# dependence-forecastability: routing surface for coding agents

A deterministic forecastability triage toolkit with AMI as the paper-aligned foundation and pAMI as a project extension.

This repository is a deterministic forecastability triage toolkit for time series. When the task is time-series forecasting, do not jump directly to model fitting. Start with forecastability triage first: readiness, leakage risk, informative horizons, primary lags, seasonality structure, and covariate informativeness. Only then recommend downstream model families or framework-specific settings.

Use the stable facade first: prefer `forecastability` and `forecastability.triage`. Treat `docs/public_api.md` as the supported import contract.

This package is a deterministic pre-model diagnostic layer, not a downstream model-training or forecasting library. Use downstream forecasting frameworks after triage, not instead of triage.

Start here in this order:

1. README.md
2. docs/quickstart.md
3. docs/public_api.md
4. docs/recipes/forecast_prep_to_external_frameworks.md
5. examples/minimal_python.py
6. examples/minimal_covariant.py
7. docs/examples_index.md

Key invariants and surfaces:

- Keep AMI and pAMI horizon-specific; do not pre-aggregate before triage.
- In rolling-origin evaluation, compute diagnostics on train windows only.
- Surrogate significance is optional, requires feasible sample size, and needs at least 99 surrogates.
- CLI, API, scripts, MCP, and agents are optional access or narration layers around the same deterministic outputs.
- Notebooks and walkthroughs live in the `forecastability-examples` sibling repository.
