Source health
Every source Vitna reads, and whether it is still answering.
GET /api/v1/sources/healthcurl https://vitna.ai/api/v1/sources/healthResponse
{
"sources": [ { "source": "...", "provider": "...", "enabled": true, "stale": false, "consecutiveFailures": 0, "lastFetchAt": "...", "lastOkAt": "...", "cadenceMinutes": 60 } ],
"meta": { "generatedAt": "...", "source": "database", "count": 21, "staleCount": 0, "failingCount": 0, "ok": true }
}meta.ok is the one-glance answer: true when nothing enabled is stale and nothing enabled has failed three times running.
meta field | Meaning |
|---|---|
staleCount | Enabled sources past their cadence |
failingCount | Enabled sources at three or more consecutive failures |
ok | Both counts are zero |
A source that has been switched off is still listed, with enabled: false and the failure count it stopped at, and counts toward neither. It is no longer read, so that count can never move again, and /witness shows the source as held rather than failing.
Public on purpose
A red source is part of the record.
A tape that silently stopped updating because a pricing page moved looks exactly like a market that stopped moving. Publishing source health is what lets a reader tell those apart without asking anyone.
503 rather than a green light
If the record cannot be read, this answers 503 record_unavailable.
It never falls back to the committed snapshot. A health endpoint reporting green from a file on disk would be the one lie this whole system exists to prevent, and it would be reporting on sources it had not looked at.
Errors
| Status | Code | Cause |
|---|---|---|
| 503 | record_unavailable | The record is not reachable from this instance |
Caching
no-store. A cached green is a claim about the past.
