Successful CLASS/Cobaya run – looking for HPC validation with full likelihoods

[ProcessTriality]

Short summary: I implemented the Process-Triality extension in CLASS/Cobaya and successfully reproduced a full multi-likelihood run in a CPU-only Colab environment. I’m now looking for independent validation on a full local HPC setup including all Planck Clik likelihoods.

For the validation run of my Process-Triality extension, I intentionally restricted the likelihoods to a subset that is known to be robust and computationally manageable in a Colab environment. Specifically, I included:
• Planck 2018 low-ℓ TT and EE
• BAO (6dF)
• Supernovae: Pantheon, Pantheon+, Pantheon+SH0ES, DES-Y5, JLA, JLA-Lite, Union3

The following likelihoods were not included in the primary run:
• Planck 2018 high-ℓ TTTEEE
• Planck 2018 lensing
• BAO SDSS MGS
• BOSS DR12 consensus

These were excluded not because of issues with CLASS or Cobaya, but because:
1. The high-ℓ TTTEEE likelihood is computationally heavy, which makes long MCMC chains impractical in a Colab runtime environment.
2. Planck lensing and several BAO likelihoods require external Clik datasets that are difficult to deploy reliably in cloud environments.
3. I wanted to avoid false negatives or artificial convergence problems caused by missing Clik libraries rather than by the model itself.

All likelihoods that were included ran stably and produced well-behaved posteriors, so the core implementation appears to be functioning as expected.

Reproducibility

To make the setup fully reproducible, I uploaded:
• all input YAML files,
• the full output chain results,
• the entire lauffähige Python notebook environment,
• CLASS modifications,
• and the complete configuration used for the run

to the following repository:

👉 https://github.com/ProcessTriality/protri-class

Everything can be executed directly (e.g., via Colab), so all results are completely transparent and reproducible without additional setup steps.

Request for assistance

I would be extremely grateful if someone with a local CLASS/Cobaya installation (including the full Planck 2018 Clik suite) could run the model with the complete set of likelihoods, especially:
• Planck 2018 high-ℓ TTTEEE
• Planck 2018 lensing
• BOSS DR12 full consensus
• SDSS MGS BAO

This would be enormously helpful for verifying that the behaviour remains consistent under the full likelihood combination.

Thank you very much in advance!