Small observations. Moving interests.
Synthetic labelled groups, not discovered clusters. Coordinates are explicit features, not model output. The 2D display shows two of three coordinates; all three determine the L2 ranking.
See where the request actually goes.
Scroll normally. Use the toolbar to enable diagram interaction, zoom, export or open fullscreen.
An educational replay of a protective reverse-proxy topology, not live Stylo.Bot traffic or a complete detector implementation.
24 users. Watch their journeys diverge.
Heads follow the measured behavioural centroids on a shared UTC timeline. The two feature coordinates are fixed; the Separate tracks control adds fixed vertical display lanes so coincident paths remain visible. These offsets have no statistical meaning; switch them off to see the exact feature projection. Forecasts also use velocity, acceleration and elapsed time in the source space. Compare direction or pace to see why the same observed position can have different futures.
24 synthetic timestamped visits against a frozen model trained on 600 earlier episodes. The UTC clock and playback speed use recorded event time. Forecast horizon: 60 seconds from the displayed observation. Fast centroid half-life: 2 seconds. Service normal-reference half-life: 24 hours; only qualified healthy observations update it. Unfinished visits are not labelled as failures. No intervention effect is simulated.
View the current data as a table
| Entity | Values / evidence | Result |
|---|