Settings — Scoring

When Sublarr searches a provider it gets back a list of candidate subtitles. Scoring is the rule set that turns that list into a ranking. Whichever candidate ends up on top — provided it clears the profile cutoff — is the one that gets downloaded. Tweak scoring when you find Sublarr consistently picking the wrong release group, the wrong release type, or the wrong source.

How a score is built

Every candidate starts at a base score and gets adjusted by a chain of modifiers:

base_score (5.0)
  + release_match_modifier  (0 to +3)
  + source_match_modifier   (0 to +1.5)
  + resolution_match        (-1 to +1)
  + group_modifier          (-2 to +2)
  + provider_modifier       (-1 to +1)
  − penalty_rules_total     (−10 to 0)
= final_score (clamped 0-10)

The Library’s scoring detail panel shows the breakdown for every candidate so you can see exactly what nudged a subtitle up or down.

Default modifiers

Modifier	Default	Effect
Release match	+3.0	Filename release tags (e.g. 1080p.x264-RARBG) match the source release. Strongest signal.
Source match	+1.5	Source type matches (Blu-ray, WEB-DL, HDTV). Slightly weaker than release match.
Resolution match	+1.0	Resolution matches (1080p, 2160p).
Resolution mismatch	−1.0	Subtitle from a 720p release applied to a 1080p video.
Hash match	+5.0	Some providers ship hash-keyed subtitles guaranteed to fit the file. Very rare but a strong signal.

Adjust these in Settings → Subtitles → Scoring → Default modifiers.

Penalty rules

Penalty rules subtract score for known-bad signals. The default ruleset covers common pitfalls; add your own when you spot a pattern.

Rule	Default penalty	When it fires
Empty / placeholder file	−10	File size below 1 KB — likely empty.
Auto-translated	−4	Filename contains “auto”, “machine”, “AI translated”.
Wrong language tag	−6	Detected language doesn’t match the requested language.
HI when prefer-not	−2	HI subtitle when profile says prefer-not.
Forced when prefer-not	−2	Same logic for forced subtitles.
Suspicious group	−3	Filename matches the user-maintained “bad groups” list.

Custom rules add to the table:

Field	Effect
Pattern	Substring or regex (toggle) tested against the filename.
Penalty	Score subtraction (−10 to 0).
Notes	Free text — shown in tooltips.

Provider modifier

Different providers have different quality reputations. The provider modifier nudges scores up or down based on which provider returned the candidate:

Provider	Default modifier	Reasoning
OpenSubtitles	+0.5	Largest dataset, generally good.
Jimaku	+0.7	Anime-specialised, high quality.
SubDL	+0.3	Curated subset of OpenSubtitles.
Subscene	0.0	Mixed quality.
MyMemory	−0.5	Auto-translation source.

Edit the modifier per provider on the same page. The change applies to all future searches.

Auto-prioritisation

When Auto-prioritise providers is on (Settings → Providers → Auto-prioritise), Sublarr re-orders providers based on their rolling success rate. The provider modifier is then applied on top of the auto-derived order — combining static quality preference with empirical performance.

How rerank affects scoring

Sublarr’s reranking pass runs after the initial provider list comes back. It re-evaluates the top N candidates against extra signals (download count, age, peer review) when those signals are available:

Signal	Effect
Download count	Each download adds a small +modifier (capped at +1.0).
Recency	Recent uploads are favoured; six-month-old uploads with low download count get a slight − nudge.
Reviewer score	Some providers expose user ratings — used as ±0.5.

Configure rerank thresholds in Settings → Providers → Provider Reranking.

Score thresholds per language

Some languages have intrinsically lower-quality subtitles available. Override the global cutoff per language under Settings → Subtitles → Languages → Threshold per language:

Language	Reasonable cutoff
de	4.5
en	5.0
ja	3.5 (less mature dataset)
tr	3.0

A per-language threshold takes precedence over the profile cutoff for that language only.

Why a candidate “won”

Open Library → Series detail → episode → Subtitle history. For each subtitle download, Sublarr stores the final score and the breakdown:

Column	Effect
Final score	Number after all modifiers and penalties.
Modifiers applied	List of each non-zero modifier.
Penalties applied	Same for penalties.
Cutoff used	Whether profile cutoff or per-language override decided eligibility.

Scoring transparency means when something seems off you can audit the decision instead of guessing.