shadcn.io is not affiliated with official shadcn/ui
Guardrails Violation Log Monitoring Block for Solid
Four safety policies as a radio list with each share of 48,200 requests, a tick at the 1% promise and an Over or Within word, plus the six recent violations of the chosen policy with the offending words drawn as bars and a Confirmed / False positive toggle on each.
Installation
Overview
Check whether a safety gateway keeps its promise, and whether it is wrong too often, with this violation log for Solid with Ark UI. It is Railwarden, an LLM safety gateway, and it checked 48,200 requests in the last 7 days; the header counts how many of its four policies (personal data, toxicity, prompt injection, off-topic) are over the 1% promise. In a hairline frame with a plus mark on each corner, the policies are one radio group: each row has a bar with a tick at 1%, the share with the word Over or Within, the count and a seven-day line. Choosing a policy lists its six recent violations with the time, the request id in monospace, what the gateway did (Blocked, Redacted or Flagged), and the excerpt with the offending words drawn as a solid bar beside the reason, never the words. Each violation has a Confirmed / False positive toggle, and the foot works out the false-positive count and share again ("6 of 24, 25%"). Guides draw the frame and its bands; the entrance is CSS only.
FAQ
How do I use my own policies and violations?
Edit POLICIES (seven daily counts each; the total and the share of REQUESTS come from them) and SAMPLE, whose entries are policy|time|request|action|before|length|after|reason|false. The bar width, the Over word, the header sentence and the false-positive count are worked out from them. The promise is the PROMISE constant.
Why are the offending words never shown?
The excerpt is built from the text before and after, with a solid bar of the right length in between and "redacted" for screen readers. That is how a review screen can show context without displaying the personal data or the attack itself.
What do Confirmed and False positive do?
They are a local label per violation: pressing False positive moves it into the count in the foot and in the band heading, pressing Confirmed takes it out again. Nothing is sent to the gateway; in your app write the label to your review queue and use it to tune the policy.