pnclENGINEGet API access

CAPACITY NOTES

When you must shed load: what to drop first

A shedding order written before the busy slate: detail calls before boards, intervals before sports, tooling before customer reads.

The 429s are stacking up and something has to give. The worst version of this moment is a loop that keeps trying everything and loses everything. The better version is a shedding order decided in advance: what your pipeline drops first, what it protects to the last, and in what sequence. Write it down before the busy Saturday, because the order is obvious now and invisible then.

Protect the board, shed the detail

The first cut is depth, not coverage. A board request keeps every event in a sport current; a detail call deepens one event. When the ceiling presses, the detail calls go first: you lose alternates and depth on individual matches and keep the whole board honest. The alternative, dropping boards and keeping detail, is coverage with holes and detail nobody can navigate.

Per-event calls are also where load spikes come from, as the capacity planning guide shows the two request shapes, checked 2026-11-01. The thing that is easiest to cut is usually the thing that overflowed.

Widen intervals before dropping sports

The interval is a smooth dial and scope is a binary switch. Halving freshness everywhere keeps every sport's record complete and merely older; dropping a sport erases it entirely. Shed with the dial first, the switch last.

The interval math is the forgiving kind: doubling the interval halves the load, with no gaps to explain later. A slower board for an hour is a footnote. A missing sport is a missing product.

Match the shedding to the phase

Live and prematch are separate loads, and which one you protect depends on what your product is for. An alerting product protects the live loop and lets prematch coast at minutes-wide intervals. A comparison site protects prematch boards and lets live detail wait. There is no universal right answer, only an unexamined one.

This is why the order belongs in a config file, not in someone's memory. The middle of an incident is the wrong place to discover that your gut and your colleague's gut disagree about which phase matters.

Keep the reserve, shed the extras

The rate limits guide recommends steady state under roughly 70 percent of the per-second quota, reserving the rest for retries and tooling, checked 2026-11-01. The shedding order protects that reserve in reverse: admin tooling, second environments and nice-to-have refreshes go before any customer-facing read.

And when the moment passes, shed in reverse too. Restore detail calls and tighter intervals gradually while watching the 429 count, because the busy slate that caused the spike is still running and a full-speed snapback just re-finds the ceiling.

Write the order once, keep it in config, and the next big slate becomes a test you already passed.