formulation

DH data pipeline for formulaic language in REPUBLIC (Dutch States General) resolutions: find and type templated openings, decisions, and related skeletons via formula_detection, with paths and provenance through data_manifest.toml + data_io.

Plan: PLAN.md. Research notions: docs/formula-conceptual-overview.md.

Setup

# Bootstrap already ran `uv sync` and installed the Jupyter kernel.
# Re-run `uv sync` if you later change dependencies.
# Committed data_manifest.toml uses in-repo tier roots.
# Optional per-machine overrides (external drives, full S4 file):
# cp data_manifest.local.toml.example data_manifest.local.toml
# Data registry view (manifest-declared datasets + on-disk status)
uv run python -m data_io.check

Open the workspace profile in Cursor or VS Code:

cursor formulation.code-workspace
# or: code formulation.code-workspace

Layout

Path	Role
AGENTS.md	Agent instructions (Cursor and VS Code Copilot)
PLAN.md	Headlines checklist and dataset table
plans/	Milestones and step guides (fill when multi-step)
docs/DATA.md	Tiers, phases, manifest workflow
data_manifest.toml	Tier roots and registered datasets
notebooks/	Jupyter workspace (explore phase)
output/	Scratch-tier root (gitignored; see output/README.md)
data_io/	Manifest I/O and provenance helpers

Agents (Cursor / VS Code)

Read AGENTS.md first. The same standards load from:

• Cursor: .cursor/rules/project-standards.mdc
• VS Code Copilot: .github/copilot-instructions.md