Classifier mode: own providers, confidence lines, debug inspector - #4
Merged
Merged
Conversation
… debug inspector
A decision model now judges posts by default: one request per post, a
probability back, no written reason. It is reached through OpenRouter,
Vercel AI Gateway or TypeSafe AI, each under a key of its own, apart from
the chat model's. A chat model still works, and either connection can be
set up and tested without the other.
- The score travels with the verdict and is cached. The hide line and the
close-call margin are read against it at display time, so moving either
re-buys nothing.
- Images are described in words by a vision model, and only for posts
whose text-only score lands in a range the reader picks. A range
covering every score skips the text-only pass.
- Banners say how sure the hide is ("93% sure", "only 62% sure"), never a
model name or a bare figure.
- Debug mode puts a chip on every judged post that opens an inspector:
request, response, timings per step, image thumbnails and descriptions.
Traces live in the tab's memory only and are never cached.
- Concurrency: a worker-wide gate for the classifier, up to 64 posts at
once; the chat model's request cap goes to 12.
- Settings: X gets Appearance and Advanced tabs for what is X's own
(hidden-post style, mode, speed or spending, confidence, images,
lookahead). General keeps AI connections, routing, debug and
maintenance. Sections are drawn as titled sheets.
- New hosts are asked for on save rather than declared, so an update
does not make existing installs re-approve the extension. A reader who
set up a chat model outside OpenRouter keeps it deciding after updating.
|
Note Currently processing new changes in this PR. This may take a few minutes, please wait... ⚙️ Run configurationConfiguration used: Organization UI Review profile: CHILL Plan: Advanced Run ID: 📒 Files selected for processing (30)
✨ Finishing Touches📝 Generate docstrings
🧪 Generate unit tests (beta)
Comment |
X serves the reader's bookmarks at /i/history as well as /i/bookmarks. Both are pages the reader filled themselves, so nothing on them is judged or hidden.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
A decision model now judges posts by default, in place of a chat model. It is faster, nearly free, and returns a probability, which the rest of this PR is built around.
What changes for a reader
google/gemma-4-26b-a4b-it.Upgrade safety
ai-gateway.vercel.sh,api.typesafe.ai) are requested on save, not declared in the manifest, so the update does not disable existing installs pending re-approval.Not verified live
The TypeSafe and Vercel request shapes come from their docs and an open-source client; neither was run against a real key. OpenRouter was.
117 tests pass;
bun run checkis clean.