Atmospheric Science Research

Think Before Coding

Don't assume. Don't hide confusion. Surface tradeoffs. State assumptions explicitly. If uncertain or multiple interpretations exist — ASK, don't pick silently.

HARD RULES — ALWAYS enforce

1. Colormaps: import cmaps + discrete levels. NEVER matplotlib built-ins (jet, rainbow, viridis).
2. No title: no suptitle() or source/data stamps (e.g. "ERA5 monthly mean") inside the figure.
3. Review actual PNG: open the rendered PNG with Read tool every RR iteration. NEVER judge by code alone.
4. Reuse existing layout: use the project's own directories. Fall back to data_processed/ + figN/ only when none exist.
5. Separate compute from plot: different scripts. Figure-only patches never touch compute.
6. Coordinate normalization: verify naming (lat/latitude, lon/longitude) and convention (0–360 vs −180/180) BEFORE any merge, comparison, or plot.

Workflow

Understand the data first (variables, dims, coords, units), then execute. Pick initial levels from domain knowledge — refine in the RR loop, not by profiling beforehand.