CANDEL is a GPU-accelerated hierarchical Bayesian framework for the local distance ladder and peculiar velocities, built on JAX and NumPyro. It forward-models distance indicators from geometric megamaser anchors through Cepheids and TRGB to peculiar-velocity tracers, and infers $H_0$, $S_8$ and velocity-field parameters directly from the data.

Clone the core and the probes you need side by side in one folder, which also holds the shared data and results:

candel-cosmo/
  CANDEL/  candel-pv/  candel-ch0/  ...   git checkouts
  data/                                   inputs: catalogues, fields, field caches
  results/                                run outputs
  plots/  remote_logs/                    figures, logs pulled from clusters

Keep data/, results/, plots/ and remote_logs/ in this folder, outside every checkout. Where they belong elsewhere, for example on a cluster's scratch or data filesystem to stay within the home quota, set root_data and root_results in CANDEL/local_config.toml to the folders holding data/ and results/, and symlink plots/ and remote_logs/ into candel-cosmo/.

Full steps, including the Python environment and local_config.toml, are in the installation instructions (also on Read the Docs).

CANDEL is a core library plus one repository per probe.

• CANDEL is the core: inference (NUTS, evidence), selection integrals, field reconstructions, cosmography, run scripts and job submission, the documentation, and local_config.toml.
• Probe packages import the core; the core never imports them. Each one registers a candel.Probe under the candel.probes entry point, and the core's main.py picks it up from the config's model.which_run once the package is installed.
• Clone the probes you need next to CANDEL. Data and results sit in the same folder, outside every checkout, and all repositories find them through root_data and root_results in CANDEL/local_config.toml.

• Stiskalek et al. (2025), The Velocity Field Olympics — arXiv:2502.00121

• Stiskalek et al. (2025), A 1.8 per cent measurement of $H_0$ from Cepheids alone — arXiv:2509.09665

• Stiskalek et al. (2025), No evidence for $H_0$ anisotropy from Tully--Fisher or supernova distances — arXiv:2509.14997

• Stiskalek (2025), $S_8$ from Tully--Fisher, Fundamental Plane and supernova distances — arXiv:2509.20235

• Stiskalek et al. (2026), Forward-modelling Milky Way Cepheids — arXiv:2603.09880

• Stiskalek & Desmond (2026), A reanalysis of the megamaser Hubble constant — arXiv:2609.17684

• Stiskalek et al. (2026), $H_0$ from the Tip of the Red Giant Branch and geometric anchors alone — arXiv:2609.29996

• Documentation: candel.readthedocs.io

• Contact: Richard Stiskalek (University of Oxford)