22.5 Folder Hygiene: Raw vs Clean vs Outputs

By Ravindra Bagale Last updated: 23 October 2026

Why. Overwriting data.csv with a cleaned version destroys the audit trail. Festival spikes get blamed on “missing rows” that you dropped. Folders save friendships.

What. A simple layout:

project/
  data/
    raw/           # never edit by hand
    clean/         # outputs of cleaning notebook
  notebooks/
  figures/         # PNG only
  outputs/         # CSV / XLSX for humans and systems
  README.md        # one paragraph on grain + sources

How.

Hygiene rules

Copy vendor files into data/raw/ with the original name + date.
Cleaning notebook reads only from raw/, writes only to clean/.
Analysis notebook reads clean/, writes outputs/ and figures/.
Never double-click-edit a CSV in Excel and save over raw/ (Excel can break encoding).
Put random_state and filter rules in the notebook top cell.

Worked example A — Paths in code.

from pathlib import Path

ROOT = Path(".")
RAW = ROOT / "data" / "raw" / "blinkit_maha_22-10-2026.csv"
CLEAN = ROOT / "data" / "clean" / "orders_clean.csv"
OUT = ROOT / "outputs"
FIG = ROOT / "figures"
for d in [RAW.parent, CLEAN.parent, OUT, FIG]:
    d.mkdir(parents=True, exist_ok=True)

Worked example B — Naming.

Bad	Better
`final.csv`	`orders_clean_2026-10-22.csv`
`chart.png`	`pune_delivered_gmv_2026-10-22.png`
`newnew2.xlsx`	`city_status_2026-10-22.xlsx`

Worked example C — README stub.

# Blinkit Maharashtra lab (fictional)
Grain: one row = one order.
Cities: Pune, Solapur, Nashik, Sambhaji Nagar, Kolhapur, Nagpur.
Currency: INR. Dates: dd-mm-yyyy. Timezone: IST.
Raw files are immutable. Clean files are rebuilt from notebooks.

Salman clones the folder months later and still understands it.