conscRAG
Run now writes several queries, four graph contexts, and ranked findings

Changelog

Product updates only. Older educational notes live in Guides and Blogs.

  • 2026-09-04

    What conscRAG added to Run

    Run now writes several OpenAlex queries, ranks about twelve base papers, hops the top three references of each, builds up to four graph-derived contexts, and writes a large set of graph-grounded ideas. You still see four ranked findings first. The rest sit behind Show all.

  • 2026-09-04

    What we removed from the MVP run

    The latest Run update deleted the MVP shortcuts: one title search, “first six hits then twelve hops,” one selected triple list, limitation-driven and analogical-transfer strategies, and the elapsed-second progress labels. Old saved runs still open. New runs do not use those paths.

  • 2026-09-04

    How a conscRAG run changed

    Press Run and the server now walks profile → queries → retrieve → graph → contexts → generate. You get a profile you can read, a query list, a provenance-labeled corpus, context chips on Figure 1, four ranked findings, and extra candidates you can expand. A run can take several minutes.

  • 2026-09-04

    Optional open PDFs on Run, off by default

    By default conscRAG still extracts triples from titles and abstracts. If you check “Also read open PDFs via OpenAlex,” the server fetches only URLs OpenAlex already marks as open access, pulls a short methods or limitations slice, and uses that in extraction. The box is off unless you turn it on.

  • 2026-09-04

    Why conscRAG shows four findings and hides the rest

    The backend plans directions first, then writes ideas, and can store many candidates. The page shows four by default, preferring one of each strategy, because a researcher can actually read four cited drafts. Show all, then filter by strategy, if you want the rest.