Problem

isRelayHealthy() and checkRelayHealth() in src/utils/relay-health.ts don't read the relay's pool state fields, causing false-healthy reports when the relay pool is degraded.

What the relay returns (v1.26.1)

{
  "status": "ok",
  "version": "1.26.1",
  "nonce": {
    "poolAvailable": 20,
    "poolReserved": 0,
    "conflictsDetected": 47,
    "circuitBreakerOpen": true,
    "effectiveCapacity": 1,
    "poolStatus": "critical",
    "lastConflictAt": "2026-03-28T17:53:50.250Z",
    "recommendation": null
  }
}

What the code reads

• isRelayHealthy(): HTTP 200 + status === "ok" — misses pool degradation entirely
• checkRelayHealth(): reads status and version only — same gap

What should be checked

At minimum, checkRelayHealth() should flag:

• nonce.circuitBreakerOpen === true → issue: "Relay circuit breaker is open"
• nonce.poolStatus === "critical" → issue: "Relay nonce pool is critical"
• nonce.effectiveCapacity < 5 → issue: "Relay effective capacity degraded (N/20)"
• nonce.conflictsDetected > 10 → issue: "Relay has N nonce conflicts"

isRelayHealthy() (the hot-path probe) should at minimum check nonce.circuitBreakerOpen !== true.

Additional: only wallet 0 nonces are checked

Both functions query Hiro for the hardcoded sponsor address SP1PMPPVCMVW96... (wallet 0 of 10). The relay now exposes GET /nonce/state with per-wallet state. checkRelayHealth() should either:

• Use /nonce/state to get per-wallet health, or
• At minimum, surface the relay's self-reported pool health fields

New endpoint available

GET /nonce/state (v1.26.1) returns per-wallet: chainFrontier, assignmentHead, pendingTxs, gaps, circuitBreakerOpen, healthy, queueDepth, settlementTimes, plus global probeQueue state.

🤖 Generated with Claude Code