Evidence

Evidence

Others hand you screenshots; we hand you records. A link that stops opening turns its own light off, and zero samples report “not measured” rather than 0% — those are rules written into the product, not promises made by marketing.

Every number Sourced delivers has a record behind it you can open: link verification results, raw AI transcripts, the four-layer attribution chain. This page shows what they look like, and how we make faking them impossible in code.

Four kinds of evidence

Link verification records

When you paste back a URL from an encyclopedia, a map listing or a community post, we fetch it for real and record the HTTP status, the title we got, and whether your entity name still appears on the page. Fail any of the three and the channel does not count.

The record holds: fetch time, status code, and the matched entity snippet.

Raw AI transcripts

Monitoring is not simulated. We actually ask Doubao, DeepSeek, ChatGPT and the rest, and keep the full answer. Both halves of any mention rate trace back to specific transcripts.

The record holds: platform, prompt, full answer, hit or miss, and position.

The four-layer attribution chain

AI mentions you, someone clicks, a lead appears, a deal closes — each layer recorded separately, so you can see which one it breaks at. Thin samples are labelled as thin; nothing is interpolated.

The record holds: sample size per layer, conversion, and an explicit low-sample flag.

Evidence pack export

All of the above, packaged into one deliverable for your boss or your own client. It contains links and raw records rather than a re-typeset conclusion.

Exports as: weekly Markdown, client-ready Excel, or raw JSON.

link_verify · redacted
  • urlzhihu.com/question/… · HTTP 200
  • entity on pagematched entity · paragraph 2
  • ownershipconfirmed as submitted by this project
  • retrieval0 samples → “not measured”, never 0%

Three rules we hold ourselves to

All three live in the code rather than in a deck. Ask us to demonstrate any of them.

A link that won’t open doesn’t count

Verification re-runs on a schedule. A page that works today and gets deleted next month turns its own light off and lands back in your to-do list.

Never measured, never 0%

With no samples, the scorecard says “not measured”. Rendering work you never did as 0% deceives you as much as your client.

Two hard gates, or it isn’t live

One verified third-party URL, one real monitoring run with samples. Miss either and the workbench keeps it at the top of the home screen — it is also the only way we judge internally whether a project has actually delivered.

Want to see real evidence?

Open the workbench and run it yourself, or let us walk you through a live project.

Free lookOpen Xiaoyuan