mostlyrightmd-core 2.1.0

Python SDK core for quants, ML engineers, and AI agents — schema-versioned types, temporal safety, validation, alignment, and snapshot primitives. Domain adapters ship separately; `pip install mostlyrightmd` pulls the full SDK. Imports as `mostlyright`.

mostlyrightmd-core

The public-data SDK core for quants, ML pipelines, and AI agents.

mostlyrightmd-core is the Python core distribution: types, schemas, validators, temporal-safety primitives, align() / spine(), and snapshot helpers. Weather, markets, economy, and finance adapters ship as separate distributions; credentialed and hosted options belong to their domain packages rather than the core-only install.

Want the whole SDK in one command? Install the meta-distribution mostlyrightmd, which pulls the core plus every domain. Individual domains are mostlyrightmd-weather, mostlyrightmd-markets, mostlyrightmd-economy, and mostlyrightmd-finance.

Install

pip install mostlyrightmd-core           # core only (helpers + snapshot primitives)
pip install mostlyrightmd                # the FULL SDK (core + weather/markets/economy/finance)
pip install mostlyrightmd-weather        # weather data sources only (brings core transitively)

mostlyrightmd-core and mostlyrightmd-weather share the mostlyright.* Python namespace but ship as separate PyPI distributions, so users who only need the helpers can skip the heavier weather deps. weather.training_table() lazy-imports mostlyright.weather and raises a clear error if the weather package is not installed.

Point-in-time composition

The core package provides the lower-level align() and spine() helpers used to compose targets and sources against per-row cutoffs. They live under mostlyright.experimental: they are documented and usable, but semver-exempt until they graduate.

Use the domain packages for source reads such as weather.observations() and economy.series(). See docs/dataset.md for a complete composition example.

Training-table example

If you install the weather package, training_table() provides one small, predefined supervised-learning recipe:

from datetime import date
from mostlyright import weather

table = weather.training_table("KNYC", date(2025, 1, 6), date(2025, 1, 12))
print(table.columns.tolist())

Use it when that recipe matches your target. For your own target or sources, use the composition helpers above. See docs/reference/migration-2.0.md for the 2.0 naming changes.

Documentation

Quickstart, concepts, and the full API reference live at https://mostlyright.md/docs/sdk/.