parallax-bench 0.0.1__tar.gz

parallax_bench-0.0.1/.github/workflows/ci.yml

@@ -0,0 +1,50 @@
+name: ci
+on:
+  push:
+    branches: [main]
+  pull_request:
+  workflow_dispatch:
+
+jobs:
+  test:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.12", "3.13"]
+    steps:
+      - uses: actions/checkout@v4
+        with: { fetch-depth: 0 }
+      - uses: actions/setup-python@v5
+        with: { python-version: "${{ matrix.python-version }}" }
+      - run: pip install -e ".[dev]"
+
+      # 1. dataset integrity
+      - run: parallax-bench validate
+
+      # 2. the quickstart contract — offline, no GPU, under two minutes
+      - name: smoke end-to-end
+        run: |
+          timeout 120 parallax-bench run --system baseline-local --subset smoke
+          parallax-bench score
+          parallax-bench report
+
+      # 3. hygiene
+      - run: ruff check src tests benchmark/build
+      - run: mypy src/parallax_bench
+      - run: pytest -q
+
+  secrets:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@v4
+        with: { fetch-depth: 0 }
+      - name: gitleaks (full history)
+        run: |
+          curl -sSfL https://github.com/gitleaks/gitleaks/releases/download/v8.24.3/gitleaks_8.24.3_linux_x64.tar.gz | tar xz gitleaks
+          ./gitleaks git --no-banner --redact --exit-code 1 .
+      - name: no production hostnames or instance topology
+        run: |
+          ! grep -rEn "edgeant\.cz|instances\.conf|JWT_SECRET_KEY *= *[^$ ]" \
+              --include="*.py" --include="*.toml" --include="*.md" --include="*.json" \
+              --exclude-dir=.git . \
+            || { echo "::error::production identifier or secret in the diff"; exit 1; }

parallax_bench-0.0.1/.github/workflows/release.yml

@@ -0,0 +1,38 @@
+name: release
+on:
+  release:
+    types: [published]
+  # manual re-publish of an existing tag (e.g. the release event got lost):
+  #   gh workflow run release.yml --ref v0.1.0
+  workflow_dispatch:
+
+jobs:
+  build:
+    runs-on: ubuntu-latest
+    steps:
+      - name: refuse manual runs that are not on a tag
+        if: github.event_name == 'workflow_dispatch' && !startsWith(github.ref, 'refs/tags/')
+        run: |
+          echo "::error::hatch-vcs derives the version from the tag — a run on a branch would publish a dev version. Use: gh workflow run release.yml --ref vX.Y.Z"
+          exit 1
+      - uses: actions/checkout@v4
+        with: { fetch-depth: 0 }      # hatch-vcs needs tags
+      - uses: actions/setup-python@v5
+        with: { python-version: "3.12" }
+      - run: pipx run build
+      - run: pipx run twine check dist/*
+      - name: verify bundled data made it into the wheel
+        run: python -m zipfile -l dist/parallax_bench-*.whl | grep -q "_data/smoke/queries.jsonl"
+      - uses: actions/upload-artifact@v4
+        with: { name: dist, path: dist/ }
+
+  publish:
+    needs: build
+    runs-on: ubuntu-latest
+    environment: pypi
+    permissions:
+      id-token: write               # OIDC — the whole trick
+    steps:
+      - uses: actions/download-artifact@v4
+        with: { name: dist, path: dist/ }
+      - uses: pypa/gh-action-pypi-publish@release/v1

parallax_bench-0.0.1/.gitignore

@@ -0,0 +1,11 @@
+.idea
+.venv
+__pycache__/
+*.pyc
+.pytest_cache/
+.mypy_cache/
+.ruff_cache/
+dist/
+*.egg-info/
+corpus-store/
+.parallax/

parallax_bench-0.0.1/CHANGELOG.md

@@ -0,0 +1,18 @@
+# Changelog
+
+## Unreleased
+
+- Data format (`queries.jsonl`, `qrels.txt`, `manifest.jsonl`) and
+  `parallax-bench validate`
+- Adapter protocol (`describe` / `index` / `search` / `generate`) with two
+  adapters: `baseline_local` (built-in BM25 + optional dense) and `edge_ant`
+- Retrieval metrics (nDCG, Recall, MRR, Success@1) verified against
+  `pytrec_eval`; bootstrap CIs, paired Wilcoxon, Holm correction
+- Mechanical generation metrics: citation accuracy, response language
+  correctness
+- Resumable DB-backed runner (SQLite/Postgres), `run` / `score` / `report`
+- `smoke` subset: 10 documents × 3 languages with texts, offline quickstart
+- Reproducibility level 1+2: two-hash manifest (`sha256_source` informative,
+  `sha256_text` binding, versioned extractor), frozen text normalization,
+  content-addressed `corpus-store/`, network-free verified ingest, per-run
+  `manifest.lock`, redacted resolved `config.json`, `verify` drift report

parallax_bench-0.0.1/CITATION.cff

@@ -0,0 +1,16 @@
+cff-version: 1.2.0
+message: "If you use parallax-bench, please cite it as below."
+title: "parallax-bench: Measuring language-induced retrieval displacement in RAG systems using parallel corpora"
+type: software
+authors:
+  - family-names: "Brůha"
+    given-names: "Ondřej"
+    email: "ondrej.bruha@alphacodes.eu"
+repository-code: "https://github.com/ondrejbruha/parallax-bench"
+license: Apache-2.0
+keywords:
+  - retrieval-augmented generation
+  - multilingual retrieval
+  - benchmark
+  - parallel corpora
+  - Czech

parallax_bench-0.0.1/LICENSE

@@ -0,0 +1,202 @@
+
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for reasonable and customary use in describing the
+      origin of the Work and reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer failure or malfunction, or any and all
+      other commercial damages or losses), even if such Contributor
+      has been advised of the possibility of such damages.
+
+   9. Accepting Warranty or Additional Liability. While redistributing
+      the Work or Derivative Works thereof, You may choose to offer,
+      and charge a fee for, acceptance of support, warranty, indemnity,
+      or other liability obligations and/or rights consistent with this
+      License. However, in accepting such obligations, You may act only
+      on Your own behalf and on Your sole responsibility, not on behalf
+      of any other Contributor, and only if You agree to indemnify,
+      defend, and hold each Contributor harmless for any liability
+      incurred by, or claims asserted against, such Contributor by reason
+      of your accepting any such warranty or additional liability.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright [yyyy] [name of copyright owner]
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+   See the License for the specific language governing permissions and
+   limitations under the License.

parallax_bench-0.0.1/PKG-INFO

@@ -0,0 +1,144 @@
+Metadata-Version: 2.5
+Name: parallax-bench
+Version: 0.0.1
+Summary: Measuring language-induced retrieval displacement in RAG systems using parallel corpora
+Project-URL: Homepage, https://github.com/ondrejbruha/parallax-bench
+Project-URL: Repository, https://github.com/ondrejbruha/parallax-bench
+Author-email: Ondřej Brůha <ondrej.bruha@alphacodes.eu>
+License-Expression: Apache-2.0
+License-File: LICENSE
+Keywords: benchmark,czech,evaluation,multilingual,rag,retrieval
+Classifier: Development Status :: 3 - Alpha
+Classifier: Intended Audience :: Science/Research
+Classifier: Programming Language :: Python :: 3.12
+Classifier: Programming Language :: Python :: 3.13
+Classifier: Programming Language :: Python :: 3.14
+Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
+Requires-Python: >=3.12
+Requires-Dist: httpx[http2]>=0.27
+Requires-Dist: pandas>=2.2
+Requires-Dist: pydantic>=2.7
+Requires-Dist: scipy>=1.13
+Requires-Dist: sqlalchemy>=2.0
+Requires-Dist: tenacity>=8.3
+Requires-Dist: typer>=0.12
+Provides-Extra: baseline
+Requires-Dist: sentence-transformers; extra == 'baseline'
+Provides-Extra: dev
+Requires-Dist: mypy; extra == 'dev'
+Requires-Dist: pytest; extra == 'dev'
+Requires-Dist: pytrec-eval-terrier; extra == 'dev'
+Requires-Dist: ruff; extra == 'dev'
+Provides-Extra: web
+Requires-Dist: fastapi; extra == 'web'
+Requires-Dist: uvicorn; extra == 'web'
+Description-Content-Type: text/markdown
+
+# parallax-bench
+
+*Measuring language-induced retrieval displacement in RAG systems using
+parallel corpora.*
+
+Parallax is the effect where **the same object appears at a different position
+when viewed from a different vantage point**. This benchmark measures that
+**the same document lands at a different rank when you ask about it in a
+different language** — with everything else (content, models, configuration,
+hardware) held constant.
+
+The PARALLAX benchmark uses parallel EU legislative corpora: the same document
+exists in many languages under a stable CELEX identifier. Ingesting each
+language into its own index and running every query variant against every
+index yields a fully controlled X×X matrix:
+
+- **diagonal** — monolingual RAG (query language = document language)
+- **off-diagonal** — cross-lingual retrieval
+- **all indexes at once** — MultiRAG
+
+Because ground truth is at *document* level and the document keeps the same ID
+across languages, **a single TREC-format qrels file is valid for every cell of
+the matrix** — monolingual and cross-lingual runs are scored by identical code
+over identical judgements.
+
+## Quickstart (two minutes, offline)
+
+```bash
+pip install parallax-bench
+parallax-bench run --system baseline-local --subset smoke
+parallax-bench score
+parallax-bench report
+```
+
+No GPU, no API key, no server, no Postgres — the `smoke` subset ships with
+document texts and the local baseline (BM25, optionally + dense via
+`pip install 'parallax-bench[baseline]'`) runs anywhere. CI runs exactly these
+commands on every PR.
+
+## Benchmarking a real system
+
+```bash
+pip install parallax-bench
+parallax-bench fetch    --data-version v1        # download ONCE into corpus-store/, verify sha256_text
+parallax-bench ingest   --system my-system       # reads only the frozen snapshot — never the network
+parallax-bench run      --system my-system --phase retrieval
+parallax-bench run      --system my-system --phase generation
+parallax-bench score    --run <run_id>
+parallax-bench report   --run <run_id>
+parallax-bench verify   --data-version v1        # measure source drift; fixes nothing
+```
+
+Systems are defined in `systems.toml` (see `systems.example.toml`) as an
+adapter class plus configuration. Ablations are expressed as two configured
+instances of the same adapter — never as extra protocol parameters. The
+adapter protocol has exactly four methods (`describe`, `index`, `search`,
+`generate`); see [docs/adding-a-system.md](docs/adding-a-system.md).
+
+`run` is resumable: the task queue lives in a database (SQLite by default,
+any SQLAlchemy URL — e.g. Heroku Postgres — via `--db` or `$DATABASE_URL`).
+Interrupt at any point; `run --resume <id>` picks up where it stopped.
+`score` is separate from `run`, so metrics can be recomputed under different
+definitions without repeating a forty-hour collection.
+
+## Results & reproducibility
+
+Aggregated metrics for our runs live in [`runs/`](runs/) as small, diffable
+CSVs; raw rankings are attached to GitHub releases. Every run directory
+carries full provenance: `config.json` (resolved configuration, secrets
+visibly redacted), `system.json` (read from the running system, never
+assumed), and `manifest.lock` (the binding `sha256_text` of every document
+that went into the index).
+
+Results are **verifiable**: every published number carries the full
+provenance needed to check it (`runs/<id>/`). The corpus snapshot used for
+these runs is **not yet publicly archived**, so third-party re-execution on
+the identical documents is not possible at this time. The manifest records
+`sha256_text` for every document, so an archive published later can be
+verified against the numbers already reported. Rebuilding the corpus from
+EUR-Lex is best-effort; `parallax-bench verify` measures source drift
+rather than assuming its absence.
+
+To add your own system's results, open a PR adding a `runs/<id>/` directory —
+CI validates the schema. There is no submission leaderboard by design.
+
+## Dataset
+
+`benchmark/v1/` contains queries, TREC qrels and a corpus **manifest** —
+`(celex, lang) → URL + sha256`. Document texts are **not redistributed**
+(licensing, size, and trust: it is immediately visible that nobody touched
+the documents). `parallax-bench fetch` downloads them from EUR-Lex and
+verifies checksums; if a source document changes, the run fails loudly.
+
+The single exception is `benchmark/smoke/` (10 documents × 3 languages,
+including texts) which exists so the quickstart works offline.
+
+## Licensing
+
+Dual-licensed on purpose: **code** (`src/`, `tests/`, `benchmark/build/`)
+under [Apache-2.0](LICENSE); **queries + qrels** (`benchmark/*/`) under
+[CC BY 4.0](benchmark/LICENSE-DATA). Corpus documents are not redistributed;
+EUR-Lex content is reusable under Decision 2011/833/EU with attribution.
+
+## Citation
+
+See [CITATION.cff](CITATION.cff). Code versions follow semver; dataset
+versions (`v1`, `v2`, …) are independent, each with its own Zenodo DOI, and a
+released dataset version is never modified in place.

parallax_bench-0.0.1/README.md

@@ -0,0 +1,108 @@
+# parallax-bench
+
+*Measuring language-induced retrieval displacement in RAG systems using
+parallel corpora.*
+
+Parallax is the effect where **the same object appears at a different position
+when viewed from a different vantage point**. This benchmark measures that
+**the same document lands at a different rank when you ask about it in a
+different language** — with everything else (content, models, configuration,
+hardware) held constant.
+
+The PARALLAX benchmark uses parallel EU legislative corpora: the same document
+exists in many languages under a stable CELEX identifier. Ingesting each
+language into its own index and running every query variant against every
+index yields a fully controlled X×X matrix:
+
+- **diagonal** — monolingual RAG (query language = document language)
+- **off-diagonal** — cross-lingual retrieval
+- **all indexes at once** — MultiRAG
+
+Because ground truth is at *document* level and the document keeps the same ID
+across languages, **a single TREC-format qrels file is valid for every cell of
+the matrix** — monolingual and cross-lingual runs are scored by identical code
+over identical judgements.
+
+## Quickstart (two minutes, offline)
+
+```bash
+pip install parallax-bench
+parallax-bench run --system baseline-local --subset smoke
+parallax-bench score
+parallax-bench report
+```
+
+No GPU, no API key, no server, no Postgres — the `smoke` subset ships with
+document texts and the local baseline (BM25, optionally + dense via
+`pip install 'parallax-bench[baseline]'`) runs anywhere. CI runs exactly these
+commands on every PR.
+
+## Benchmarking a real system
+
+```bash
+pip install parallax-bench
+parallax-bench fetch    --data-version v1        # download ONCE into corpus-store/, verify sha256_text
+parallax-bench ingest   --system my-system       # reads only the frozen snapshot — never the network
+parallax-bench run      --system my-system --phase retrieval
+parallax-bench run      --system my-system --phase generation
+parallax-bench score    --run <run_id>
+parallax-bench report   --run <run_id>
+parallax-bench verify   --data-version v1        # measure source drift; fixes nothing
+```
+
+Systems are defined in `systems.toml` (see `systems.example.toml`) as an
+adapter class plus configuration. Ablations are expressed as two configured
+instances of the same adapter — never as extra protocol parameters. The
+adapter protocol has exactly four methods (`describe`, `index`, `search`,
+`generate`); see [docs/adding-a-system.md](docs/adding-a-system.md).
+
+`run` is resumable: the task queue lives in a database (SQLite by default,
+any SQLAlchemy URL — e.g. Heroku Postgres — via `--db` or `$DATABASE_URL`).
+Interrupt at any point; `run --resume <id>` picks up where it stopped.
+`score` is separate from `run`, so metrics can be recomputed under different
+definitions without repeating a forty-hour collection.
+
+## Results & reproducibility
+
+Aggregated metrics for our runs live in [`runs/`](runs/) as small, diffable
+CSVs; raw rankings are attached to GitHub releases. Every run directory
+carries full provenance: `config.json` (resolved configuration, secrets
+visibly redacted), `system.json` (read from the running system, never
+assumed), and `manifest.lock` (the binding `sha256_text` of every document
+that went into the index).
+
+Results are **verifiable**: every published number carries the full
+provenance needed to check it (`runs/<id>/`). The corpus snapshot used for
+these runs is **not yet publicly archived**, so third-party re-execution on
+the identical documents is not possible at this time. The manifest records
+`sha256_text` for every document, so an archive published later can be
+verified against the numbers already reported. Rebuilding the corpus from
+EUR-Lex is best-effort; `parallax-bench verify` measures source drift
+rather than assuming its absence.
+
+To add your own system's results, open a PR adding a `runs/<id>/` directory —
+CI validates the schema. There is no submission leaderboard by design.
+
+## Dataset
+
+`benchmark/v1/` contains queries, TREC qrels and a corpus **manifest** —
+`(celex, lang) → URL + sha256`. Document texts are **not redistributed**
+(licensing, size, and trust: it is immediately visible that nobody touched
+the documents). `parallax-bench fetch` downloads them from EUR-Lex and
+verifies checksums; if a source document changes, the run fails loudly.
+
+The single exception is `benchmark/smoke/` (10 documents × 3 languages,
+including texts) which exists so the quickstart works offline.
+
+## Licensing
+
+Dual-licensed on purpose: **code** (`src/`, `tests/`, `benchmark/build/`)
+under [Apache-2.0](LICENSE); **queries + qrels** (`benchmark/*/`) under
+[CC BY 4.0](benchmark/LICENSE-DATA). Corpus documents are not redistributed;
+EUR-Lex content is reusable under Decision 2011/833/EU with attribution.
+
+## Citation
+
+See [CITATION.cff](CITATION.cff). Code versions follow semver; dataset
+versions (`v1`, `v2`, …) are independent, each with its own Zenodo DOI, and a
+released dataset version is never modified in place.

parallax_bench-0.0.1/benchmark/DATASHEET.md

@@ -0,0 +1,57 @@
+# Datasheet — the PARALLAX benchmark dataset
+
+Following the structure of *Datasheets for Datasets* (Gebru et al., 2021),
+abbreviated to what applies.
+
+## Motivation
+
+**Why was the dataset created?** To measure language-induced retrieval
+displacement in RAG systems: how the rank of the *same document* changes when
+the *same question* is asked in a different language, everything else held
+constant. No existing benchmark isolates the language variable this way, and
+Czech is heavily underrepresented in RAG evaluation resources.
+
+**Who created it?** Ondřej Brůha, as part of a master's thesis on evaluating
+multilingual RAG systems using parallel legislative corpora.
+
+## Composition
+
+- **Queries** (`queries.jsonl`) — parallel language variants of the same
+  underlying question, bound by `query_group`. Each targets exactly one
+  source document.
+- **Relevance judgements** (`qrels.txt`) — TREC format, *document*
+  granularity. Because documents keep the same CELEX id across languages, a
+  single qrels file is valid for every query-language × index-language cell.
+- **Corpus manifest** (`manifest.jsonl`) — `(celex, lang) → URL + sha256 +
+  length`. **Document texts are not part of the dataset** (see below).
+- **Smoke subset** (`smoke/`) — 10 documents × 3 languages (cs, en, de)
+  *including texts*, 20 hand-authored query groups. Exists so the quickstart
+  runs offline; not intended for drawing conclusions.
+
+## Collection & generation
+
+Corpus documents come from EUR-Lex (parallel EU legislation, stable CELEX
+identifiers). Query generation is documented verbatim — prompt, model,
+date — in each version's `generation.md`. Where an LLM generated or
+translated queries, this is disclosed there; primary sets are
+pivot-generated (English) and translated (`origin: "translated"`), with a
+native-generation validation sample (`origin: "native"`).
+
+## Distribution & licensing
+
+Queries and qrels: **CC BY 4.0** (`LICENSE-DATA`). Corpus texts are **not
+redistributed** — three reasons: upstream dataset licences vary; size; and
+trust (it is immediately verifiable that documents are untouched — `fetch`
+re-downloads from the source and checks sha256). EUR-Lex content itself is
+reusable under Commission Decision 2011/833/EU with source attribution.
+
+## Known limitations
+
+See `docs/limitations.md` — translationese in translated queries,
+document-level (not chunk-level) ground truth, legal domain as primary,
+automatically derived relevance without full human verification.
+
+## Maintenance
+
+Dataset versions (`v1`, `v2`, …) are immutable once released; each gets its
+own DOI via the GitHub–Zenodo integration. Corrections produce a new version.