pnclENGINEGet API access

CAPACITY NOTES

How much data a day of polling actually moves

Requests are only half the volume. A full board can re-read 1,522 events per call while a cursor delta reads a handful, and the habit decides the rest.

Capacity planning counts requests, and that is the right headline number. But there is a second volume worth a line in the notebook: how much data those requests actually move. The answer swings by two orders of magnitude depending on one habit, and it decides whether a day of polling is a rounding error or a bandwidth line item.

The two numbers of a poll

Every board request has a request cost, one call against your rate allowance, and a payload cost, the bytes of the response. The request cost is fixed by your schedule. The payload cost is set by how much of the board you ask for again.

The REST polling guide carries the documented contrast, checked 2026-10-27: a full prematch soccer board can return 1,522 events, while the delta read after a quiet minute returned 7 changed events. Same endpoint, same schedule, and the payload difference is the difference between re-reading a library and reading a Post-it note.

What a day of each looks like

Run the arithmetic on a steady prematch board, one call every 30 seconds around the clock: 2,880 reads a day. Done as full snapshots, that is on the order of 4.4 million event payloads a day, the same thousand-plus events re-read nearly three thousand times. Done as cursor deltas at the documented quiet-minute ratio, it is tens of thousands of event payloads, mostly concentrated near the close when prices actually move.

Neither number will trouble a modern connection. But the larger one troubles everything downstream: the parser re-processes events that did not change, the store de-duplicates rows that were never new, and the monitoring graphs spend their range on noise. Payload volume is request volume's shadow, and it falls on every stage after the fetch.

The cursor is the payload budget

The since cursor is usually introduced as a freshness feature or a politeness to the API. It is also your data-volume control. A poller that persists its cursor and resumes from it asks only for what changed, and a day of changes is small even on a busy board, because most events sit still most of the time.

The practical rule: full snapshots are for first boot and for rebuilding after state loss, as the polling guide frames them, and delta reads are for everything in between. If your logs show full-board responses in steady state, your data volume is roughly two hundred times what it needs to be, and the fix costs nothing but the cursor you already get back.

When the payload still matters

Delta polling shrinks the steady day, not the exceptional one. The minutes before a big slate's kickoffs change most of the board, so deltas grow toward full-board size exactly when you care most. Budget bandwidth and parse time for the busy hour, not the average one, and let the quiet hours be the savings.

Count requests for the plan, count payloads for the pipeline. The cursor keeps both small, and the day stays cheap in both currencies.