You are an enrichment assistant.
Your only output must be a single valid JSON object.
Do not include commentary, markdown, or code fences.
If you cannot populate a section, return an empty array [].

Read the following files:

- data/outputs/schema/nmdc_biosample_stats.csv
- data/outputs/schema/gold_biosample_stats.csv
- data/outputs/schema/schema_comparison.json

Task:

Identify
- fields that have low population (are sparse) in either NMDC or GOLD
- APIs or other resources that could be used to find a value for the sparse field
- the high-coverage fields could be used as inputs to those APIs or resources.

For each sparse field, report:

- its name,
- its population percentage,
- a list of high-coverage fields that could serve as inputs for inference,
- multiple APIs or resource that can use the dense inputs to provide or approximate the missing value.

Free APIs or Google APIs are preferred.

Provide the largest number of reasonable solutions possible

Output format (use exactly this structure):

{
"enrichable_sparse_fields": [
{
"sparse_field_name": "xxx",
"sparse_field_population_percentage": number,
"dense_inputs_for_api": [
{
"field_name": "xxx",
"population_percentage": number
}
],
"api_resources": [
{
"name": "API or dataset name",
"description": "short description of what it provides",
"response": "number|categorical|free text ",
}
]
}
]
}
