TOOL NOTES

The request estimate is a floor, not a budget

The cost calculator's request count assumes a perfect scheduler. Add the failure and growth multipliers before you defend the number.

The API cost calculator turns a polling schedule into a request count: one request every 10 seconds, 24 active hours, 30 days and one request per cycle comes to 259,200 requests, per the formula on its page, checked 2026-10-21. That number is exact, and it is also a floor. It assumes a scheduler that never retries, a scope that never grows and a calendar that matches reality. Production grants none of those. The budget you defend should carry the overhead on top.

What the clean number assumes

The formula is (3,600 divided by the interval in seconds) times active hours, days and requests per cycle. Every term is a decision you control, which is what makes the tool useful. The same exactness is the trap: the page itself lists what it leaves out, retries, pagination, event growth and idle periods among them. Those omissions do not disappear because the arithmetic is tidy; they move to your side of the ledger.

Multiplier one: failures

A healthy pipeline still retries. Timeouts, a 5xx under load, a restart that forces one full board read per sport and phase to rebuild the cursor: each is rare and each costs requests. On a stable setup this padding is small, a few percent of the total. Price it anyway, because a month with one bad night is the month the unmultiplied estimate misses.

Pick your own figure from your logs once you have them. Until then, a single-digit percentage is a planning assumption, not a fact about the API, and writing it down as an assumption is the point.

Multiplier two: scope creep

The larger risk is not failure; it is success. A second phase doubles requests per cycle. Halving the interval doubles the whole product, and the formula says so plainly. The preset plans make the same distinction from the billing side: the paid REST tiers carry no volume cap and are limited by rate, 10 requests a second at $99 a month and 30 at $229, while the free key is 100 requests a day, per the calculator's preset notes, checked 2026-10-21.

So run the calculator twice: once for the schedule you launch with and once for the schedule you will defend in three months. The second number is the budget; the first is the invoice month one.

The honest budget

An honest estimate is the floor times a multiplier you can explain: the clean count, plus single digits for failures, plus the growth you already know is coming. A budget of 293,000 built from 259,200 that way survives a retry storm and a new league. A bare 259,200 survives nothing but the demo.

The same posture covers the rate side. Whatever interval you choose, its per-second rate needs the same headroom, because the ceiling a plan publishes is a limit, not an allotment. The tools page collects the calculators; run the estimate, pad it, then pick the plan that carries the padded number.

A number you can defend beats a number that is merely right. The floor comes from the calculator; the multiplier comes from your logs.