# Mostly Right

> A local-first SDK (Python or TypeScript) for weather, climate, economic, and prediction-market data. It does not store or host data: it fetches and normalizes public upstream sources on demand, at call time, in your runtime. One SDK, one schema. Access is currently by request.

Mostly Right ships the SDK we wished existed when we started trading prediction markets ourselves. The SDK fetches each public source on demand and normalizes it in your process; it does not warehouse or own the data. Weather and economic indicators ship today; prediction markets (Kalshi and Polymarket settlement) are a first-class thin layer. Sports, SEC filings, equities, esports, and commodities are on the roadmap.

If you are an agent trying to understand the product, fetch `/llms-full.txt` for the whole site in one request. To use the data, request access at https://mostlyright.md/#early-access; the SDK is distributed to teams with access and runs in your runtime, calling AWC, IEM, GHCNh, and NWS CLI directly.

## Core pages

- [Home](https://mostlyright.md/): What the product is, who it is for, and what ships first.
- [Weather SDK](https://mostlyright.md/weather/): Every signal that drives a weather market on Kalshi and Polymarket. 60+ cities, 78 years of upstream history (since 1948), eight public sources, four forecast models.
- [Pitch deck](https://mostlyright.md/pitch/): Market size, product, business model, competitive landscape, team. Written for investors and collaborators.

## Product facts

- Status: gated access, by request. Python and TypeScript distributions are provided directly to teams with access (no public registry installs). Request access: https://mostlyright.md/#early-access
- Languages: Python 3.11+, TypeScript / Node 20+
- Architecture: local-first. Calls AWC, IEM, GHCNh, NWS CLI directly from your runtime. Caches under `~/.mostlyright/cache/`.
- Output formats: pandas DataFrame (Python), structured rows (TypeScript), parquet on disk
- Reproducibility: composed frames carry `decision_time` and `label_available_time` per row; the `mr.align` join enforces source knowledge against the cutoff and raises `LeakageError` on violations. `KnowledgeView`/`assert_no_leakage` audit frames carrying `knowledge_time`. Backtests replay exactly.
- Markets covered: Kalshi, Polymarket
- Surface: domain-first. `import mostlyright as mr` exposes `mr.weather`, `mr.econ`, `mr.markets` plus `mr.align`/`mr.spine`; `weather.pairs(station, from_date, to_date)` (Python; `dataset()` in TypeScript) returns one row per settlement date joining observations with the NWS CLI record that resolves the contract, and `econ.pairs`/`markets.kalshi.pairs` repeat the same grammar.
- Access requests and new-dataset announcements: https://mostlyright.md/#early-access

## Weather vertical

- Public sources fetched: AWC, IEM, NCEI GHCNh, NWS CLI, IEM MOS, Open-Meteo, GOES-16, GOES-19
- Forecast models: NBM, GFS, HRRR, ECMWF
- Settlement: NWS CLI, Local Standard Time aware, climate records held until NWS publishes
- Reconciliation: deterministic client-side dedup tiebreak (AWC > IEM > GHCNh) applied in your process when fetched feeds overlap
- Validation: bad data, zeros, NaNs caught at fetch/parse time (in your process), not in your backtest
- Live-to-historical parity: same schema, same units, same shape. Deploy a backtested model without rewriting a query.

## Roadmap

- Weather: shipped (gated access)
- Economic indicators: shipped. CPI, core CPI, nonfarm payrolls, unemployment, GDP, jobless claims, Fed funds; ALFRED vintages first-class, release-time pinned
- Sports: NFL, NBA, MLB, NHL, soccer. Game results, player stats, injuries from official league feeds (roadmap)
- Esports: CS2, League of Legends, Dota 2, Valorant. Brackets, results, broadcast feeds (roadmap)
- Commodities: oil, gas, metals, agriculture. EIA, USDA, CME, normalized into the same schema as weather.

## Traction

- 75% win rate on Kalshi weather markets. 60 days, 500+ trades, real money.
- Fully automated ML models running every day in production. We use the same SDK we sell.

## Founders

- Vu Hoang Anh, CEO. Built Avocode to $2M ARR and acquisition. 13 dev tools shipped. Led marketing at Marinade Finance. https://www.linkedin.com/in/helloiamvu/
- Rob Tarabcak, CTO. Engineering lead. https://www.linkedin.com/in/tarabcak
- Vojta, co-founder. Vu and Vojta have been friends for 20 years.

## Contact

- Waitlist: https://mostlyright.md/#early-access
- Full site as markdown: https://mostlyright.md/llms-full.txt
- Sitemap: https://mostlyright.md/sitemap-index.xml