Section: Configuration & Environment
Problem requirements.txt currently lists every dependency with no version pin at all (fastapi, torch, transformers, peft, etc. — zero == constraints anywhere in the file).
Why this matters
For a project whose core claim is reproducible evaluation, shipping an unpinned dependency file means a fresh pip install -r requirements.txt next month could silently change model-loading, tokenization, or numeric behavior underneath the reported FinQA results. This is a direct threat to the project's own reproducibility claims.
Acceptance Criteria
Files likely to modify
finverify-terminal/backend/requirements.txt
Skills required: Python, dependency management
Estimated difficulty: 1/5
Estimated effort: 2-3 hours
Labels: backend, reproducibility, good first issue
Milestone: v0.2 - Foundations & Testing
Dependencies: None
Section: Configuration & Environment
Problem requirements.txt currently lists every dependency with no version pin at all (fastapi, torch, transformers, peft, etc. — zero == constraints anywhere in the file).
Why this matters
For a project whose core claim is reproducible evaluation, shipping an unpinned dependency file means a fresh pip install -r requirements.txt next month could silently change model-loading, tokenization, or numeric behavior underneath the reported FinQA results. This is a direct threat to the project's own reproducibility claims.
Acceptance Criteria
Every dependency in requirements.txt is pinned to the exact version currently used in the working environment (pip freeze from a known-good venv, filtered to direct dependencies)
A requirements-dev.txt or extras section separates test/dev-only dependencies if any are added
README/CONTRIBUTING note that dependency upgrades should be deliberate PRs, not silent drift
Fresh pip install -r requirements.txt in a clean venv followed by python test_dvl.py still passes
Files likely to modify
finverify-terminal/backend/requirements.txt
Skills required: Python, dependency management
Estimated difficulty: 1/5
Estimated effort: 2-3 hours
Labels: backend, reproducibility, good first issue
Milestone: v0.2 - Foundations & Testing
Dependencies: None