KnouandClaude Opus 5
867a07cecb
Close gaps found in a review of the documentation: - Track data/metrics/ and data/metric_sources.json in git so the data checker passes on a fresh clone (finding 23; decision logged) - State the Alaska and Hawaii coverage gap in the README limitations and extend finding 12 - File findings 24-26: the data-sources doc lacks gridMET and several pipeline commands; wettest/driest month are computed twice; solar GHI is written by the extreme-temperature apply step - Add the stale "fallback values" note to filter 2's tasks Tidy the document system: - Add docs/reviews/README.md with the numbering rules and a finding index - Rename koppen-mixed-display-plan.md to koppen-mixed-display.md and fix its stale Puerto Rico and "stage 5" text - Add the precipitation-month step to the README enrichment list - Describe the Current method / Previous method pattern in plan section 5 - Add CLAUDE.md with the project guardrails and doc layout Format filter-calculations.md so it renders on GitHub and in VS Code: inline math uses $...$, ranges use en dashes, and implementation references name functions instead of line numbers. Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
2.2 KiB
2.2 KiB
Project notes for Claude
US County Climate Explorer: a static Leaflet app (index.html, app.js,
styles.css) over data/climate-data.csv, built by the Python scripts in
scripts/. The 12 filters are being reviewed one at a time before the CSV is
restructured; see docs/pipeline-plan.md.
Guardrails
- Never rerun
scripts/build_county_climate_data.pyoverdata/climate-data.csv. It drops live columns and overwriteskoppenZone(pipeline plan §7). - Change one filter at a time. After every apply step, run the checker and compare changed counties against expectations.
- Run scripts from the project root with the virtual environment:
.venv\Scripts\python.exe scripts\check_climate_data.py.venv\Scripts\python.exe -m unittest discover -s tests
- When
app.jsor the data files change, bumpAPP_ASSET_VERSIONinapp.js(it also versions the CSV and GeoJSON URLs) and the matchingapp.js?v=query inindex.html.styles.csshas its own query inindex.html. - The map and every filter cover the 50 states and DC. Puerto Rico rows stay in the data files but are never shown.
- Do not start the Phase 3 CSV restructure until all 12 filters are reviewed.
Writing
- Write "Köppen" with the umlaut in prose, UI text, help text, and messages.
Use ASCII
koppenonly in identifiers, file names, and data paths.
Documentation layout
Each fact has one home:
docs/pipeline-plan.md: current pipeline, target design, phases, the per-filter checklist (§5), the filter tracker (§6), and guardrails (§7). Keep it about 200 lines.docs/reviews/: one file per filter with its findings, decision links, and tasks;00-cross-filter.mdfor findings that affect several filters.docs/reviews/README.mdhas the numbering conventions and the finding index.docs/decisions.md: open decisions and a dated log of decided ones. An item leaves "Open decisions" only when the project owner decides it.docs/filter-calculations.md: what each calculation is. No review status.scripts/common/README.md: rules for shared helpers.
Keep the tracker status in the plan and the review-file checkboxes in sync.