DTI CLASS API

External exploratory compute backend for DTI-Core Grand Auditor.

Boundary

This API is exploratory and non-canonical.

It is not:

• a likelihood evaluation
• a posterior comparison
• a Planck validation pipeline
• a manuscript checkpoint updater

Endpoints

• GET /health
• POST /class/compute

Local run

pip install -r requirements.txt
bash run_local.sh

Optional CLASS/PyCLASS:

pip install -r requirements-classy.txt

Test

bash test_api.sh http://127.0.0.1:8000

Render deployment

This repository includes:

• Dockerfile
• render.yaml

Recommended Render settings:

• Service type: Web Service
• Environment: Docker
• Repository: fujikix1102/dti-class-api
• Branch: main
• Health check path: /health
• Start command: handled by Dockerfile

After deployment, check:

https://YOUR-RENDER-URL/health

Expected response includes:

"status": "ok"

If classy_available is true, CLASS/PyCLASS is available on the deployed backend. If classy_available is false, the API still works as a safe wrapper but does not run CLASS propagation.

Streamlit frontend connection

The Streamlit frontend should later call:

POST https://YOUR-RENDER-URL/class/compute

Boundary:

• exploratory backend
• non-canonical
• not a likelihood evaluation
• not a posterior comparison
• not a Planck validation pipeline

AxiCLASS + Cobaya DESI DR2 BAO endpoint

Bounded endpoints:

• GET /axiclass/desi-dr2-bao/health
• GET /axiclass/desi-dr2-bao/provenance
• POST /axiclass/desi-dr2-bao

The POST endpoint currently accepts only the locked Planck-2018-baseline-like parameter contract. It executes the pinned Linux AxiCLASS build through a Cobaya Theory provider and evaluates the inherited DESI DR2 BAO theory_fun and logp methods using explicitly bound frozen mean and covariance files.

Boundaries:

• no sampler;
• no posterior;
• no MCMC;
• no historical-chain reproduction;
• no normalized likelihood claim;
• no EDE-branch claim;
• no reuse of 30.06.