Sharp X Timeline Filter
When the TypeSafe classifier is selected, Sharp sends a Noul question asking whether to hide each eligible X post under the reader’s plain-language filter instruction.
When the TypeSafe classifier is selected, Sharp sends a Noul question asking whether to hide each eligible X post under the reader’s plain-language filter instruction.
When the TypeSafe classifier is selected, Sharp sends a Noul question asking whether to hide each eligible X post under the reader’s plain-language filter instruction. A single-post request uses key hide; batched requests use p1, p2, and so on. The state contains each post’s author and text, with optional reply context and image descriptions, plus recent reader corrections when present. The returned yes probability is compared with the reader’s hide line to produce the verdict.
Apply the reader’s filter instruction literally; judge this post on its own merits and say whether it should be hidden.
One post with author handle, text (or “(no text)”), optional replying_to context, and optional image descriptions. If present, prior reader corrections are also included. The question includes the reader’s filter instruction.
The extension observes newly mounted timeline posts and scans the timeline; unchanged post IDs are not sent for judgment twice.
Blocked words and per-author or other local rules can decide a post before the model is called.
New eligible posts near the viewport are grouped. A full batch is sent at once; a partial batch waits for the batch window.
Jev returns hide probabilities. The configured hide line determines whether a post stays visible, is hidden, or is presented as a close call; image descriptions are requested only for posts that need a second look.
The extension turns a reader’s natural-language timeline rule into per-post yes/no decisions. It batches posts that arrive together, and carries prior corrections as examples in later states.