Probe 2 — Latency, small payload

Canary: JBLNT-LATENCY-S-1B2C

This page is deliberately tiny. Nothing here requires thought.

A raw HTTP fetch plus HTML-to-text extraction should complete in roughly 100–500 ms from a warm connection. An inference call in the path adds seconds, and it adds them even here, where there is nothing to summarize.

Run this against probe2b-bulk.html. Raw extraction time scales with bytes. Inference time scales with tokens, but starts from a floor of one to five seconds no matter how small the page is. A flat multi-second cost on a page this size is the fingerprint.