A Content Security Policy that broke our own logo
We shipped a strict CSP, scored well on every header scanner and quietly broke the site's own logo. Here is what inline style attributes do under CSP and why a passing scan is not proof.
Technical notes on security headers, right-to-left interfaces and running our own models.
Browse the articles
What broke, why it broke and how we fixed it.
We shipped a strict CSP, scored well on every header scanner and quietly broke the site's own logo. Here is what inline style attributes do under CSP and why a passing scan is not proof.
Mirroring a left-to-right layout produces a Persian site that technically works and reads as foreign. What actually has to change when the reading order flips.
Commercial AI APIs are not reachable from where we build. That constraint decided Sevoria's architecture, and some of the results were better than the default would have been.