Content Moderator AI Block for Svelte

Everything. The twelve posts and their ranks are typed at the top of the file as example figures; no model scores anything, and "Apply" only changes the status line: nothing is removed, hidden or sent. To make it real, load posts with scores from your classifier, send the person's calls and the slider's threshold to your API in apply, and keep the banding function, which only needs a rank and a threshold.

Two lines are drawn from one number. Posts ranked below the flag are auto-allowed, posts from the flag up go to review, and posts ranked 30 or more above it are auto-blocked; moving the flag toward Strict sends more posts to a person, toward Lenient lets more through. The counts above the table and the label on the button come from the same function as the rows, so they always agree.

They settle one post that needs review: its status becomes "Allowed by you" or "Blocked by you", the count above the table follows, and Undo puts it back in the queue. A decision only counts while the post is still in the review band; if the slider moves it out, the band decides again. Focus moves to Undo after a decision and back to Allow after an undo.

The slider reads "Flag at 50, block at 80" and announces the result once, in the status line, when you let go. The table has a hidden caption, column headings and a row heading for every post, and its buttons are named for their post ("Allow post by SL"). Status is a word and a glyph, never a color.

Yes. The slider and the labels follow the direction, posts keep the direction of their own text, and the ranks and counts stay together inside the right-to-left text.

The table becomes a list of cards with the policy label and risk on top, the post in two lines, and the status with its buttons at the bottom. Every card has the same height, so a decision never moves the list.