Observed reliability

Capture interval by dataset

Seconds between consecutive captures of the same scope, pooled across scopes. It bounds how long a change can go unseen, and it is not latency. p95 needs 20 samples and p99 needs 100; below that they are withheld. A scope captured once adds no interval, so rarely visited scopes wait longer than these figures show.

Sportsbook (ESPN-attributed mainlines): Measured from source_snapshots, which keeps only the newest 500 raw responses, so this window is short by design and moves forward.

Sportsbook (native scopes): Measured between non-quarantined capture receipts, including receipts requiring resync. This is observation cadence, not proof that every receipt applied a price update or that a quiet feed has a gap.

Capture interval by league and scope

Ingest time

Seconds from a capture's observed_at to the completed_at of the batch run that contained it. This is an upper bound, because a capture commits before its batch closes. The API reads the same tables with no application cache, so this bounds capture-to-available.

Worker uptime from job runs

Each job has a target interval. We split the window into bins of that size and count the bins holding at least one stored capture. Targets are configuration, not a promise.

pick6: Share of 5-minute bins holding at least one stored Pick6 capture receipt.

dfs: Share of 15-minute bins holding at least one stored DFS capture receipt.

sportsbook: Share of 5-minute bins holding a retained ESPN response or a stored sportsbook batch. Older bins rely on batches, which are written only when a payload changes, so this is a lower bound.

Longest gaps

Gaps have bin precision. Known causes are on incidents.

Pick6 batch runs

Since Nov 3, 2026, 17:05 UTC: 62 runs, 46 success, 16 partial, 0 failed, 0 still running or interrupted. Latest run started Nov 3, 2026, 22:44 UTC.

Latest job state

Worker uptime from heartbeats

Each scheduled refresh records whether a live worker was seen, meaning a heartbeat within 90 seconds, at about two samples an hour.

API handlers

Responses recorded inside the application's GET handlers. Client errors and our own planned tests are excluded from the share.

Window Nov 3, 2026, 18:25 UTC to Nov 3, 2026, 21:30 UTC, 17 finished. 6 client errors and 4 planned tests excluded. 0 recording errors.

Scope health: active, current and stale

Active says the scope still exists at the source as far as we can tell. Current and stale say how old our last capture of it is. A stale active scope is a real gap in our data.

Pick6

A scope is one discovered sport, group and category. Active: enabled and seen by discovery in the last 24 hours. Current: its last complete capture is at most 900 seconds old, the same test as the API stale flag. Stale: older. Never captured: no complete capture stored.

DFS

A scope is one draft group. Active: still listed by the last slate catalog. Current: player list captured within 900 seconds. Stale: older. Never captured: no player list stored.

Sportsbook (ESPN-attributed mainlines)

A scope is one league. Active: has an unfinished event seen in the last two days. Current: an event in that league was captured within 900 seconds. Stale: older.

Sportsbook (native scopes)

A scope is one subscribed native market scope. Active: holds an authoritative baseline. Current: the scope's own health check passes, which needs a receipt or transport heartbeat within seconds, not 900. Stale: everything else.

Game context (ESPN)

A scope is one league. Current: a context capture within 900 seconds, the same test as the API stale flag. These captures follow game summaries, so stale outside game windows is expected. Active is unknown and is not asserted.

Independent statistics (ESPN)

A scope is one league of retained game samples. There is no freshness target for this dataset, so stale here only means no capture in the last 900 seconds. Active: holds a game not yet final.