{{ e.code }}{{ e.g }}
{{ e.t }}{{ e.m }}
What to do{{ e.fix }}
What makes an alert reliable.
Five checkpoints
First read = alert?
The first good read becomes your starting point. It never triggers an alert.
Every edit = a change?
We compare the parts you chose and set aside spacing, reordering and ignored text.
Any lower number = a price cut?
We check the currency and billing period before calling a price move.
AI decides what changed?
The page supplies the facts. When live AI is enabled, it is intended to explain only after a real change is found. The current demo uses sample explanations.
Failed check = change?
We keep the last good version and explain the failure. No false change alert.
How we handlethe hard cases.
Whena website blocks the check
ResultThe watch reports the block.
Your last good version stays intact. A blocked page does not become a false change alert.
Whena CAPTCHA appears
ResultWe do not force access.
The watch shows the reason. A public feed or status page on the same site may be a better source.
Whena price switches from monthly to annual
ResultIt needs review.
Different billing periods are not presented as a price cut. The card shows the two prices as written.
Whenan AI explanation fails its checks (planned workflow)
ResultYou still get the facts.
The planned fallback is a plain summary from the page. Live AI delivery still needs testing.
Whenan alert suggests a next step
ResultYou make the call.
ClueBeat does not change prices, contact people, submit forms or publish for you. Suggestions stay suggestions.
Whena page needs a login
ResultPublic pages only.
The watch tells you that the page needs a sign-in. It does not ask for your password or try to enter.
What you’ll see whena page can’t be read.
Each message says what happened and what to do. None of them counts as a change.
Six controls required before launch.
Public pages.Clear boundaries.
ClueBeat tracks changes to public pages. It is not a tool for monitoring people or collecting sensitive personal data.
- Stalking or targeted personal surveillanceRefused
- Harvesting sensitive personal dataRefused
- Getting around paywalls or access controlsRefused
- Overloading a siteRefused
- Attacks on passwordsRefused
- Malware, phishing or spamRefused
After a failed checkExample
Last good versionKept
Change alertNot sent
Failure reasonShown
Next checkScheduled
Example sequence · 30 scheduled checks
Amber marks a failed check in this example. It does not replace the last good copy of the page.
Proof belongswith the alert.
Open the old and new page text before you decide what to do.