Community · E2 · artifact verified
Hide spoiler danmaku with typed judgments
A danmaku filter with the motto 'keep danmaku on, keep spoilers out': each scrolling comment is judged by Jev for whether it spoils the video, and spoilers are hidden while everything else stays - with A/B scripts comparing question phrasings.
01 · Role in the system
What Jev does here
Danmaku flies by too fast to moderate by hand, and spoiler comments ruin shows for everyone else. This filter sends each danmaku to Jev as typed questions - is this a spoiler for the video, how confident - and hides only the ones that pass the spoiler threshold, leaving normal chat visible. Because the judge returns calibrated probabilities, the hide line is tunable rather than binary. The repo includes provider tests, version tests and A/B scripts over question types - evidence the author measured which phrasings judge spoilers best. Built for Chinese video platforms; README bilingual.
02 · Control boundary
Where Jev sits
Jev as the spoiler judge over a danmaku stream: per-comment typed spoiler questions with calibrated confidence, tunable hide threshold, and A/B-tested question phrasings - filter without turning comments off.
Code owns the loop, permissions, thresholds, validation, and side effects. Jev owns only the bounded judgments described above.
03 · Known limits
What this evidence does not prove
- Chinese danmaku platforms; other languages need new criteria.
- 0 stars; spoiler precision/recall not formally measured.
- Spoiler judgment is inherently context-dependent - late-episode comments may false-positive.
04 · Attribution
Public sources
This is a Community record: the project was published by a third-party community author.
- meetchen ↗Community · github · public · checked 2026-10-09