paperguard 2.0.1__tar.gz

paperguard-2.0.1/.env.example

@@ -0,0 +1,13 @@
+# Email for "polite pool" of free APIs (OpenAlex, CrossRef, Unpaywall)
+# Replace with your own email - this is REQUIRED for higher rate limits
+PAPERGUARD_EMAIL=your.email@example.com
+
+# Optional: NCBI API key for higher PubMed rate limits
+# Get free at https://www.ncbi.nlm.nih.gov/account/
+NCBI_API_KEY=
+
+# Cache directory
+PAPERGUARD_CACHE_DIR=~/.paperguard/cache
+
+# Default RNG seed for reproducibility
+PAPERGUARD_SEED=42

paperguard-2.0.1/CHANGELOG.md

@@ -0,0 +1,350 @@
+# Changelog
+
+All notable changes to PaperGuard are documented in this file. Format follows
+[Keep a Changelog](https://keepachangelog.com/) and the project adheres to
+[Semantic Versioning](https://semver.org/).
+
+## [2.0.1] — 2026-05-19 — macOS arm64 CI fix
+
+### Fixed
+- **`greenlet>=3.0`** added explicitly to the `[webui]` and `[dev]`
+  extras. SQLAlchemy 2.x needs `greenlet` for its async-to-sync
+  bridge; it is normally a transitive dep on Linux and Windows, but
+  the macOS arm64 wheel marks it optional in some versions, which
+  caused all 20 multi-tenant tests to error on the macos-latest
+  CI matrix. Local `pip install -e .[dev]` on macOS now pulls it in.
+- No source-code changes; only the dependency declaration moved.
+
+## [2.0.0] — 2026-05-19 — Paper-mill graph + Carlisle automation + multi-tenant Web UI
+
+Second stable major. Folds in the two dev-1 shipped items (M1 paper-mill
+citation-graph signatures; deeper Carlisle automation including multi-arm
+RCT support and PDF→C1 trial-ID auto-extraction) and adds an opt-in
+multi-tenant Web UI surface. 223 tests passing, mypy strict clean,
+30 built-in detectors plus plugin entry-point.
+
+### Added (multi-tenant Web UI)
+- **`/app/*` invite-only multi-tenant surface** mounted on top of the
+  existing anonymous `/scan` endpoints. Activated by setting
+  `PAPERGUARD_DB_URL` or `PAPERGUARD_MULTITENANT=1`; otherwise no behaviour
+  changes from 1.x.
+- **User + InviteCode + Project + ScanReport ORM** (SQLAlchemy 2.0 async).
+  SQLite by default; any async engine (PostgreSQL via `asyncpg`, MySQL via
+  `aiomysql`) via `PAPERGUARD_DB_URL`.
+- **Per-report visibility**: `private` / `org` / `public`. Public reports
+  are listed at `/app/shared` and readable anonymously.
+- **Admin invite flow**: admins mint single-use codes at
+  `/app/admin/invites`; invitees redeem at `/app/redeem/{code}` with a
+  password ≥ 10 characters.
+- **Bootstrap admin from env**: `PAPERGUARD_ADMIN_EMAIL` +
+  `PAPERGUARD_ADMIN_PASSWORD` create the first admin idempotently on
+  startup.
+- **Sessions** in HttpOnly, SameSite=Lax cookies signed with
+  `itsdangerous` (key from `PAPERGUARD_SECRET_KEY`); 14-day TTL.
+- **Passwords** hashed with `bcrypt` directly (no passlib; bcrypt 4.x/5.x
+  compatible).
+- New optional dependencies in `paperguard[webui]`: `sqlalchemy>=2.0`,
+  `aiosqlite>=0.20`, `bcrypt>=4.0`, `itsdangerous>=2.2`.
+- **24 new tests** covering bootstrap, auth, invite redemption (single-use,
+  weak-password rejection, email binding), project isolation, visibility
+  enforcement (private / org / public), legacy `/scan` survival,
+  multi-tenant off-by-default, and session signing.
+- New documentation: `docs/webui_multitenant.md` with architecture,
+  env-var reference, invite flow, visibility semantics, and production
+  checklist.
+
+### Added (Carlisle automation deepening)
+- **C1 multi-arm RCT support**: `BaselineVariable.arms` accepts any number
+  of `(n, mean, sd)` arms. Pairwise Welch t between every arm pair;
+  per-variable median pairwise p folded into the Stouffer combination.
+  Backward-compatible (legacy `n1/mean1/sd1 + n2/mean2/sd2` still works).
+- **Robust baseline-table parser**: handles 3+ arm tables, header-embedded
+  N (`Treatment (n=42)`), categorical `n (%)` rows recorded separately,
+  multiple table caption formats (`baseline`, `demographics`, `study
+  population`, `participant characteristics`, `table 1/i`).
+- **Auto-extract trial registration IDs** from PDF/docx text
+  (`extractor/trial_ids.py`): supports NCT, ISRCTN, ChiCTR, ACTRN,
+  EudraCT, DRKS. CLI surfaces found IDs in scan output.
+- 14 new tests for multi-arm Welch, trial-ID extraction, arm-column
+  identification, mean±SD / paren / categorical row parsing.
+
+### Added (M1 paper-mill citation graph)
+- **M1 — Paper-Mill Citation Graph Signatures** detector
+  (`detectors/m1_paper_mill_graph.py`).
+- **Citation-graph fetcher** (`fetcher/citation_graph.py`).
+- **`--check-paper-mill` flag** on `paperguard scan`.
+- 5 unit tests against synthetic citation subgraphs.
+- New core dependency: `networkx>=3.0`.
+
+### Planned for 2.x (still open)
+- Full Cabanac 2025 PDCN model on a 5M-node citation graph (the M1
+  detector is the local-subgraph version of the same signatures)
+- Reviewer-fraud signal extraction (no public data source yet)
+- ML-trained Western-blot specific image model (requires labeled corpus)
+- Multi-tenant Web UI with shared scan history
+
+## [1.0.0] — 2026-05-18 — First stable release
+
+### Added (since 0.9.0)
+- **PDF→C1 auto baseline extraction** (`extractor/baseline_tables.py`)
+  — CLI now automatically scans PDFs for baseline characteristics tables
+  and feeds them to C1 Carlisle.
+- **Expanded T4 dictionary** — tortured phrases went from ~50 to **150+**
+  curated entries covering CS/ML, statistics, optimization, biomedical,
+  energy, common terms, and GPT-disguise phrases.
+- **`paperguard server`** — production-mode daemon with X-API-Token auth,
+  multi-worker, `/health` endpoint.
+- **`docs/detectors/`** — 29 deep-dive markdown pages auto-generated from
+  source (one per detector + index README).
+- **`scripts/generate_detector_docs.py`** — regenerate detector docs from
+  source metadata.
+- **GitHub Actions enhancements**: codecov upload, dependabot config,
+  release-please workflow.
+- **Coverage badge** (73% on tracked source files).
+- **`README.zh.md`** finalized.
+
+### Changed
+- Status promoted from Alpha to Beta in PyPI classifiers (the project
+  is now stable enough for first-real-users).
+
+### Decided NOT to do in 1.0
+- **Extras split** (`paperguard[image]` / `paperguard[text]` / etc.) —
+  current dependencies are all small enough that the additional
+  packaging complexity doesn't pay for itself. May revisit in 2.0
+  if dependency tree grows.
+
+### Post-1.0 polish
+
+- **Fetcher disk-cache** (`src/paperguard/fetcher/cache.py`) — OpenAlex
+  / CrossRef / Unpaywall responses cached to `diskcache` for 7 days,
+  reducing API load on repeat scans.
+- **`paperguard list-detectors`** — table / json / ids output, optional
+  `--cluster` filter.
+- **`paperguard fetch-rw`** — downloads the Retraction Watch CSV.
+- **`paperguard fetch-ori`** — writes a starter ORI sanctions template.
+- **`docker-compose.yml`** — production reference deployment.
+- **`MANIFEST.in`** — ensures docs and fixtures ship in the sdist.
+- **`examples/plugin_example/`** — fully installable plugin template
+  with entry-point wiring.
+- **`paperguard.reporter.schema`** — emits a JSON Schema describing the
+  audit-report shape.
+- **Coverage** 73% → **74%** with 16 new boost tests (180 total).
+
+## [0.9.0] — 2026-05-18 — Polish + 4 specialized detectors
+
+### Added detectors (4 new, 25 → 29)
+- **A7 Last-Digit 0/5 Preference (Geng method)** — Binomial test on
+  P(末位 ∈ {0,5}) ≠ 0.2; specialized refinement of A1 χ², bidirectional
+  (catches both excess and depression). Direct internalization of the
+  2025 Geng Hongwei method.
+- **B8 SPRITE plausibility** — Heathers et al. (2018) iterative
+  reconstruction: given (mean, SD, N, scale_min, scale_max), tries to
+  construct any valid integer sample; failure → SUSPICIOUS.
+- **F5 EXIF Cross-Image Clustering** — multi-image consistency: span
+  > 5 years, > 2 distinct camera models, or identical second-precision
+  timestamps across ≥ 3 images.
+- **T6 AI-Generated Text Heuristic** — two-layer detection:
+  uncleaned LLM response leakage (CRITICAL) + AI-overused phrase
+  density (Kobak 2025 word patterns).
+
+### Added fetchers
+- **`fetcher/openalex.py`** — new `get_author_retraction_rate(author_id)`
+  using OpenAlex `is_retracted` flag (Retraction Watch–synced upstream).
+- **`fetcher/pubmed.py`** — Biopython `Entrez` wrapper for PMID lookup
+  and DOI→PMID resolution. Saves us re-implementing E-utils.
+
+### Added documentation
+- **`README.zh.md`** — 中文 README full translation.
+- `docs/fraud_case_studies.md` (from 0.8.0) — cross-referenced.
+
+### Tests
+- **`tests/test_golden.py`** — anti-regression gate: golden findings
+  count on the paired fixtures.
+- 12 new tests for A7/B8/F5/T6.
+- 164 total passing.
+
+### Reused open-source components
+- **biopython** for NCBI Entrez (vs hand-rolled E-utils HTTP).
+- imagehash, pdfplumber, pymupdf, opencv-python-headless, PIL, piexif:
+  all already in use. No new heavyweight dependencies in 0.9.0.
+
+## [0.8.0] — 2026-05-18 — Real-world fraud case internalization
+
+### Added (6 new detectors derived from real-case study)
+- **A6 Implausible Values** — column-name-aware range checks + sentinel
+  values (999 / -999) detection. Internalizes Wansink's "700 pizza slices".
+- **B7 P-Curve** (Simonsohn 2014) — p-curve shape analysis; left-skew or
+  near-α pile-up signals p-hacking. Internalizes Wansink's email-leaked
+  data-mining patterns.
+- **D1 Residual Smoothness** — block-variance stability check;
+  internalizes the Stapel-case signature of "too clean" data.
+- **D2 Missing-Data Pattern** — flags 0-missing datasets with low
+  column-σ variation; internalizes Carlisle's RCT-fraud observation.
+- **F4 Cross-Paper Image Duplication** — persistent SQLite pHash store
+  for cross-paper image reuse. Internalizes Masliah (2024) and Hwang
+  (2005) findings.
+- **T5 Stylometry** — Markowitz-Hancock 2014 PLOS ONE linguistic
+  fingerprint (methodology / certainty / adjective density ratios).
+  Internalizes the Stapel-text findings.
+
+### Documentation
+- **`docs/fraud_case_studies.md`** — Stapel, Fujii, Hwang, Schön,
+  Macchiarini, Wansink, Masliah, Geng-targets, Bik 2016: each case maps
+  to specific detectors with honest "would catch" vs "cannot catch"
+  assessment.
+
+### Detector count
+- Built-in: 25 (was 19). Added A6, B7, D1, D2, F4, T5.
+
+### Tests
+- 152 passing (was 138). Added 13 new tests for the 6 new detectors.
+
+## [0.7.0] — 2026-05-18
+
+### Added
+- **B5 TIVA** (Schimmack 2014) — z-score variance test on a set of independent
+  study p-values; insufficient variance → potential p-hacking / selective
+  reporting.
+- **B6 GRIMMER** (Anaya 2016; Allard 2018) — `(mean, SD, N)` triple-consistency
+  test. Stricter than B1 GRIM (which only checks mean × N).
+- **T4 Tortured Phrases** (Cabanac 2021) — 50+ machine-translation
+  fingerprint phrases ("profound neural organization" → "deep neural network").
+  Detects paper-mill / synonym-laundered text.
+- **B4 statcheck upgrades** — added **Q-test** for meta-analysis heterogeneity;
+  whole-text one-tailed scan (if "one-tailed/one-sided/单尾" anywhere in the
+  manuscript, switch matching t/r/z to one-tailed when that gives consistency).
+- **`paperguard selfcheck`** — runs internal fixtures through all detectors as
+  a sanity check on installation.
+- **`paperguard explain`** — LLM-explanation of a specific finding from a JSON
+  report (needs `PAPERGUARD_LLM_PROVIDER`).
+- **`paperguard diff before.json after.json`** — track changes between two
+  scan reports.
+- **Auto-OA-PDF download** — `paperguard scan --doi X` (without `-f`) now
+  attempts to download the OA PDF via Unpaywall.
+- **docs/** — `detectors.md` (per-detector reference) and
+  `epistemic_position.md` (vocabulary rule, innocent-explanation rule).
+- **examples/04_full_pipeline_demo.py** — exercise every detector class.
+
+### Detector count
+- Built-in: 19 (was 16). Added B5, B6, T4.
+
+### Tests
+- 138 passing (was 120). Added: T4 ×5, B5 ×4, B6 ×4, CLI extras ×5.
+
+## [0.6.0] — 2026-05-18
+
+### Added
+- **T3 — Data Availability + Ethics Audit** detector. Flags missing data
+  statements, vague "available on request" without verifiable accessions,
+  missing IRB/IACUC, missing trial registration (NCT/ISRCTN/ChiCTR/EudraCT),
+  missing competing-interests disclosure.
+- **F3 — Splice / Copy-Move Forensics** detector. Patch-level statistical
+  signatures (mean / std / Laplacian variance) with translation-vote
+  consistency to find pixel-level cloning that ORB-based F2 misses.
+- **CLI auto-runs T3** on extracted PDF/docx text.
+
+### Fixed
+- **PDF image extraction** now filters out tiny embedded bitmaps (math
+  symbols, font glyphs) by size (≥ 200×200 px, ≥ 8 KB) and SHA-256-dedups,
+  eliminating massive F1 false-positive cascades on typeset PDFs.
+- **G4 publisher-creator whitelist** — Springer / Elsevier / Wiley / LaTeX /
+  pdfTeX / Acrobat Distiller / Word / LibreOffice etc. no longer trigger
+  the "creator not in authors list" CONCERN. This was a 100% false positive
+  on every published PDF.
+
+### Tested on
+- 2 real Nature Communications papers (ecology, OA) — both correctly
+  classified as PASS with 0 findings.
+
+## [0.5.0] — 2026-05-18
+
+### Added
+- **F2 — Bik-style internal image duplication** detector. ORB keypoint
+  self-matching + RANSAC affine consensus to find copy-pasted patches inside
+  a single image. Rotation/scale tolerant.
+- **T1 — Text similarity** detector. 5-gram word-shingling + Jaccard against
+  a user-supplied corpus (no network). For self-plagiarism and re-use.
+- **T2 — Clinical-trial outcome consistency** detector. Compares paper's
+  reported primary outcomes to ClinicalTrials.gov v2 API registration.
+  Catches outcome switching (Goldacre 2019).
+- **ORI sanctions** local CSV lookup (`paperguard.fetcher.ori_sanctions`).
+- **LLM explainer** (opt-in via `PAPERGUARD_LLM_PROVIDER`) — supports
+  OpenAI / Anthropic / Ollama. Hard-coded system prompt forbids the LLM
+  from claiming fraud or inventing evidence.
+- **Statcheck one-tailed support** — recognizes "one-tailed / 单尾" in the
+  reporting context.
+- **i18n: es, ja, de** language packs (now 5 total).
+- **WCAG 2.1 AA** for HTML reports: focus-visible outlines, ARIA roles,
+  semantic header/main/footer, `prefers-reduced-motion`, severity colors
+  re-tuned for ≥ 4.5:1 contrast on white.
+- New dependency: `opencv-python-headless` (for F2).
+- 21 new tests (133 total).
+
+## [0.4.0] — 2026-05-18
+
+### Added
+- **Plugin system** — third-party packages can register detectors via
+  `paperguard.detectors` entry-point group. `DetectorRegistry.load_plugins()`
+  discovers and instantiates them with safe error handling.
+- **i18n** — report framework now supports `en` and `zh-CN` via a lightweight
+  dict-backed `t()` helper (no gettext / .po toolchain). `--lang` flag added
+  to `scan`. `PAPERGUARD_LANG` env var also honored.
+- **Web UI** (`paperguard webui`) — FastAPI app with upload form, language
+  selector, `/detectors` introspection endpoint, and `/scan.json` for
+  programmatic use. Available via the `paperguard[webui]` extras.
+- 14 new tests (91 total): i18n, plugin loader (mocked entry points), and
+  webui (TestClient).
+
+### Changed
+- `paperguard` package now exposes `__version__`.
+- `register_default(load_plugins=True)` is the new default; pass `False`
+  to opt-out (used in tests).
+
+## [0.3.0] — 2026-05-18
+
+### Added
+- **C1 Carlisle** baseline-imbalance detector for RCTs (Welch t per variable + Stouffer combination).
+- **F1 image-duplication** detector via perceptual hash (`imagehash` library).
+- **G1 image EXIF temporal forensics** — flags shooting time before claimed experiment start, after submission, and Photoshop signatures.
+- **G3 docx rsid forensics** — identifies python-docx / pandoc-generated files via missing or homogeneous `w:rsid` values.
+- **Image extractors** for .docx (word/media/) and .pdf (via pymupdf).
+- **Retraction Watch CSV loader** — local lookup against the official dataset (no network).
+- Tests: 19 new (67 total).
+
+## [0.2.0] — 2026-05-18
+
+### Added
+- **A2 Benford** first-digit detector with applicability gate (≥ 2 decades of range).
+- **B4 statcheck** — recompute reported `t / F / χ² / r / z` p-values from
+  manuscript text, flagging decision-reversals as SUSPICIOUS and numeric
+  inconsistencies as CONCERN.
+- **PubPeer** client — surfaces existing public comments on a DOI.
+- **PDF text and table extraction** via pymupdf + pdfplumber.
+- **Docx inline-number classification** — extracts and classifies numbers from
+  prose (p-values, percentages, mean ± SD, generic decimals).
+- **HTML report export** (`--output-html`) — self-contained styled HTML.
+- **Batch mode** (`paperguard batch --glob 'papers/*.pdf'`) for many files at once.
+- **Dockerfile** for containerized usage.
+- Roadmap, contributing guide, security policy, GitHub Actions CI.
+
+### Changed
+- `clean-meta` subcommand removed from PaperGuard; moved to a separate
+  standalone tool to keep this project narrowly scoped to detection.
+- `scan` now auto-handles .xlsx, .csv, .tsv, .docx, .pdf — both table data and
+  free-text run through detectors as appropriate.
+
+### Removed
+- `src/paperguard/utils/docx_meta_writer.py` and the `clean-meta` CLI command.
+  Cleanup tooling lives in a separate private repo to avoid coupling
+  detection with anti-detection in one shipped product.
+
+## [0.1.0] — 2026-05-18
+
+### Added
+- Initial MVP release with five detectors: A1 (terminal digit), A3 (inter-column
+  arithmetic), A5 (decimal consistency), B1 (GRIM), G4 (file metadata forensics).
+- Click-based CLI with `scan` and `search` subcommands.
+- Rich terminal report + JSON export + immutable audit log.
+- OpenAlex, CrossRef, and Unpaywall clients (`scan --doi` integration).
+- BH–FDR p-value correction and severity escalation (PASS → CRITICAL).
+- 22 tests, full `mypy --strict` + `ruff` clean.

paperguard-2.0.1/CITATION.cff

@@ -0,0 +1,52 @@
+cff-version: 1.2.0
+message: "If you use PaperGuard in your research, please cite it."
+title: PaperGuard
+abstract: >
+  Statistical anomaly screener for tabular research data. Flags anomalies,
+  not fraud. Detectors cover terminal-digit, Benford, inter-column
+  arithmetic, decimal consistency, GRIM, statcheck, Carlisle baseline
+  balance, image duplication (pHash), EXIF temporal forensics, docx rsid
+  forensics, and file metadata forensics.
+authors:
+  - name: PaperGuard Contributors
+version: 2.0.0.dev1
+date-released: 2026-05-18
+license: MIT
+keywords:
+  - research-integrity
+  - scientific-misconduct
+  - statistical-forensics
+  - data-fabrication
+  - peer-review
+repository-code: https://example.com/paperguard
+references:
+  - type: article
+    authors:
+      - family-names: Mosimann
+        given-names: J. E.
+    title: "Data fabrication: Can people generate random digits?"
+    journal: Accountability in Research
+    year: 1995
+  - type: article
+    authors:
+      - family-names: Brown
+        given-names: N. J. L.
+      - family-names: Heathers
+        given-names: J. A. J.
+    title: "The GRIM Test: A simple technique detects numerous anomalies in the reporting of results in psychology"
+    journal: Social Psychological and Personality Science
+    year: 2017
+  - type: article
+    authors:
+      - family-names: Nuijten
+        given-names: M. B.
+    title: "The prevalence of statistical reporting errors in psychology (1985-2013)"
+    journal: Behavior Research Methods
+    year: 2016
+  - type: article
+    authors:
+      - family-names: Carlisle
+        given-names: J. B.
+    title: "Data fabrication and other reasons for non-random sampling in 5087 randomised controlled trials"
+    journal: Anaesthesia
+    year: 2017

paperguard-2.0.1/CONTRIBUTING.md

@@ -0,0 +1,69 @@
+# Contributing to PaperGuard
+
+Thanks for considering a contribution. Most useful contributions are new
+detectors. The codebase makes adding one straightforward.
+
+## Setup
+
+```bash
+git clone <repo>
+cd PaperGuard
+python -m venv .venv
+# Linux/macOS: source .venv/bin/activate
+# Windows:     .\.venv\Scripts\Activate.ps1
+pip install -e ".[dev]" types-openpyxl
+pre-commit install   # optional
+```
+
+Run the validation suite before any PR:
+
+```bash
+pytest -m "not network" -v
+ruff check src/ tests/
+mypy src/
+```
+
+## Adding a new detector
+
+Use `src/paperguard/detectors/a1_terminal_digit.py` as the canonical template.
+Every detector must:
+
+1. Subclass `BaseDetector`.
+2. Define class-level `id`, `name`, `description`, `academic_basis`,
+   `data_requirements`, `assumption_cluster` (all as `ClassVar`).
+3. Implement `check_applicability(data) -> tuple[bool, str]` — return
+   `(False, reason)` rather than raising when data doesn't fit.
+4. Implement `_detect(data, seed) -> list[Finding]`.
+5. Register the detector in
+   `src/paperguard/core/registry.py:DetectorRegistry.register_default()`.
+
+Each `Finding` must include at least three `innocent_explanations`. This is
+non-negotiable — the epistemic posture of the tool depends on it.
+
+## Tests
+
+For each detector, add a test file under `tests/test_detectors/`. At minimum:
+
+- One test that confirms the detector flags `fabricated_data` at
+  `CONCERN` or higher.
+- One test that confirms it does not flag `genuine_data` at `SUSPICIOUS+`.
+- One test for inapplicability (e.g., wrong data type or too-small N).
+
+Tests requiring network calls must be marked with `@pytest.mark.network` so
+CI can skip them.
+
+## Code style
+
+- Python ≥ 3.11. Use `X | Y` not `Union[X, Y]`; `list[X]` not `List[X]`.
+- Type hint every public function. We run `mypy --strict`.
+- Use pydantic v2 for data validation; `click` for CLI; `rich` for terminal.
+- Avoid hard-coding secrets. New API clients should read credentials via
+  `pydantic-settings` in `config.py`.
+
+## Epistemic rules (not negotiable)
+
+The tool never outputs the words "fraud", "造假", "misconduct",
+"cheating" in any report. Use "anomaly", "statistical inconsistency",
+"unexplained pattern". Every report ends with the standard disclaimer.
+
+A finding without `innocent_explanations` is a bug.

paperguard-2.0.1/LICENSE

@@ -0,0 +1,21 @@
+MIT License
+
+Copyright (c) 2026 PaperGuard Contributors
+
+Permission is hereby granted, free of charge, to any person obtaining a copy
+of this software and associated documentation files (the "Software"), to deal
+in the Software without restriction, including without limitation the rights
+to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+copies of the Software, and to permit persons to whom the Software is
+furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice shall be included in all
+copies or substantial portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+SOFTWARE.

paperguard-2.0.1/MANIFEST.in

@@ -0,0 +1,20 @@
+include LICENSE
+include README.md
+include README.zh.md
+include CHANGELOG.md
+include CITATION.cff
+include CONTRIBUTING.md
+include SECURITY.md
+include ROADMAP.md
+include pyproject.toml
+include .env.example
+
+recursive-include docs *.md
+recursive-include src/paperguard *.py
+recursive-include tests/fixtures *.csv *.py
+
+global-exclude __pycache__
+global-exclude *.py[cod]
+global-exclude *.so
+global-exclude .DS_Store
+global-exclude *.bak