coding-standards.md

Coding Standards

These standards are lightweight and repo-specific. They are meant to help keep training code, utilities, and Android code aligned.

General

• Prefer small, explicit command surfaces over hidden behavior.
• Keep filenames, manifest formats, and output paths stable unless there is a clear migration plan.
• Update docs when changing user-facing commands, paths, or asset names.

Python

• Keep CLI scripts argument-driven and runnable from the repo root.
• Prefer readable orchestration over clever metaprogramming in data and training workflows.
• Preserve CSV schema compatibility when possible.
• Fail early on missing files, invalid manifests, or dependency gaps.

Android / Kotlin

• Keep processing behavior explicit and configuration-driven.
• Preserve CPU fallback behavior when touching ModelManager.
• Treat budget-device memory limits as a first-class constraint.
• Keep WorkManager payload keys and repository mapping logic in sync.

Data and experiment hygiene

• Avoid mixing curated datasets with staged raw corpora by accident.
• Keep benchmark outputs clearly named and grouped.
• Use manifests instead of hand-assembled pair lists where possible.

Documentation

• Prefer ASCII unless a file already intentionally uses Unicode.
• Remove stale examples instead of leaving placeholders that no longer run.
• Link docs to code that exists now, not code that used to exist.