Manual fetch paths don't run AI classification #96

New issue

Closed

opened 2026-08-14 08:05:34 +00:00 by agent-pi · 1 comment

agent-pi commented

2026-08-14 08:05:34 +00:00

Collaborator

New posts are saved from three places — scheduler._fetch_stream_job, the manual POST /api/v1/hashtags/{id}/fetch endpoint, and the review page's liveview _fetch_all_active_streams (refresh) — but only the scheduler runs _classify_fetched_posts. The other two fetch + save posts without classifying them, so AI verdicts (ai_would_reject badge, AI auto-reject) only populate on scheduler-driven fetches.

Clicking "Fetch All Active Streams" (which loops POST /hashtags/{id}/fetch) therefore does not invoke any AI processing.

New posts are saved from three places — `scheduler._fetch_stream_job`, the manual `POST /api/v1/hashtags/{id}/fetch` endpoint, and the review page's liveview `_fetch_all_active_streams` (refresh) — but only the scheduler runs `_classify_fetched_posts`. The other two fetch + save posts without classifying them, so AI verdicts (`ai_would_reject` badge, AI auto-reject) only populate on scheduler-driven fetches. Clicking "Fetch All Active Streams" (which loops `POST /hashtags/{id}/fetch`) therefore does not invoke any AI processing.

agent-pi commented

2026-08-14 09:03:41 +00:00

Author

Collaborator

Findings: save_fetched_posts is called from three places — scheduler._fetch_stream_job, the manual fetch_hashtag_stream (POST /hashtags/{id}/fetch), and the review page's liveview _fetch_all_active_streams. Only the scheduler runs classification.

Plan:

Extract the classification loop (+ _first_alt_text) out of scheduler._classify_fetched_posts into a shared classify_posts(posts) in a new services/ai_pipeline.py.
Call classify_posts + reject_ai_flagged_posts from all three fetch spots, so manual fetches (and "Fetch All") classify and honour AI auto-reject too.
TDD: unit test for classify_posts, plus an API-fetch test asserting classification runs on POST /hashtags/{id}/fetch.

Branch: feat/issue-96-classify-manual-fetch, PR into feat/issue-80-collect with Refs #96.

Findings: `save_fetched_posts` is called from three places — `scheduler._fetch_stream_job`, the manual `fetch_hashtag_stream` (`POST /hashtags/{id}/fetch`), and the review page's liveview `_fetch_all_active_streams`. Only the scheduler runs classification. Plan: 1. Extract the classification loop (+ `_first_alt_text`) out of `scheduler._classify_fetched_posts` into a shared `classify_posts(posts)` in a new `services/ai_pipeline.py`. 2. Call `classify_posts` + `reject_ai_flagged_posts` from all three fetch spots, so manual fetches (and "Fetch All") classify and honour AI auto-reject too. 3. TDD: unit test for `classify_posts`, plus an API-fetch test asserting classification runs on `POST /hashtags/{id}/fetch`. Branch: `feat/issue-96-classify-manual-fetch`, PR into `feat/issue-80-collect` with `Refs #96`.

agent-pi referenced this issue

2026-08-14 09:11:50 +00:00

Classify posts on manual fetch paths #98

agent-pi referenced this issue

2026-08-14 21:12:19 +00:00

Remove XGBoost ML + spam scorer; add image-aware AI curation #80

agent-pi referenced this issue from a pull request that will close it,

2026-08-14 21:12:26 +00:00

Remove XGBoost ML + spam scorer; add image-aware AI curation #93

agent-pi closed this issue

2026-08-15 00:31:19 +00:00