Statistics

The Statistics page is Sublarr’s analytics dashboard. It collects everything Sublarr does — subtitle downloads, machine translations, sync runs, provider performance, and host health — into six sections that share a single time-range selector. Open it from the sidebar; the page fetches each section independently and shows a spinner until the first data arrives.

Statistics data flow

Time range

A row of buttons in the header controls the window for the time-sensitive sections. The default is 30 days.

Range	Window
24h	Last 24 hours
7 days	Last 7 days
30 days	Last 30 days
All	Every record, no cut-off

The range applies to Overview counts, the Trends chart, the Subtitles breakdowns, and the Translation section. The Providers table and the System section are always live snapshots and ignore the range.

Note

Each section caches independently. Snapshot sections are cached for about five minutes server-side; the System section refreshes on a shorter cycle and re-polls roughly every 30 seconds while the page is open.

Overview

Six headline tiles summarising the current state and the selected window:

Tile	Meaning
Subtitle files	Total subtitle downloads recorded (all time).
Downloads	Subtitle downloads within the selected range.
Translations	Translation events within the selected range.
Wanted	Items currently in the wanted backlog.
MT awaiting original	Provisionally machine-translated items still waiting for a real source to finalise.
Avg download score	Mean match score of downloads in range.

Trends

A multi-series line chart over the selected range, plotting three daily counts:

• Downloads — subtitle files written per day
• Translations — successful translation events per day
• Syncs — successful sync runs per day

The chart reads from a pre-aggregated daily rollup rather than scanning the raw event tables, so it stays fast even over long windows. See How the data is produced.

Note

On a fresh install the chart may be empty until the first rollup runs. The rollup backfills up to 90 days of history on its first execution, so past activity appears after the first nightly run (or after a manual run of the stats_rollup scheduler job).

Subtitles

Three breakdown panels covering downloads within the selected range:

Panel	Breaks down by
By source	Where the subtitle came from (e.g. provider download, embedded extract, translation).
By language	Language of the downloaded subtitle.
By provider	Which subtitle provider supplied it.

Each panel is a horizontal bar list; empty dimensions collapse to a “no data” note.

Translation

Four tiles plus a per-backend breakdown, all scoped to the selected range and counting only successful (ok) translation events:

Tile	Meaning
Characters translated	Sum of output characters across successful translations.
Estimated cost	Estimated spend in USD, derived from per-event cost estimates.
Quality-gate pass rate	Share of translation events that passed (successful events divided by all events).
Avg latency	Mean translation latency in milliseconds.

The By backend panel breaks the successful translations down by the engine that produced them (for example your configured LLM or DeepL).

Providers

A live table of every subtitle provider that has recorded activity. Rows with no searches and no downloads are hidden. Sorted by download count.

Column	Meaning
Provider	Provider name. A · disabled marker appears when the provider has been auto-disabled by the circuit breaker.
Searches	Total searches issued to the provider.
Hit rate	Successful searches divided by total searches.
Downloads	Successful downloads from the provider.
Avg score	Mean match score of its downloads.

This section is a live snapshot of the provider statistics table and is not affected by the time-range selector.

System

A live snapshot of the host and database:

Tile	Meaning
CPU	Current CPU usage of the host.
Memory	Current memory usage.
Disk	Disk usage of the root volume.
Uptime	How long the Sublarr process has been running.
Database size	On-disk size of the database (SQLite file size, or the reported database size on PostgreSQL).

Note

The system tiles depend on psutil. If it is unavailable in your environment the resource tiles read zero, while database size and uptime still report.

How the data is produced

Most sections query Sublarr’s live tables directly — subtitle downloads, translation events, provider statistics, and the wanted backlog — and cache the result briefly at the API layer.

The Trends chart is the exception. A scheduled rollup job aggregates each calendar day’s downloads, translations, and syncs into a single row per date, so the chart never has to scan the raw tables:

• The stats_rollup scheduler job runs once a day and rewrites today and yesterday (to catch late-arriving records), then fills in any missing historical day within a 90-day window.
• On first run against an existing library it backfills up to 90 days of history, skipping days with no activity so a new install doesn’t accumulate long runs of empty rows.
• The job is idempotent — re-running it for any date recomputes and overwrites that date’s row.

You can trigger a rollup on demand from Settings → System → Scheduler by running the stats_rollup job, which is useful right after first install to populate the trend chart immediately.

Tip

Rollup timestamps and day boundaries are computed in UTC. If your server’s local day differs from UTC, a download near midnight may land in the adjacent day’s bar on the Trends chart.