# Rights, use, and takedown notice

This repository combines three layers with different rights:

1. The packaged Python utilities are copied from Miles and distributed under
   Apache License 2.0; see `LICENSE-CODE`.
2. Benchmark questions, labels, and teacher-derived metadata originate from
   `artillerywu/DeepResearch-9K` revision
   `9eaf02da8608b91dc9440503ce10889160c14347`. That upstream repository did not
   declare a dataset license at the time this release was built. No additional
   rights over upstream content are granted here.
3. `search/corpus.sqlite3` contains extracted text from many public web sources.
   Copyright and other rights remain with the respective source owners. This
   release does not grant a blanket license over third-party webpage text.

The snapshot is provided for research reproducibility and internal benchmark
diagnostics. Users are responsible for determining whether their intended use,
jurisdiction, redistribution, and access to each source are permitted. Do not
use it to identify, profile, contact, or make decisions about individuals.

The crawl used public HTTP, respected robots policy for direct-origin requests,
applied bounded per-host access, and rejected credential-bearing URLs. Before
distribution, session/API-like query values in URL metadata were redacted and
the SQLite file was vacuumed. These technical measures are not a legal opinion
or a substitute for source-specific terms.

For a correction or removal request, open a discussion on
`Shangy/DR-256` and include the canonical URL
or `url-...` docid. A source document can be blocked or removed in a subsequent
version; that change must receive a new database hash because it can affect
retrieval behavior.

THE DATA AND SOFTWARE ARE PROVIDED "AS IS", WITHOUT WARRANTIES OR CONDITIONS
OF ANY KIND. See `LICENSE-CODE` for the software warranty disclaimer.