pnclENGINEGet API access

CAPACITY NOTES

Stale odds: ask the feed before your own logs

A freshness incident starts with one question: whose clock stopped. The feed's own health and ping endpoints answer it in one request each.

Your dashboard shows stale odds and the instinct is to open your own logs. Hold that instinct for five minutes. The first question in any freshness incident is whose clock stopped, yours or the feed's, and the feed answers it directly. Debugging your own pipeline before asking is how a provider hiccup becomes a wasted afternoon.

Ask the feed before your own logs

Two status endpoints exist for exactly this moment, as the health check guide documents, checked 2026-10-30. A public /ping returns a plain ok for probes that cannot authenticate. An authenticated /health returns the operational state of both feeds: status, live_write_age_ms, the source currently carrying live data, and a version field that is the same cursor your board endpoints return as last.

One request each, and the /health call counts against your plan like any other request, so once a minute per environment is the right cadence.

Read the whole answer, not the top line

The top-level fields describe the live feed. The prematch service is a nested object with its own status, write age and version, and the top-level status does not change for a prematch problem. A monitor that reads only the top field is blind to half the feed.

Two traps matter in an incident. A prematch status of unreachable means the health check itself could not reach that service: treat it as unknown, not down. And write ages normally sit well under a second, so a reading in the tens of thousands of milliseconds is already a story, whatever the status field says.

The cursor comparison settles the blame

The strongest field in the response is version, because it is comparable. Your poller stores the cursor from its last read. If the feed's version matches your stored cursor, the feed has not moved past you: the staleness is yours. If the feed's version is ahead and still writing, the feed is alive and your pipeline is the one not reading.

That single comparison replaces the entire speculative phase of the incident. No guessing about deploys, no re-reading your parser. Either the feed stopped, or your side did, and the cursor says which.

When the feed really is degraded

A degraded read is a reason to hold, not to hammer. Keep serving your 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 budget trip your own alerts rather than the error loop.

The pipeline monitoring guide covers the consumer-side metrics that watch the rest of the incident: snapshot age on your side, cursor progress, error rates. The health endpoints answer the prior question those metrics cannot: whether the problem is upstream of everything you control.

Five minutes, two endpoints, one cursor comparison. Debug in that order and most freshness incidents end before your own logs are even open.