FEED HEALTH
Watch the feed's own heartbeat.
When a board goes stale, the first question is whose fault it is: your
poller or the source. The Pinnacle odds API answers it directly on two
status endpoints: an authenticated /health with per-feed
detail, and a public /ping that returns a plain
ok. Poll them on a slow loop and a degradation shows up in
your dashboard before it shows up in your product.
What /health reports
One GET with your API key returns the operational state of both feeds.
The top-level fields describe the live feed only: status is
healthy or degraded with an optional
degraded_reason, live_write_age_ms is the
milliseconds since the last live write, source says whether
live data currently arrives over ws, rest or
none, and version is the same cursor the board
endpoints return as last.
{
"status": "healthy",
"degraded_reason": null,
"live_write_age_ms": 546,
"source": "ws",
"events_in_store": 145,
"version": 50352678,
"prematch": {
"enabled": true,
"status": "healthy",
"write_age_ms": 71,
"version": 50352723
}
}
The nested prematch object is a second service, not a
detail. It carries its own status, which can also be
unreachable, and its own write_age_ms and
version. One side can be degraded while the other is fine,
and the top-level status never changes for a prematch problem, so a
monitor that reads only the top field is blind to half the feed.
Wire it into your monitor
Poll /health once a minute per environment; each call is one
request against your plan, so it is a cheap heartbeat, not a second
poller. Alert on two conditions: status leaving
healthy on either feed, and live_write_age_ms
or prematch.write_age_ms crossing your freshness budget.
Both ages normally sit well under a second, so a reading in the tens of
thousands is already a story.
Two subtleties keep the alerts honest. Treat prematch
unreachable as unknown, not down: it means the health check
itself could not reach the prematch service while the data may be fine.
And compare version against the cursor your own poller last
stored, because matching cursors prove your pipeline is reading the same
feed the status page describes. For load balancers and uptime probes that
cannot send an auth header, /ping is public and returns
ok in plain text.
When the feed is degraded
A degraded read is a reason to hold, not to hammer. Keep serving the last
good snapshot with its age visible, widen your poll interval so you do
not add load while the feed recovers, and let the freshness numbers on
the next /health call decide when to tighten again. Page a
human only when the degradation outlasts the tolerance your product
announced.
This endpoint watches the source; your own pipeline still needs its five consumer-side metrics from the pipeline monitoring guide, and a degraded period can surface as ordinary error responses, which the error runbook covers. The status page on the pnclPULSE SSE setup guide is the reference for what a healthy stream looks like over time.
Questions this raises
Is /health authenticated?
Yes, it takes the same API key as any other endpoint, in the header or the key parameter. /ping is the public one, meant for probes that cannot send credentials.
How often should I poll /health?
Once a minute per environment is plenty. Every call counts as one request against your plan, so five dashboards each polling every five seconds add up to real quota for no extra signal.
Does prematch unreachable mean prematch is down?
No. It means the health check could not reach the prematch service; the data itself may be fine. Treat the reading as unknown and confirm with your own prematch poll's freshness.
Can I trust the top-level status alone?
No. It describes the live feed only. A prematch outage never moves it, so watch prematch.status and prematch.write_age_ms as separate series.
Status fields and endpoint behaviour follow the vendor's published documentation, checked on 29 October 2026. Published .