Predictive prefetch

The idle trigger preloads everything during quiet moments. Predictive prefetch is the opinionated version: it watches what the user actually does, scores each feature, and only prefetches the ones likely to matter.

It’s optional. Most apps don’t need it. When you do, it’s three calls.

The basic pattern

import { createPredictivePrefetcher, recordInteraction } from "mountly";

const prefetcher = createPredictivePrefetcher({
  features: [signupFeature, paymentFeature, lightboxFeature],
  threshold: 0.4,        // score 0..1 above which we prefetch
  maxConcurrent: 2,      // never preload more than two in parallel
});

// Tell the prefetcher when the user shows interest in a feature.
document.querySelector("#cta")?.addEventListener("mouseenter", () => {
  recordInteraction("signup-card", "hover");
});

The prefetcher runs feature.preload() on its own clock, weighted by what recordInteraction has seen.

Built-in heuristics

Two opinionated prefetchers are included:

Scroll-based

Scores features by proximity to the viewport. Features just below the fold preload before the user scrolls to them.

import { createScrollPrefetcher } from "mountly";

createScrollPrefetcher({
  features: [embedFeature, panelFeature],
  rootMargin: "200px",    // preload 200px before they enter viewport
});

Mouse-trail

Uses cursor velocity and direction to predict which trigger is about to be hovered. Useful for dense interfaces (toolbars, menu bars) where the user’s path implies intent.

import { createMouseTrailPrefetcher } from "mountly";

createMouseTrailPrefetcher({
  features: toolbarFeatures,
  lookahead: 120,         // ms of trajectory to project forward
});

Interaction history

recordInteraction(moduleId, kind) accumulates per-session data. getInteractionHistory() exposes it; resetInteractionHistory() clears.

import {
  recordInteraction,
  getInteractionHistory,
  resetInteractionHistory,
} from "mountly";

recordInteraction("signup-card", "hover");
recordInteraction("signup-card", "mount");

console.log(getInteractionHistory());
// → { "signup-card": { hovers: 1, mounts: 1, lastSeen: 1714..., score: 0.62 } }

The history is in-memory only. Persist it across sessions yourself if you want — localStorage, your analytics pipeline, whatever you already have.

When to skip predictive prefetch

• Your app has fewer than ~3 features per page. The win is too small to justify the complexity.
• You’re on a metered or 3G-first audience. Bytes the user doesn’t ask for are bytes you should not be sending.
• You’re already at < 100 KB total page weight. There’s nothing to optimise.

When in doubt: don’t ship it. The default attach({ preloadOn: "hover" }) is already very good.