What's next
Your first watch is running. Pick the thread that matches what you want next.
Make it more accurate
Section titled “Make it more accurate”Run it seriously
Section titled “Run it seriously”Build on it
Section titled “Build on it”Understand it properly
Section titled “Understand it properly”The engine is worth twenty minutes if you plan to rely on it: How it works → Decision rules → Failures and retries.
Three things that catch most people out
Section titled “Three things that catch most people out”Reading these now saves an afternoon later.
Content is judged exactly once. Watcher remembers what it has already read, so improving a condition does not make it re-examine pages it already consumed. To test a change, create a watch with a new id. See What Watcher remembers.
An ungraded source cannot confirm anything. Watcher recognises only a small built-in set of domains. Everything else starts ungraded and is capped at the “unconfirmed” tier, however confident the model is. One trust rule for the site you actually care about usually fixes a whole class of surprise. See Source trust.
Your real check frequency is the slower of two clocks — the watch’s own cron, and how often the instance runs. A five-minute cron under an hourly job is an hourly watch. See Scheduling.
A good first week
Section titled “A good first week”- Run one watch you care about and leave it alone for a few days.
- When alerts arrive, label them 👍 or 👎 in the panel.
- Read the run history for a watch that stayed quiet — confirm it is quiet for the reason you expect, not because collection is failing.
- Then tune, once, using Tune precision.