Reference

API reference

Mostly Right 2.0 groups methods by weather, markets, economy, and finance. Domain reads return source and timing fields. A small training-table helper covers a few predefined examples.

Python

import mostlyright as mr

mr.__all__
# ['ContractError', 'LeakageError', 'MostlyrightError', 'NoDataError',
#  '__version__', 'discover', 'economy', 'finance', 'markets', 'weather']

The four domains appear when their distributions are installed. A full pip install mostlyrightmd installs all four.

Namespace	Stable starting points	What it covers
`mr.weather`	`observations()`, `climate()`, `forecasts()`, `satellite()`, `live.`, `cwop.`, `nwp.*`, `training_table()`	Weather observations, forecasts, satellite pixels, citizen stations, labels, and live reports
`mr.economy`	`series()`, `snapshot()`, `releases()`, `training_table()`	Economic indicators and first-print vintages
`mr.markets.kalshi`	`settlement_days()`, `parse_ticker()`, `training_table()`	Kalshi weather-market labels and resolution rules
`mr.markets.polymarket`	`settlement_days()`, `discover()`, `settle()`, `training_table()`	Polymarket discovery and settlement
`mr.markets.economy_trades`	`candles()`	Kalshi economy-market price history
`mr.finance.transcripts`	`resolve()`, `derive()`, transcript schemas, ledgers, and parsers	Earnings-call data and mention resolution
`mr.discover()`	Cross-domain catalog search	Stations, schemas, and available data

Raw observations and live reports

Use observations() for archived per-report rows:

reports = weather.observations("KNYC", from_date, to_date)

Use weather.live.latest() for the newest Python report:

import asyncio

now = asyncio.run(weather.live.latest("KNYC"))

TypeScript exports the live method directly from weather:

const now = await weather.latest("KNYC");

obs() was removed in 2.0. There is no compatibility alias.

Experimental composition

Custom join functions live under mostlyright.experimental, not at the package root:

from mostlyright.experimental import align, spine

target = spine(
    target_frame,
    entity="station",
    decision_time="decision_time_utc",
    y="label",
)
table = align(target, source_a, source_b)

align(), spine(), DeferredSource, and the contributor registry are documented and usable, but semver-exempt until they graduate.

Training-table example

weather.training_table() is a small predefined recipe used by the weather quickstart. It returns one row per station and local-standard day, with the official outcome beside observation features and cutoff fields.

from datetime import date
from mostlyright import weather

example = weather.training_table(
    "KNYC",
    date(2025, 1, 6),
    date(2025, 1, 12),
)

The default label="cli" attaches NWS CLI daily-summary labels. Use label="daily_extremes" for the international daily-extremes target.

The TypeScript meta package exposes the same example in camelCase:

import { trainingTable } from "mostlyright";

const { rows: example, provenance } = await trainingTable(
  "KNYC",
  "2025-01-06",
  "2025-01-12",
);

Python returns a DataFrame. TypeScript returns a DataResult with rows and provenance. Serialized column names stay in snake_case across both SDKs. Use the direct domain reads when this recipe does not fit.

2.0 naming changes

Removed in 2.0	Use now
`obs()`	`observations()`
`pairs()`, `research()`, `dataset()`	`training_table()`
`research_economy()`, `research_econ()`	`economy.training_table()`
`economy.history()`	`economy.series()`
`econ` namespace	`economy`
TypeScript `dataset()`, `research()`, `pairs()`	`trainingTable()`

Complete API reference

This page is the short map. The generated references document every published module, function, option, return type, and exception: