DROP ALERT NOTES
A synthetic alert through the live path
A quiet market proves nothing about your chain. Inject one clearly marked fake alert a week and watch the whole path do its ordinary work.
Your relay passes its tests, the market is quiet, and a small doubt grows: would the channel actually fire tonight if a real drop happened? There is a way to know without waiting for the market: a synthetic alert, injected by hand, walked through the whole live path once a week.
The quiet-market problem
Every part of your chain can be healthy while the chain itself is dead. The reader runs, the dedupe store hums, the webhook credentials sit unexpired, and the first real alert in a week discovers the channel was renamed on Tuesday. Nothing in a quiet stretch proves the path works, because nothing travels it.
The recorded-stream test, replaying a fixture through the relay, covers the logic. What it cannot cover is the live edges: the webhook token, the channel permissions, the phone that actually buzzes. A synthetic alert covers exactly those.
Build one honest fake
A synthetic alert is just a dict wearing the real fields: home, away, sport, sect, outcome, from_price, to_price, nvp, id, alerted, the same shape the setup guide documents for the stream's events, checked 2026-11-08. Make it obviously synthetic: a fixture named for what it is, prices that read as a test. It enters your pipeline after the stream reader, at the same seam the real alerts use.
The id deserves one line of care. Give the fake a recognizable id, and let your dedupe store treat it like any other: the point is to watch the full path, dedupe included, do its ordinary work on an extraordinary event.
What the weekly drill proves
Run it on a schedule, not when you remember. Each run answers, end to end: the relay accepted the event, the rules fired, the cooldowns let it through, the webhook delivered, the message rendered in the channel with the fields in the right places.
When a run fails, the failure is small and cheap and found on a Tuesday morning instead of a Champions League night. That is the entire argument for doing it.
Keep the drill honest
Two rules keep the test meaningful. Mark it clearly, so nobody mistakes the fake for market news, and so you can filter it out of any later analysis of alert volume. And resist the urge to special-case it: a synthetic alert that skips the dedupe or the cooldown tests a path that production never uses, which proves nothing.
A quiet market proves nothing about your chain. A weekly fake, walked through the same doors as the real ones, proves everything you actually need to know before the busy night arrives.