Infrastructure: Submission format and input standardization

[duncanwp]

Description

Define and implement the standard submission format for model output.

Requirements

• CMIP-standard variable names (see Table A2 in protocol)
• Specify required grid resolution for submissions
• Handle both coupled and atmosphere-only submissions
• Metadata collection: ECS, TCR, model type, training data description
• Validation script to check submissions are complete and correctly formatted

Notes

• Must accommodate diverse model architectures (ESMs, neural GCMs, emulators, hybrids)
• input4MIPs forcing data access and format

[duncanwp]

Proposed solution: adopt ICONEval's template system for recipe generation

ICONEval's recipe template system addresses submission standardization by generating recipes programmatically:

• Placeholder-based templates ({{dataset_list}}, {{timerange}}) allow a single recipe template to work for any model. ClimateBench2 should adopt this with {{model_list}} placeholders so submissions only need to provide a model name.
• Tag-based selection (#TAGS tier1 energy-balance) enables running subsets of diagnostics (e.g., --tags tier1 for entry-ticket tests only).
• Data availability check — ICONEval's recipe_check_data_availability validates that all required variables exist before running the full evaluation. This should be the first step in any ClimateBench2 submission pipeline.

Implementation: create esmvaltool/generate_recipes.py that reads templates + a model config and produces runnable recipes. This standardizes the input format to a simple model name + data path.

Phase: 1 (Foundation) in the integration plan
ICONEval reference: iconeval/_templates.py, recipe_check_data_availability.yml