A schema you can read. A contract you can discuss.
Explore a proposed data dictionary for network-specific records and evidence-linked context. These examples are local, illustrative files—not live API responses.
Start with the record’s meaning.
Use the same envelope for the source context while preserving the native identifier model. The example below contains field names, not a current balance or price.
{
"network": "solana",
"asset_type": "token",
"identifier_type": "mint_address",
"fields": [
"raw_amount",
"decimals",
"slot_context"
],
"mode": "illustrative_schema"
}Small fields. Important differences.
| FIELD | TYPE | MEANING |
|---|---|---|
network | string | The network context for this schema; do not identify an asset by symbol alone. |
asset_type | string | A native asset or a token record. Keep the underlying identity model explicit. |
identifier_type | string | The network-specific kind of identifier required by the actual record. |
fields | array of strings | A proposed field selection, not a claim that a dataset has been provisioned. |
mode | string | Always illustrative_schema in these local examples. |
source_record_id | string or null | The evidence or observation reference used to trace an output. |
observed_at | timestamp or null | When the source observation was collected, distinct from its effective time. |
review_state | defined enum | The actual review performed, not a broad safety or investment rating. |
Specify missing values before writing a client
In this proposed contract, null means unavailable. An empty collection means a completed query with no matching entries. Zero means a measured quantity. Define the policy explicitly rather than substituting one for another. A real feed must document the conventions that actually apply to its records.
Make a schema change reviewable
Version changes to meaning, units, filtering, or observation policy—not just renamed properties. Include schema and coverage versions in an export manifest. A client should be able to distinguish new source data from a changed interpretation of the same information.
Test recovery with a small acceptance sample
Repeat a delivery, interrupt a historical range, revise an observation, and remove optional metadata. Verify that stable identifiers prevent double counting and that missing data remains visible. Add evidence-age and unsupported-classification cases for RWA and AI-related research.
Your next build starts with better data.
Define your assets, fields, evidence, and delivery needs. Start with a clear brief—not an assumed service package.