130Build a browser-side analytics delivery system with batching, sampling, offline queueing, and deduplication

At scale, analytics costs real money and real bandwidth. Build a delivery pipeline that sends fewer, better events:

createAnalyticsPipeline({ transport, storage, sampleRate = 1, random = Math.random, online = true, batchSize = 20, dedupeWindowMs = 1000, now = Date.now, retryDelay = 1000, createId = () => crypto.randomUUID() }) returns { ready, track(name, props, { id, critical }), flush(), setOnline(online), queueSize }.

• Sampling per session: decide once whether this browser session is sampled (random() < sampleRate) and store the decision in storage under "analytics:sampled", so reloads keep it. Unsampled sessions drop ordinary events; events tracked with critical: true (errors, purchases) are always kept.
• Deduplication: drop an event whose id was already tracked (id defaults to a new createId()), and drop an event with the same name and props as one tracked less than dedupeWindowMs ago (double clicks, re-renders).
• Batching: events { id, name, props, ts } are queued; when online and the queue reaches batchSize, flush.
• Offline queueing: the queue is saved in storage under "analytics:queue" after every change and loaded on start (ready resolves after loading), so nothing is lost across reloads or offline periods. setOnline(true) flushes.
• Delivery: flush() sends batches of up to batchSize, oldest first, with await transport.send(batch). A successful batch is removed; a failure stops the flush, keeps the events and schedules a retry after retryDelay.

storage has getItem, setItem and removeItem.