> ## Documentation Index > Fetch the complete documentation index at: https://docs.getvector.me/llms.txt > Use this file to discover all available pages before exploring further. # Finding confidence and feedback > Judge how much to trust a monitoring finding, resolve the ones held for review, and tell Vector when a finding was wrong so detection improves. Not every "your [competitor](/competitors/add-competitors) changed something" finding deserves the same trust. Some sites show automated checks different content than real visitors, and some pages reorder or rotate content without changing anything that matters. Vector checks for this before showing you a finding, and tells you how confident it is. ## Judge how much to trust a finding In the **Check log**, a change carries a badge: **high confidence**, **medium confidence**, **low confidence** or **suspected decoy**. Hover it to see why — which service fetched the page, whether that service counts as datacenter or human-like, and whether a second check agreed: | Badge | Typically means | | ----------------- | ------------------------------------------------------------------------------------------------------------------------------ | | high confidence | A second, human-like check agreed, and the page looked genuine | | medium confidence | Only one check, but from a human-like service, and the page looked genuine | | low confidence | Only one datacenter check, or the page looked implausible, or its content changes on every load | | suspected decoy | A second check disagreed with the first — the site may be showing bots fake content. Both versions are kept for you to compare | Agreement between two checks raises confidence; it doesn't prove a change is real. Treat a high-confidence finding as trustworthy, not certain. Which pages count as high-stakes, and which services count as human-like, is set in [Connect web scrapers](/ai/web-scrapers). ## Resolve a finding held for review A low-confidence change on a high-stakes page, such as pricing, waits for you instead of flowing into [automatic re-evaluation](/monitoring/automatic-re-evaluation) or the comparison. On most pages you still get an **Inbox** notification marked **suspected decoy — needs review**; a held pricing change sends none. Find the change in the **Check log**: open the run's **details**, and on the change choose: * **Confirm** — it's a real change. Processing carries on, including automatic re-evaluation if you have it on. * **Dismiss** — discard it. It's never used as a baseline or acted on. High-confidence changes never need this step. ## Tell Vector a finding was wrong Open a finding from your **Inbox**. Under **Your feedback**, pick a verdict: * **✗ False positive** — nothing really changed on the site. * **✓ Correct** — the change is real. * **\~ Noise** — real, but not worth flagging. Add why if it helps — for example, "the pricing block just reorders on each load" — and click **Save feedback**. The verdict stays on the finding for anyone who looks at it later. Each finding has one verdict for the whole workspace. Anyone can set it, and **Update feedback** replaces it for everyone — the latest verdict is the one that counts. Pricing-change and re-evaluation notifications don't offer **Your feedback**. A competitor's **🔔 Monitoring** tab shows how many of their findings were marked false positives. ## Fix the mistakes that keep coming back On the **Proposals** tab of **Monitoring**, click **✦ Generate proposals**. Vector reads the false positives you've marked and suggests changes to how detection works — the recurring mistakes, stated as something you can act on. Nothing is applied on its own. **Accept** a proposal or **Reject** it, with an optional reason. Accepting also adds a regression test, so the same false positive can't quietly come back. ## Where to go next Where findings arrive. Setting how strictly pages are checked.