benchmarks/dynamical_systems/ibm_double_pendulum/_status.json

{
  "entry_id": "ibm_double_pendulum",
  "download_sh_written": true,
  "data_file_built": false,
  "data_file_path": "data/data.csv",
  "n_rows": null,
  "n_cols": 16,
  "references_downloaded": [
    "asseman2018.pdf",
    "asseman2018.bib",
    "ibm_double_pendulum_dax.bib",
    "ibm_double_pendulum_readme.md",
    "ibm_double_pendulum_index.md",
    "wikipedia_double_pendulum.html",
    "wikipedia_double_pendulum.bib",
    "levien_tan_1993.bib"
  ],
  "n_formulas_written": 3,
  "formula_ids": ["eom_theta1", "eom_theta2", "hamiltonian"],
  "metadata_yaml_written": true,
  "description_md_written": true,
  "reference_scores_computed": false,
  "reference_scores_by_formula": {
    "eom_theta1":  {"nmse": null, "r2": null, "rmse": null, "kendall_tau": null, "mape": null},
    "eom_theta2":  {"nmse": null, "r2": null, "rmse": null, "kendall_tau": null, "mape": null},
    "hamiltonian": {"nmse": null, "r2": null, "rmse": null, "kendall_tau": null, "mape": null}
  },
  "contamination_tier": "high",
  "notes": "Primary data source (IBM DAX CDN at dax-cdn.cdn.appdomain.cloud) was not DNS-resolvable from the authoring environment; download.sh was authored and tested (it fails cleanly with a hand-download fallback message), but data.csv was not built and reference_scores are therefore null. Paper (Asseman, Kornuta, Ozcan 2018, NeurIPS MDSD workshop) is saved at reference/asseman2018.pdf; it documents the apparatus (l1 = 91 mm, l2 = 70 mm, camera 400 Hz, 480x480 px) but NOT the bob masses m1, m2 or any dissipation model. Formulas are the textbook simple-pendulum (two point-mass) Lagrangian form: eom_theta1 and eom_theta2 are Euler-Lagrange, hamiltonian is T + V. All three were verified bit-exact against an RK4-integrated trajectory (max |f(X) - rhs| = 0.0 on 500 test points; H conserved to ~1e-5 J/kg over 20 s integration). Mass convention m1 = m2 = 1 kg because IBM does not report values; data.csv column H_mpernorm uses the same convention. Upstream columns (x_red, y_red, x_green, y_green, x_blue, y_blue in 5x-upscaled pixels) are preserved; derived columns (theta1, theta2, omega1, omega2, and their second finite differences) are documented in data/README.md. Default data.csv = video 0 of 21; data_all_videos.csv is the full concatenation. Contamination tier 'high' because the Lagrangian form is textbook standard, but the specific apparatus dimensions and real dissipation structure are low-contamination. Suggested use: Track C (red-team) probe for the Lagrangian form + Track B scoreable on the dissipation-dominated H residuals."
}