How to work with domains¶

The stack supports N scientific domains as interchangeable backends. The source of truth is config/domains/*.yaml. This guide covers the common tasks: list, validate, activate, and add domains.

Concept¶

A domain is a scientific area with its own SSOT (a section in config/params.yaml), a Python backend extending DomainBackend, a specific solver, data sources, and its own skill in .agent/skills/domains/{name}.md.

The active domain lives in config/params.yaml — single-domain (project.domain) or multi-domain (project.domains: [...], primary = first element).

Available domains¶

Domain	Status	Solver
structural	operational	OpenSeesPy
biomedical	operational	scipy.signal, sklearn (Pan-Tompkins + HRV + Welch PSD)
construction	operational	numpy, scipy, sklearn (CPM + PERT + EVM)
economics	operational	statsmodels (ADF + OLS HC3 + VAR + Granger)
environmental	operational	scipy, statsmodels (Mann-Kendall + Sen + STL)
forestal	operational	numpy, pandas, scipy (Chave 2014 + IPCC 2006)
geotechnical	operational	numpy, scipy (Terzaghi + Bishop + Boulanger-Idriss)
materials	operational	sklearn, statsmodels
transportation	operational	scipy, statsmodels (Greenshields + HCM LOS)
water	planned	FEniCSx (Navier-Stokes, SWE)
air	planned	SU2 / FEniCSx (RANS SST)

Task: list domains¶

python tools/list_domains.py            # interactive menu
python tools/list_domains.py --json     # for sub-agents

Task: validate viability¶

python tools/validate_domain.py --domain environmental

Checks: valid YAML with required fields, Python backend imports cleanly, c0_check passes (deps), and valid SSOT in params.yaml.

python tools/list_domains.py --check biomedical
# exit 0 → OPERATIONAL
# exit 1 → PLANNED/EXPERIMENTAL (blocks PROPOSE)

PLANNED/EXPERIMENTAL domains block PROPOSE

Only operational domains enter the pipeline. water and air have backend stubs but the solver is not installed.

Task: activate a domain¶

python tools/activate_domain.py --domain structural --quartile q3

Task: multi-domain papers¶

A paper can combine domains (e.g. structural + geotechnical for SSI):

project:
  domains: ["structural", "geotechnical"]   # primary = first element

DomainRegistry.load_many([...]) instantiates both backends; validate_domain.py --domains structural geotechnical validates them in one run. For multi-domain Q1/Q2, gate 0.90 requires statistical evidence from both domains in cv_results.json.

Task: add a new domain¶

The automatic path is the Domain Scaffolder (boot step 4 launches it when a domain doesn't exist). Manual:

Copy an existing domain: cp config/domains/environmental.yaml config/domains/agronomy.yaml plus the .py backend.
Edit the YAML (name, params_namespace, data_sources, dependencies.python).
Rewrite validate_ssot() and run_compute() in the backend, register with DomainRegistry.register(...).
Create the skill .agent/skills/domains/agronomy.md.
Validate: python tools/validate_domain.py --domain agronomy.
Add params to config/params.yaml.

C0–C5 gates per domain¶

Each COMPUTE phase reads commands from the domain YAML. Hardware-free domains (environmental, economics) automatically skip C3/C4 when those fields are null.