For researchers

You want to run a macro-forecasting study, get the artifacts, and read the results. You do not want to author new model classes or modify the package source. Start here.

If you want to …

Goal	Page
Set up the package and run your first recipe	Quickstart
Build your first study with a working recipe template	First study
See what the runtime executes today vs. what is schema-only	Runtime support matrix
Understand the output directory and manifest layout	Understanding output
Look up FRED-MD / FRED-QD / FRED-SD column dictionaries	FRED datasets
Check FRED-SD T-code policy (transform-code defaults for state series)	FRED-SD transform policy, inferred T-codes, v0.1 review
Browse the 38 bundled recipes (with extras + smoke-test status)	Recipe gallery
Use the high-level Python facade (mf.forecast / mf.Experiment)	Simple API
Bring your own monthly or quarterly data (CSV / Parquet)	Bring your own data

Public API at a glance

import macroforecast as mf

# Run any recipe end-to-end. Iterates every {sweep: [...]} cell, applies
# L0 failure_policy + seed, returns ManifestExecutionResult.
result = mf.run("recipe.yaml", output_directory="out/")

# Re-execute the stored manifest and verify per-cell sink hashes match
# bit-for-bit.
replication = mf.replicate("out/manifest.json")
assert replication.sink_hashes_match

For the full curated reference see encyclopedia/public_api.md. Browse every recipe axis / option in the encyclopedia.