pnclPULSEGet API access

DROP ALERT NOTES

Recovering the drops you missed while offline

The stream does not replay what fired during an outage. The REST drops endpoint holds the record; here is how to merge it without doubling.

The reader was down for an hour: a deploy, a crashed process, a laptop that slept. The stream reconnects cleanly, but reconnection only covers what happens next. The drops that fired while you were gone are not replayed. Recovering them is a deliberate step, and the REST drops endpoint is the tool for it.

Accept that the gap is real

The stream pushes alerts as drops are detected and holds no backlog for you. When your reader reconnects, it starts from now. Anything that moved during the outage exists only in the record the API keeps, which is exactly what the endpoint serves: the same drops, pullable on demand, as the stream and endpoint comparison lays out, checked 2026-10-28.

So the recovery has two halves: get the stream live again, then ask the endpoint what you missed. Skipping the second half leaves a hole that later reads as a quiet market.

Ask the endpoint for the window

Pull the drops endpoint for the outage window, filtered to your sports. The comparison page documents the shape difference that matters here: the endpoint returns objects with from, to, market, side and sport_name, and it adds three fields the stream does not carry: drop_pct, age_s and is_live. The percentage and the age are computed for you, which is convenient, and the field names are different, which is the trap.

Write the mapping once and keep it with the relay: from_price reads from, to_price reads to, sect reads market, outcome reads side. Code that applies the stream's field names to endpoint objects silently loses the numbers, the kind of bug that looks like an empty result.

Merge without doubling

Your dedupe store decides what the outage cost. Match each endpoint row against the alerts you logged before the gap by event and outcome, as the dedupe guide keys them, checked 2026-10-28. Anything already in the store is a duplicate; anything new belongs in the log with a flag that says it arrived late.

That flag matters more than it looks. A backfilled alert is evidence about the market at alert time, not a signal to act now. Route recovered alerts to the log and the morning summary, not to the channel that pages people, unless your review of each one says otherwise.

Measure the outage, not just the gap

Close the loop with two numbers: how long the gap ran, and how many drops the endpoint showed for it. The first tells you whether your restart policy is doing its job. The second tells you what downtime costs in your sports, which is the number that argues for a standby reader later.

An outage handled this way leaves nothing behind: the stream is live, the record is complete, and the log says plainly which entries were recovered. The market kept moving while you were gone. Now your records agree.