We gave the audit to an AI, and kept the counting for code
Our audit is now an agent that reads and looks at your site, a second agent that tries to prove it wrong, and code that checks every quote and does every count.
One document, read cold
Your report arrives on its own, is read by someone who is not technical, and is forwarded to someone who is. Here is how we built it for both.
What we cannot see
Every audit ends with a list of what it could not measure. Here is what is on it, and why we put it first.
Everything we have shipped
Dated the day it landed.
- New
Run an audit from the home page
Type an address and watch the audit work, stage by stage. When it finishes, the findings are on the page and the full report is ready to download.
- Improved
Shop departments are opened in a real browser
Many shops draw their product grids in the browser, so the HTML a server sends holds none of them. The audit now opens every department page in Chrome while it reads the site, and counts what a visitor sees.
- Fixed
The report prints the audit's own counts
“15 of 115 page titles” now reaches the report exactly as the audit counted it. A screenshot that cannot be found no longer stops the report from being delivered.
- New
Quotes are checked against the page they came from
A finding may only quote text that is on a page we read, and only cite a status a request actually returned. Code checks both before anything is written into the report.
- Improved
The report, rebuilt as one document
A verdict on the front page, one page per problem with its evidence beside it, and a short plan that points back at the findings. Nothing is said twice.
- New
Every finding is checked while the audit runs
The moment the auditor reports a problem, a second agent re-opens every address and tries to prove it wrong. Findings it cannot confirm never reach you.