# REPRODUCIBILITY.md

## Environment
- Python ≥ 3.11; dependencies pinned in `uv.lock` (created with `uv sync`).
- LaTeX: TeX Live (tested 2026), `latexmk`.
- OS tested: macOS (Apple Silicon). No OS-specific code paths.

## Data
- Dataset: `mito0o852/OHLCV-1m` @ revision `776328445b7ac6e7815ef3a483e9c8ded1eb6d56` (pinned in `config/base.yaml`).
- Download: `python -m intraday_fragility.cli download --start 1992-01 --end 2026-03`
  (resumable; per-file sha256 verified against HF LFS metadata; manifest in
  `data/manifests/raw_files.json|.parquet|.csv`).
- Verify: `python -m intraday_fragility.cli verify-download`.
- Raw data are never committed; the manifest and checksums are.

## Pipeline
`make` targets run stages in order (see `Makefile`). Every figure has its source
data in `outputs/figures/data/`; every table exists as `.tex` + `.csv`. Numerical
claims in the manuscript come from `outputs/estimates/results_registry.json` via
generated LaTeX macros — no hand-typed numbers.

## Seeds
Global seed in `config/base.yaml` (`seed: 20260713`); stage-specific seeds logged
per run in `logs/`.

## Provenance
Each run logs Python version, package versions, dataset revision, command line,
and timestamps to `logs/<stage>/*.jsonl`.