Actual prototype screen with the fictional volunteer example loaded. CSV review runs locally in the browser.

Inspiration

A volunteer coordinator needs to share a shift list, but the CSV also contains email addresses, phone numbers and notes the recipient may not need. ShareCheck adds a deliberate review step before that file leaves the browser.

What it does

Load the fictional example or a UTF-8 CSV. Review every column, including those without flags. Choose entire columns to omit, inspect the proposed copy, and confirm before downloading. Undo restores a column without changing the original file.

The reviewed CSV quotes every field and prefixes formula-like text with an apostrophe. An optional JSON decision record stores the source SHA-256 fingerprint, column positions and counts, without original header names or cell values. Processing stays local in the browser. No account, API key, upload or AI service is needed at runtime.

How it was built

Created independently on October 30, 2026 during the event. JavaScript, HTML, CSS and browser-native Web Crypto support a strict CSV parser, deterministic pattern checks, reversible omission controls and separate exports. All example data is fictional. No commercial source or source from prior hackathon entries was incorporated.

OpenAI assistance contributed to testing under designer directions. The independent source is MIT licensed; system fonts and browser APIs supply the interface.

Challenges and testing

Review state must never survive a failed or changed import. Confirmation is invalidated when selections change, malformed input clears stale results, and pending file reads cannot repopulate a cleared session. The parser handles quoted multiline values, escaped quotes, BOM, Unicode and common line endings.

The core passed 39 tests. An independent review added 14 adversarial checks and 1,000 round-trip cases. Actual browser checks covered the synthetic workflow and downloaded outputs. These results cover specified cases, not every browser or dataset.

Scope and lessons

Simple English-header, email and long-number patterns help direct attention, but can miss names, free text, non-English headers and identifying combinations. Every column still needs human review. This is not an anonymizer, compliance assessor or complete personal-data detector. Spreadsheet escaping visibly changes some text and is not a universal security guarantee. The original download remains unchanged and unsanitized.

The decision fingerprint can link a record to a known source; share that record thoughtfully. The lesson was to keep the original immutable, make omissions reversible, and make human review explicit rather than treating a green indicator as permission to share.

Try it

Open index.html in a modern browser supporting Web Crypto, or serve the downloaded folder locally with python3 -m http.server 8000. Use only the supplied fictional example for the demonstration. Limits: 250,000 input bytes, 1,000 data rows and 50 columns.