Production progress

Only complete, newline-terminated committed batch records count. Assignments and discards both count as processed; secondary links and retries do not count again. Country completion requires a matching done.json and the expected primary mention IDs.

Throughput uses approximately the last 30 minutes, restricted to the current launcher session. Batch completion time is inferred from call timestamps and durations, not document publication dates. The overall ETA is all remaining mentions divided by this observed wall-time throughput. It includes unfinished work in failed countries. Queued country ETAs are not invented.

Hover over Recent throughput, or activate View throughput trend, for a two-hour history of the same rolling metric with total and endpoint lines. History is reconstructed from committed logs, not browser polling, and starts fresh at each launcher session. Peaks and declines reflect batches entering and leaving the rolling window; they are not instantaneous request rates.

This is a rough projection, not a deadline. Context growth, endpoint quotas, loading, retries and the final low-parallelism tail can change it. Failed or stalled countries trigger warnings, not suppression of the volume estimate: they still require attention, and time spent waiting for intervention is not included. Early estimates are provisional. ETA is withheld for stopped/unverified launchers, data errors, or no committed mentions in the throughput window. A stalled country's own ETA is withheld after 15 minutes without a commit.

Elapsed wall time includes pauses. After a restart the original start is approximated using the earliest recorded batch when the previous launcher start is no longer available. This page is read-only: it never changes the running sweep or its files.

Cost sums each call's reported USD in committed and failed-batch logs, including initial, retry and repair calls. Batch-level metadata is not added again. Calls in progress or not yet written to a batch log are excluded. Missing costs are flagged, not assumed to be zero.

Estimated total cost = logged cost / committed mentions × total target mentions. This all-run average includes observed failure overhead and survives restarts. It is a rough projection, not a billing total or spending cap: context growth, caching, retries and the changing mix of countries may change cost per mention.