data requests down to the storage drive. Because two operations cannot occupy the same physical coordinates on the hardware bus at the exact same time, the SSD controller was forced to serialize the queue, delaying Elena’s random read loop.

On Elena's screen, the flat, green $220\,\mu\text{s}$ baseline signal deformed. It erupted into a highly volatile, jagged topography of physical latency spikes ranging between $400\,\mu\text{s}$ and $900\,\mu\text{s}$.

The user's software sandboxes were completely intact, but their physical hardware was screaming.

The raw time-series wave streamed back to Elena’s server, feeding directly into a pre-trained Convolutional Neural Network. The AI treated the jagged spike clusters like an image, instantly recognizing the distinct topological "crater" left behind by the banking app's asset-loading sequence.

Elena took a sip of her coffee. The sandbox hadn't been broken; it had been bypassed entirely. By listening to the microscopic physical echoes of a shared storage drive, she knew exactly what her target was doing, one latency spike at a time.