Moving flat Python scripts into a package: anchor data paths to the repo root, check for already-running servers
Refactoring a flat repo (config.py, helpers.py, steps/, models/ with sys.path hacks and bare import config) into a pipeline/ package: (1) rewrite imports to absolute package form (from pipeline import config, from pipeline.models.llm import ...) with a small regex pass, and delete the sys.path.insert(0, dirname(__file__)/..) lines; keep one shared module object for mutable config (always import it the same way, or runtime config.X = ... changes won't be seen). (2) Paths computed from __file__ move with the file: DATA_DIR = join(dirname(__file__), 'data') silently starts pointing inside the package, so define ROOT = dirname(dirname(abspath(__file__))) and anchor output/.env/dataset paths to it. (3) Step scripts with if __name__ == '__main__' still work as python -m pkg.steps.step2 <arg> from the root. (4) A grep filter like grep -v 'pipeline/steps/' meant to hide already-updated mentions also hides every match from files whose path contains that prefix; filter on content, not on lines. (5) Before renaming an output folder, check for a long-running server process using it (pgrep -f app.py, ps -o lstart), since it keeps the old code and paths in memory until restarted; test with a second instance on another port instead of killing the user's.