|
Download README.md from Hugdrigo/reviewflow-screening-lab: direct link, hf CLI and curl.
- Browser
- Download file 5.65 kB
-
https://huggingface.co/spaces/Hugdrigo/reviewflow-screening-lab/resolve/main/README.md
- Command line
-
hf download hf://spaces/Hugdrigo/reviewflow-screening-lab/README.md
-
curl -L -o README.md https://huggingface.co/spaces/Hugdrigo/reviewflow-screening-lab/resolve/main/README.md
5.65 kB
| title: "ReviewFlow | Systematic Review Screening Studio" | |
| emoji: "📚" | |
| colorFrom: green | |
| colorTo: blue | |
| sdk: static | |
| app_file: index.html | |
| pinned: false | |
| license: mit | |
| short_description: "Systematic review screening and duplicate audits" | |
| tags: | |
| - research | |
| - literature-review | |
| - systematic-review | |
| - evidence-synthesis | |
| - data-analysis | |
| - javascript | |
| # ReviewFlow: systematic literature screening and traceable decisions | |
| **Project 69 of the Hugdrigo research and decision-intelligence portfolio.** | |
| ReviewFlow is a privacy-conscious, browser-only title-and-abstract screening application for researchers organizing literature reviews. It supports an auditable, two-reviewer workflow with deterministic duplicate detection and exports for downstream study selection. It runs as a free **Static Space** on Hugging Face, with no backend, paid inference, keys, cookies or external JavaScript packages. | |
| > **Research integrity:** All three built-in demonstrations contain **fictional publications and abstracts**. ReviewFlow does **not** search scholarly databases, retrieve full texts, adjudicate study quality, perform meta-analysis or constitute a completed PRISMA 2020 process. Screening decisions require human review. | |
| ## Capabilities | |
| 1. **Research Overview:** imported, duplicate, unique, retained, excluded and unresolved records, reconciled automatically. | |
| 2. **Screening Studio:** editable decisions from Reviewer A and Reviewer B, explicit consensus decisions, exclusion reasons, text notes, search and paginated reference table. | |
| 3. **Duplicate Audit:** conservative matching by normalized DOI/identifier, or title plus publication year; original data is preserved. | |
| 4. **Reviewer Agreement:** 2-by-2 confusion table, raw agreement and Cohen's kappa on records with two definitive reviewer decisions. | |
| 5. **Screening Flow:** a PRISMA-inspired summary for the **title/abstract stage only**, with an exportable SVG diagram. | |
| 6. **Data & Audit Exports:** import CSV, download original records, decisions, exclusion reasons, full audit JSON, a Markdown methods report, diagram SVG, and a sample CSV template. | |
| ## How screening decisions work | |
| - Identical **include/include** and **exclude/exclude** independent ratings result in an effective decision when there is no manual consensus decision. | |
| - Any other pairing, including an `unsure` or `pending`, stays pending until a researcher explicitly sets `resolution` to `include` or `exclude`. | |
| - A consensus exclusion requires an explicit recorded reason. Original independent ratings remain intact when a resolution is entered. | |
| - A duplicate is retained in exports and flagged by `duplicateOf`, but excluded from the unique-screening denominator. | |
| - Retained records are **candidates for full-text assessment**, *not final included studies*. | |
| ## Data model | |
| Required import fields: `id`, `title`, `year`. | |
| Optional import fields: `authors`, `doi`, `source`, `abstract`, `reviewer_a`, `reviewer_b`, `resolution`, `reason`, `note`. | |
| `reviewer_a` and `reviewer_b`: `pending`, `include`, `exclude`, `unsure`. | |
| `resolution`: `pending`, `include`, `exclude`. Every manually excluded record must include a `reason`. | |
| - The CSV parser handles UTF-8, quoted cells, escaped quotes, embedded newlines and both CRLF/LF. | |
| - Imports are all-or-nothing after validation, to a maximum of 5,000 rows and 3 MB. | |
| - Every ID must be unique; years must be between 1800 and 2100. | |
| - CSV export escapes potentially spreadsheet-executable text. Avoid importing confidential records into publicly shared files. | |
| ## Reproducibility and formulas | |
| For a set of `n` unique records with two definite independent ratings, observed agreement is `Po = agreements / n`. Expected agreement uses the product of each reviewer's include and exclude proportions. Cohen's kappa is `(Po - Pe) / (1 - Pe)` and is returned as **not estimable** if the denominator is zero. Uncertain and incomplete ratings are excluded *only from the kappa denominator*, not from the screening flow counts. | |
| Duplicate matching uses a simplified normalization, so it may miss ambiguous duplicates and can occasionally flag distinct records that share an identifier. Review matching outcomes before using the results in a formal publication. | |
| All metrics are descriptive. ReviewFlow does not prove study relevance, bias or scientific validity. | |
| ## Privacy and persistence | |
| The application performs processing locally in the browser, and does not send imported data to a server. Data and edits remain in application memory, **not** permanent browser storage. Export your audit before reloading or switching demonstration studies. Switching away from an edited/imported study triggers a confirmation warning. | |
| ## Quick deployment on Hugging Face | |
| 1. Create a Space under `Hugdrigo` with name `reviewflow-screening-lab`. | |
| 2. Select **Manual setup**, **Static SDK**, **Blank**, **Public**, and **MIT**. | |
| 3. Place `README.md`, `index.html`, and `LICENSE` directly at the repository root. | |
| 4. Commit to `main`, open **App**, and verify the six modules. | |
| 5. If metadata validation blocks uploading, edit the **existing repository README.md** first. Specifically, `colorFrom` and `colorTo` must be recognized colors, and `short_description` must not exceed 60 characters. | |
| ## Validation | |
| The included technical package documents 53 unit tests and 92 browser checks on desktop/mobile via Chromium. Validation checks include deduplication, reconciliation, CSV validation and preservation on error, manual resolution, downloads, reset, and responsiveness. Tests are local and do not guarantee the Space has already been published. | |
| ## License | |
| MIT. Copyright (c) 2026 Hugdrigo. | |