One contract for historical and live data.

Historical rows and current reports keep the same shared columns, units, null behavior, timestamps, and source fields. Move from research to production without rebuilding field mappings.

Read the docs

archive-and-live.py

1from mostlyright import weather
2
3live = await weather.live.latest("KNYC")
4archive = weather.observations(
5    "KNYC", "2025-01-15", "2025-01-15",
6    return_type="list",
7)[-1]

1import { weather } from "mostlyright";
2
3const live = await weather.latest("KNYC");
4const { rows } = await weather.observations(
5  "KNYC", "2025-01-15", "2025-01-15"
6);
7const archive = rows.at(-1);

Change the time window, not your data model.

Historical and live reads preserve documented field meanings. Each mode can add its own timing or fetch metadata without renaming shared measurements.

One shared contract

Historical and live rows use the same names for event time, temperature, dewpoint, and source.

Source on every row

Each result retains the provider selected by source filtering or deterministic deduplication.

Clear time fields

event_time_utc states when the observation happened. Historical rows can add the station-local calendar day.

Cache by mutability

Completed historical periods can be reused. Current periods are refetched instead of treated as final.

Reuse historical data without freezing current periods

Python historical reads reuse Parquet partitions under~/.mostlyright/cache/. Node uses~/.mostlyright/cache-ts/; browsers use IndexedDB. Current periods are skipped or refetched so incomplete partitions are not treated as final.

Archive partitions: completed periods can be reused without treating the current period as final.

Source field: each observation row names the provider that produced it.

Fetch ledger: cache writes record source, station, window, fetch time, row count, and window hash.

weather example · ~/.mostlyright/cache

~/.mostlyright/cache/
├─ v1/observations/KNYC/2025/01.parquet
└─ provenance/iem.jsonl

import mostlyright as mr

reports = weather.observations(
    "KNYC", "2025-01-01", "2025-01-31", source="iem"
)
provenance = mr.provenance(reports)

What you stop rewriting when you switch to live data

Concern	Direct provider integration	Mostly Right
Historical and live	maintain separate clients and payload models	one shared data contract
Columns and units	translate provider fields in application code	documented names such as event_time_utc and temp_f
Source identity	track provider metadata separately	source kept on every returned row
Missing data	decode provider sentinels and empty payloads	explicit nulls and typed no-data errors

One contract for historical and live data.

Change the time window, not your data model.

Reuse historical data without freezing current periods

What you stop rewriting when you switch to live data

Research with historical data. Use the same contract live.