Community · E2 · artifact verified
Write a morning research note per topic
For each topic you follow, jrp runs a pipeline: your questions steer the web search, Jev decides which findings count, and an LLM you choose writes the morning note - in English, Chinese or Japanese - with parallel stages and recorded claim-fidelity checks.
01 · Role in the system
What Jev does here
The search is steered, not crawled: your standing questions shape each day's queries, and every result that comes back is judged by Jev before it counts - typed decisions on what is relevant and what a claim actually says, with the recorded systemone request bodies kept in live cassettes so drift between runs is measurable. An LLM of your choice then writes the note from only what passed, in your language, and the pipeline runs stages in parallel with end-to-end and parallel test suites. The division is deliberate: judgment-only Jev filters, the generative model explains, and the tests prove the drafts never leak back into the judgment layer.
02 · Control boundary
Where Jev sits
Jev as the relevance and claim judge of a search-steered research pipeline: standing questions shape queries, typed decisions filter what counts, cassettes record the request bodies for drift checks, and a chosen LLM writes the note in three languages.
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
- Topic quality depends on the standing questions you write.
- 3 stars; Japanese-authored, docs in English and Japanese.
- The note-writing LLM is your choice and its cost is separate from Jev's.
04 · Attribution
Public sources
This is a Community record: the project was published by a third-party community author.
- shimo4228 ↗Community · github · public · checked 2026-10-04