Readiness and freshness

A configured source is not a healthy source

Configuration means someone wrote a setting. Readiness means the service can currently answer correctly. They are unrelated.

The endpoints

Endpoint	Answers
GET /live	Is the process running?
GET /ready	Can it answer correctly, and if not, why?
GET /health	What is the state, including unavailable?
GET /version	Which service and provider revisions?
GET /metrics	Operational metrics, where enabled

/live returning 200 says nothing about whether an answer would be correct. Read /ready.

What readiness actually checks

The NFT and Realm read model reports ready only when all of the following hold:

• a generation identifier exists;
• the indexed height is a safe integer;
• the view is not stale;
• the tip is not mixed;
• the integrity check has not failed.

When any fails, the response is 503 with the reason, the state, the source, the network, the generation identifier, the indexed height, the source revision, and the counts.

That response is designed to be actionable rather than merely negative.

The six states

State	Meaning	What a caller does
Configured	A setting exists	Nothing. Wait for readiness
Live	The process is running	Nothing. Check readiness
Ready	Can answer correctly	Proceed
Degraded	Answering from an incomplete view	Do not act on the answer
Stale	Answering from an old chain position	Check the indexed height
Mixed tip	The generation spans more than one provider tip	Do not act on the answer

Declared coverage

The ARC-20 source declares partial coverage with an explicit reason: confirmed authoritative history and complete proven holder snapshots are indexed, but the shipped adapter has no exhaustive mempool feed with stable pending lifecycle and disappearance handling.

That limitation is about pending activity. It does not weaken the confirmed scan or the holder proof requirements. Read the declared coverage rather than assuming completeness.

For product builders

Render the state, never a misleading zero:

• “We cannot reach the index right now” for unavailable.
• “Showing data from block N” for stale.
• “This view does not cover that asset type” for out of scope.
• “No assets found” only when ready and genuinely empty.

Production resilience

The production read service runs on the dedicated Universe indexer host and binds only to its local interface. A supervised private compatibility tunnel keeps existing application routes available during infrastructure changes, while automatic restart and health checks restore both the indexer and its private route without manual intervention. Endpoint consumers still apply the readiness and freshness rules above: resilience of the service does not exempt a caller from declaring coverage and lag.

Source

Unavailable index against empty balance.