Kalshi Weather Tracker

Tracks NWS hourly forecasts vs. NCEI observed highs for Kalshi high-temperature markets.

Repo structure

stations.csv                  # add new locations here
R/
  pull_forecasts.R            # NWS hourly forecast capture
  pull_actuals.R              # NCEI CDO official daily highs
data/
  forecasts_KHOU.csv          # one file per station
  actuals_KHOU.csv
.github/workflows/
  pull_forecasts.yml          # runs every 2 hours
  pull_actuals.yml            # runs daily + supports manual backfill

Setup

1. Fork / create this repo on GitHub

2. Add your NCEI token as a GitHub Secret

Go to Settings → Secrets and variables → Actions → New repository secret:

• Name: NCEI_TOKEN
• Value: your token from https://www.ncei.noaa.gov/cdo-web/token

3. Backfill historical actuals

Trigger the Pull Actuals workflow manually from the GitHub Actions tab:

• Set start_date to 2025-01-01
• Leave end_date blank (defaults to yesterday)

This will populate data/actuals_KHOU.csv with every official daily high since Jan 1, 2025.

4. Enable the scheduled workflows

GitHub requires at least one push to the default branch before scheduled workflows activate. Push a small change (e.g., edit this README) to trigger them.

---

Adding a new station

Add a row to stations.csv:

station_id,name,lat,lon,ncei_ghcnd_id,timezone
KHOU,Houston Hobby,29.6454,-95.2789,USW00012918,America/Chicago
KDFW,Dallas Fort Worth,32.8998,-97.0403,USW00003927,America/Chicago
KORD,Chicago O'Hare,41.9742,-87.9073,USW00094846,America/Chicago

The NCEI GHCND station ID can be found by searching https://www.ncei.noaa.gov/access/past-weather/ — look for the station name and use the ID shown (strip the GHCND: prefix).

On the next workflow run, new CSVs will be created automatically for any new stations.

---

Output format

data/forecasts_KHOU.csv

Column	Description
station_id	NWS station ID
station_name	Human-readable name
forecast_issued_utc	When this forecast was captured
forecast_for_date	The date being forecast
forecast_high_f	Predicted high (°F) — max of all hourly temps that day
n_hours_in_forecast	How many hourly periods were available for that date

data/actuals_KHOU.csv

Column	Description
station_id	Station ID
station_name	Human-readable name
date	Observation date
observed_high_f	Official TMAX from NCEI GHCND (°F)

---

Notes

• Forecast lag: NWS forecasts for "tomorrow" are typically stable by mid-afternoon. The 2-hour cadence captures how the forecast evolves through the day prior.
• Actuals lag: NCEI CDO data typically appears 1–4 days after the observation date. The daily workflow will keep re-checking until data appears.
• NCEI 503 errors: The CDO API is occasionally slow or down. The script retries once automatically; failures are non-fatal and the next daily run will catch up.