pnclPULSEGet API access

DROP ALERT NOTES

A week at min_drop 1: what the floor shows you

Run the stream at its floor for one deliberate week, log everything, alert on nothing new, and replay the log to set thresholds from evidence.

The stream's floor is min_drop at 1, and almost nobody should run production there. But the floor has a second use that has nothing to do with production: one deliberate week at 1 is a census of everything your sports actually do, and the log it produces is the ground truth every later threshold decision leans on.

Why the floor, once

Every threshold you set is a guess until you have seen the raw supply. The drop alerts guide documents the window arithmetic and the floor at 1, checked 2026-11-04. Running at the floor for a fixed week answers the questions you cannot estimate: how many moves per day does each of your sports produce, at what sizes, at what hours.

This is a measurement, not a lifestyle. You are buying a week of noise to learn the shape of the signal.

What the week teaches

Volume by sport is the first read. Some leagues move constantly; some sit still for days. The counts per sport per day, broken by drop size, tell you where a 3% rule is calm and where it is a siren.

The daily shape is the second read. Drops cluster near kickoffs and big team news, so your log will show the hours when the feed is loud. Those hours are where your alert budget gets spent, whatever threshold you pick, and knowing them is worth the week on its own.

Surviving the noise

A floor week needs discipline or it becomes the week nobody reads. Log everything, alert on nothing beyond your existing production rules, and let the two lanes run side by side. The floor stream writes to a log file; the production relay keeps behaving as usual. If your relay keys its dedupe by alert fields, keep the two stores apart so the census never leaks into the pager.

Check the log for ten minutes a day, not the channel. The point is the aggregate, not the moment, and aggregate reading is a morning habit.

From census to threshold

At the end of the week, replay the log against candidate thresholds: for each candidate, count how many alerts it would have fired per day, per sport, per hour. The number you choose should be the one whose daily volume you can actually act on, not the smallest number that catches everything.

The dedupe guide explains how one falling price can fire several times, which is why raw counts need a dedupe pass before they mean alerts, checked 2026-11-04. Replay with that pass included, or your census overstates everything.

One week at the floor, logged and unread until morning. After that, every threshold you ever set is a measurement, not a guess.