← all recipes

"Ambient Tutor" — teach the interface quietly based on hesitation, return, and attention. It has a memory layer (a decaying trail / hysteresis), a gravity layer (universal attraction toward mass), a propagate layer (travelling waves through the medium), and a link layer (ropes / cloth between neighbours). The active render stack is heatmap, trails, particles, links.

Memory watches hesitation and repeated returns while gravity gives the right explanation subtle priority near the point of need; propagate and link carry help to related controls.

Runtime tokens

memory gravity propagate link

Metrics

hesitation return helpfulness attention hesitation, return, helpfulness are host-supplied lanes — drive them with data-field-<metric> (or a domain model); without that, their --field-* stays inert.

Diagnostics

heatmap causality inspector

Conditions

dwell return

Render

heatmap · trails · particles · links

Reduced motion

contextual tips with a help rail and related-explanation markers

Without motion

help surfaces as a contextual tip near the point of need with a list of related explanations

<field-root></field-root>
  <div data-body="memory" data-strength="1" data-range="300" data-feedback></div>
  <div data-body="gravity" data-strength="0.7" data-range="360"></div>
  <div data-body="propagate" data-strength="0.5" data-range="320"></div>
  <div data-body="link" data-strength="0.6" data-range="280"></div>