Copycat

An evidence-led Singapore copyright triage workspace. Compare two works, record the legal context, inspect deterministic matches and export a reviewable report.

What changed in v2

• Thirteen evidence-backed questions cover subsistence, qualification, term, standing, restricted acts, territory, permission, copying, protected expression, substantial taking, independent creation, fair use and other exceptions.
• Missing facts stay unknown. Similarity never establishes access, ownership, copying or substantiality. Legal results describe the supplied assessments.
• The fair-use worksheet records all four s 191 factors and s 192 acknowledgment conditions. Other claims and historical dates are explicitly routed for review.
• Text matching handles short documents, Unicode tokens, repeated text and small excerpts without inventing overlap. DOCX tables are included. Unreadable PDFs fail.
• Image and visual-only video comparisons report their methods and limits. Missing dependencies disable those options instead of returning false zero scores.
• Case-scoped access keys protect files, jobs, previews and exports. Case inputs cannot change during analysis; edits invalidate old reports. Expiry and deletion cover originals, derivatives, database evidence and PDFs.
• A responsive, keyboard-accessible workflow supports review, reruns, session recovery, image previews, passage locations, PDF and JSON exports.

Run locally

Use Python 3.11 and Node 22 or newer. Run one local API process (without reload for persistent work). From the repository root:

# Windows / PowerShell
py -3.11 -m venv backend/.venv
backend/.venv/Scripts/python -m pip install --upgrade pip setuptools
backend/.venv/Scripts/python -m pip install -r backend/requirements.txt -r backend/requirements-image.txt
cd backend
.venv/Scripts/python -m uvicorn app.main:app --port 8000

In another terminal:

cd frontend
npm ci
npm run dev

Open localhost:3000. API documentation is at localhost:8000/docs. On Linux/macOS, use python3.11 -m venv backend/.venv and .venv/bin/python.

Video requires FFmpeg and ffprobe on PATH, or the COPYCAT_FFMPEG_BIN and COPYCAT_FFPROBE_BIN settings. The UI reads server capabilities. It does not use Whisper, OCR or an external AI service. Text comparison is lexical, not semantic or cross-language matching. See configuration.

Containers

docker compose up --build

The compose configuration binds the UI and API to localhost, uses a persistent named volume and includes FFmpeg/image dependencies. Production UI builds embed the API URL; change that build argument and CORS origins together for another host. See the operations runbook before exposing the service.

Workflow

1. Select matching media types, upload original/comparison files and identify the work category, alleged conduct date and claim route.
2. Record yes/no/unknown legal assessments. Yes/no needs evidence and reasoning. Unknown is valid for an initial comparison.
3. Run analysis. The case page resumes polling after reload in the same tab.
4. Inspect legal gaps and matches. Revise context and rerun as evidence develops.
5. Download PDF/JSON before expiry. Save the access key if you need to reopen in another tab. Anyone with that key can read or delete the case.

The retention default is 24 hours from case creation. Access expires at that time. Physical cleanup runs every five minutes while the service is active, after running work finishes. Operators must separately govern backups.

Legal foundation and limits

The Singapore framework explains the legal limbs, primary sources, version findings and verification limits. Core provisions include Copyright Act 2021 ss 39, 109–116, 123–125, 133–146, 152–159 and 190–194. The rulepack corrects the baseline's misidentified RecordTV citation.

This product records human assessments of evaluative legal requirements. It does not decide disputed facts, balance fair use automatically, certify infringement, or assess criminal liability/remedies. Authorisation, secondary infringement, other work categories and conduct outside the reviewed consolidation period need separate review. Statutory source bytes and version markers were checked; the automated provision/quotation extraction package is incomplete and is not claimed as verified. Foreign-country regulations and historical amendment impacts are case-specific review items.

Reproducibility

Reports include role-bound file hashes, full intake, component scores, evidence, dependency versions, rulepack content hash and scoring version. The report SHA-256 excludes case ID and generation time and includes the substantive inputs/results. PDF bytes include presentation metadata; the JSON fingerprint identifies the assessment. See scoring.

Verify

cd backend
python -m pytest -q
cd ../frontend
npm ci
npm run build
npx playwright install chromium
npm test
npm audit

Activate the backend venv or use its Python executable. Playwright starts a real API and UI on ports 8000/3000; stop other copies first. Browser tests exercise intake, analysis, evidence, revision, exports, access isolation, deletion, mobile layout and axe accessibility checks. GitHub Actions runs backend, frontend and browser checks. Validation and remaining release conditions are recorded in the improvement audit.