CoolFace
Datasetpublic

openthomas/weather-forecasts

OpenThomas weather forecasts Every forecast the OpenThomas weather desk read or produced, as-of. Pushed daily from the trading box, so each commit is a timestamped record of what was known when — the leak-free substrate behind the paper's replay experiments (docs/EXPERIMENTS.md in the harness repository). file what lines / bytes data/nwp-guidance-and-settlements.jsonl as-of daily extremes from seven public NWP models (Open-Meteo previous-runs API) at leads 0-5, plus… See the full description on the dataset page: https://huggingface.co/datasets/openthomas/weather-forecasts.

sourceHugging Facemitupdated 6h agoView on Hugging Face
0likes347downloads
README.md145 linesDownload Raw Back to root
1---2license: mit3tags:4- weather5- numerical-weather-prediction6- prediction-markets7- forecasting8- graphcast9- gencast10- build-in-public11pretty_name: OpenThomas weather forecasts12---13 14# OpenThomas weather forecasts15 16Every forecast the [OpenThomas](https://openthomas.com) weather desk read or17produced, as-of. Pushed daily from the trading box, so each commit is a18timestamped record of what was known when — the leak-free substrate behind19the paper's replay experiments (`docs/EXPERIMENTS.md` in the20[harness repository](https://github.com/PredictionMarketTrader/openthomas)).21 22| file | what | lines / bytes |23|---|---|---|24| `data/nwp-guidance-and-settlements.jsonl` | as-of daily extremes from seven public NWP models (Open-Meteo previous-runs API) at leads 0-5, plus official settlements (ACIS / NWS CLI) | 25,462 |25| `data/own-models.jsonl` | station-day extremes from OpenThomas' own GraphCast / Pangu inference runs, with issue timestamps | 3,500 |26| `data/gencast-spread.json` | per station-day ensemble spread of the latest GenCast run (sigma multiplier input) | 3,293 |27| `data/graphcast-tempseries.json` | coarse global 2m-temperature daily series from the latest GraphCast run | 112,918 |28| `data/baseline-nbm.jsonl` | as-of NBM daily max/min per station (IEM MOS archive), the official-guidance baseline | 2,140 |29| `data/baseline-gfsmos.jsonl` | as-of GFS MOS daily max/min per station (IEM MOS archive) | 2,140 |30 31## Frozen replay datasets and experiment results32 33`replay/` holds the frozen decision-time rows (one JSONL per window, digest34in the filename) and the experiment results scored on them. A row carries35the settlement station and day, the baseline probability at the snapshot,36the Kalshi bid/ask at that snapshot, and the outcome — never anything the37agent could not have known at decision time.38 39- `replay/e3-high9.txt`40- `replay/e3-low1.txt`41- `replay/e3-low2.txt`42- `replay/e3e4.log`43- `replay/e3e4.sh`44- `replay/e4.log`45- `replay/e4.sh`46- `replay/e4.txt`47- `replay/lineage.json`48- `replay/results-2026-07-09-to-2026-09-11.json`49- `replay/results-2026-07-09-to-2026-09-11.md`50- `replay/results-2026-07-09-to-2026-09-12.json`51- `replay/results-2026-07-09-to-2026-09-12.md`52- `replay/results-2026-07-09-to-2026-09-13.json`53- `replay/results-2026-07-09-to-2026-09-13.md`54- `replay/results-2026-07-09-to-2026-09-14.json`55- `replay/results-2026-07-09-to-2026-09-14.md`56- `replay/results-2026-07-09-to-2026-09-15.json`57- `replay/results-2026-07-09-to-2026-09-15.md`58- `replay/results-2026-07-09-to-2026-09-16.json`59- `replay/results-2026-07-09-to-2026-09-16.md`60- `replay/results-2026-07-09-to-2026-09-17.json`61- `replay/results-2026-07-09-to-2026-09-17.md`62- `replay/results-2026-07-09-to-2026-09-18.json`63- `replay/results-2026-07-09-to-2026-09-18.md`64- `replay/results-2026-07-09-to-2026-09-19.json`65- `replay/results-2026-07-09-to-2026-09-19.md`66- `replay/results-2026-07-09-to-2026-09-20.json`67- `replay/results-2026-07-09-to-2026-09-20.md`68- `replay/results-2026-07-09-to-2026-09-21.json`69- `replay/results-2026-07-09-to-2026-09-21.md`70- `replay/results-2026-07-09-to-2026-09-22.json`71- `replay/results-2026-07-09-to-2026-09-22.md`72- `replay/results-2026-07-09-to-2026-09-23.json`73- `replay/results-2026-07-09-to-2026-09-23.md`74- `replay/results-2026-07-09-to-2026-09-24.json`75- `replay/results-2026-07-09-to-2026-09-24.md`76- `replay/results-2026-07-09-to-2026-09-25.json`77- `replay/results-2026-07-09-to-2026-09-25.md`78- `replay/results-llm-2026-07-09-to-2026-09-11.json`79- `replay/results-llm-2026-07-09-to-2026-09-11.md`80- `replay/results-llm-2026-07-09-to-2026-09-12.json`81- `replay/results-llm-2026-07-09-to-2026-09-12.md`82- `replay/results-llm-2026-07-09-to-2026-09-19.json`83- `replay/results-llm-2026-07-09-to-2026-09-19.md`84- `replay/rows-2026-07-09-to-2026-09-11.jsonl`85- `replay/rows-2026-07-09-to-2026-09-12.jsonl`86- `replay/rows-2026-07-09-to-2026-09-13.jsonl`87- `replay/rows-2026-07-09-to-2026-09-14.jsonl`88- `replay/rows-2026-07-09-to-2026-09-15.jsonl`89- `replay/rows-2026-07-09-to-2026-09-16.jsonl`90- `replay/rows-2026-07-09-to-2026-09-17.jsonl`91- `replay/rows-2026-07-09-to-2026-09-18.jsonl`92- `replay/rows-2026-07-09-to-2026-09-19.jsonl`93- `replay/rows-2026-07-09-to-2026-09-20.jsonl`94- `replay/rows-2026-07-09-to-2026-09-21.jsonl`95- `replay/rows-2026-07-09-to-2026-09-22.jsonl`96- `replay/rows-2026-07-09-to-2026-09-23.jsonl`97- `replay/rows-2026-07-09-to-2026-09-24.jsonl`98- `replay/rows-2026-07-09-to-2026-09-25.jsonl`99- `replay/rows-gfsmos-2026-07-09-to-2026-09-11.jsonl`100- `replay/rows-gfsmos-2026-07-09-to-2026-09-12.jsonl`101- `replay/rows-gfsmos-2026-07-09-to-2026-09-13.jsonl`102- `replay/rows-gfsmos-2026-07-09-to-2026-09-14.jsonl`103- `replay/rows-gfsmos-2026-07-09-to-2026-09-15.jsonl`104- `replay/rows-gfsmos-2026-07-09-to-2026-09-16.jsonl`105- `replay/rows-gfsmos-2026-07-09-to-2026-09-17.jsonl`106- `replay/rows-gfsmos-2026-07-09-to-2026-09-18.jsonl`107- `replay/rows-gfsmos-2026-07-09-to-2026-09-19.jsonl`108- `replay/rows-gfsmos-2026-07-09-to-2026-09-20.jsonl`109- `replay/rows-gfsmos-2026-07-09-to-2026-09-21.jsonl`110- `replay/rows-gfsmos-2026-07-09-to-2026-09-22.jsonl`111- `replay/rows-gfsmos-2026-07-09-to-2026-09-23.jsonl`112- `replay/rows-gfsmos-2026-07-09-to-2026-09-24.jsonl`113- `replay/rows-gfsmos-2026-07-09-to-2026-09-25.jsonl`114- `replay/rows-high9-2026-07-09-to-2026-09-11.jsonl`115- `replay/rows-low1-2026-07-09-to-2026-09-11.jsonl`116- `replay/rows-low2-2026-07-09-to-2026-09-11.jsonl`117- `replay/rows-nbm-2026-07-09-to-2026-09-11.jsonl`118- `replay/rows-nbm-2026-07-09-to-2026-09-12.jsonl`119- `replay/rows-nbm-2026-07-09-to-2026-09-13.jsonl`120- `replay/rows-nbm-2026-07-09-to-2026-09-14.jsonl`121- `replay/rows-nbm-2026-07-09-to-2026-09-15.jsonl`122- `replay/rows-nbm-2026-07-09-to-2026-09-16.jsonl`123- `replay/rows-nbm-2026-07-09-to-2026-09-17.jsonl`124- `replay/rows-nbm-2026-07-09-to-2026-09-18.jsonl`125- `replay/rows-nbm-2026-07-09-to-2026-09-19.jsonl`126- `replay/rows-nbm-2026-07-09-to-2026-09-20.jsonl`127- `replay/rows-nbm-2026-07-09-to-2026-09-21.jsonl`128- `replay/rows-nbm-2026-07-09-to-2026-09-22.jsonl`129- `replay/rows-nbm-2026-07-09-to-2026-09-23.jsonl`130- `replay/rows-nbm-2026-07-09-to-2026-09-24.jsonl`131- `replay/rows-nbm-2026-07-09-to-2026-09-25.jsonl`132 133## Conventions134 135- Temperatures are °F; settlements are the integer daily extreme from the136  NWS Climatological Report (via ACIS), the number the markets settle on.137- `lead` is days ahead: lead 1 for target day *d* is what the model published138  on *d − 1*.139- NBM / GFS MOS baselines use the *d − 1* 12Z run, read from the IEM MOS140  archive as issued.141- Own-model rows carry `issued_at`; the run that produced them is identified142  by that timestamp and the model name.143 144*Paper trading. Nothing here is financial advice.*145