Feed Scoring
RSSMonster assigns each subscribed feed a trust score between 0.0 and 1.0. The score estimates how useful that source is to the current user based on recent publishing behavior and reading interactions.
Feed trust is not a fact-checking score and does not claim that every article from a high-scoring feed is correct. It is also distinct from crawl health and article quality:
- Feed trust measures source usefulness, originality, consistency, and user affinity.
- Crawl health and reliability describe whether RSSMonster can fetch the feed successfully.
- Article quality evaluates an individual article’s content signals.
Feed scores are user-specific because feeds and their article interactions are stored per user.
The Observation Window
Each recalculation examines articles published by the feed during the previous 30 days. Older articles do not affect the new observation, which lets the score adapt when a source changes while preventing its entire history from dominating forever.
If a feed has no articles in that window, RSSMonster keeps its existing trust and attention metrics unchanged.
Inputs to Feed Trust
RSSMonster derives the observed score from three positive dimensions and several penalties.
Originality
Originality uses RSSMonster’s event relationships to estimate whether a feed often provides representative reporting or repeats coverage found elsewhere.
For recent articles assigned to events, RSSMonster considers:
- the proportion selected as the representative article for their event;
- the average number of articles in those events; and
- the proportion belonging to events with at least two articles.
An article in a single-article event is treated as original for this calculation. Articles in events containing two or more articles contribute to the feed’s duplication rate. Larger event clusters and a high duplication rate reduce originality, while frequently supplying the representative article raises it.
Event membership is semantic evidence, not deterministic duplicate identity. A lower originality contribution therefore means that the source commonly covers stories also covered elsewhere; it does not necessarily mean that its articles are literal copies.
Engagement
Engagement measures how the user responds to recent articles from the feed. Each article can contribute:
| Signal | Contribution |
|---|---|
| Favorite/bookmark | 1.0 |
| One or more outbound clicks | 0.5 |
| Skimmed, attention bucket 1 | 0.25 |
| Read, attention bucket 2 | 0.75 |
| Deep read, attention bucket 3 | 1.25 |
| Highly engaged, attention bucket 4 | 1.75 |
The combined contribution of one article is capped at 2.5. RSSMonster then averages these contributions across every recent article from the feed and normalizes the result to the 0.0–1.0 range. Feeds that publish many ignored articles therefore do not receive the same engagement signal as feeds whose articles are regularly saved, opened, or read.
See Bookmarks for why favorites are particularly useful as an explicit interest signal.
Publishing Consistency
Consistency is based on recent publication frequency. It rises linearly until the feed reaches four articles per day, where this contribution is fully satisfied. This rewards sources that publish consistently without giving an unbounded advantage to very high-volume feeds.
Negative Feedback and Volume
The observed score can be reduced by:
- articles explicitly marked Not Interested;
- very high publishing volume; and
- recent feed-mute history.
The high-volume penalty starts above 25 articles per day and reaches its maximum at 50 or more articles per day. It can reduce the observed score by up to 15%. Negative feedback uses a gradual square-root penalty, also capped at 15%, so a small amount of feedback matters without letting a few observations overwhelm the entire score. Recent mute history can apply a further penalty of up to 10%.
How the Score Is Combined
The scoring calculation starts from a neutral observation of 0.75 and makes bounded adjustments:
observed trust = 0.75
+ (originality - 0.50) × 0.10
+ (engagement - 0.35) × 0.35
+ (consistency - 0.25) × 0.08
The result is clamped to 0.0–1.0, after which volume, negative-feedback, and mute penalties are applied.
Engagement has the largest positive influence. Originality and consistency move the result more gently, so a feed is not rewarded merely for publishing frequently or being the only source covering a story.
Confidence for New and Small Feeds
RSSMonster avoids making strong judgments from very small samples:
- 10 or fewer recent articles have zero sample confidence;
- confidence increases linearly from 10 to 100 articles; and
- 100 or more recent articles provide full sample confidence.
The observed result is blended toward the neutral 0.75 baseline according to that confidence. A small feed can still accumulate useful metrics, but a short run of articles cannot immediately push its trust score to an extreme.
New feed records begin with a stored score of 0.5. Recalculation moves that stored value gradually toward the confidence-weighted observation.
Smoothing Over Time
After confidence weighting, RSSMonster applies an exponential moving average:
new trust = 0.35 × new observation + 0.65 × previous trust
Only 35% of one recalculation comes from the latest observation. The other 65% comes from the previous stored score. This makes feed trust responsive over time but resistant to sudden spikes, sparse data, and one unusually good or bad batch of articles.
Attention Metrics Stored with the Score
The same recalculation stores additional per-feed behavior metrics:
- average attention among articles that received an attention bucket;
- deep-read ratio;
- skim ratio;
- ignored-article ratio;
- average clicks per article;
- proportion of articles receiving at least one click; and
- number of attention samples.
These metrics help RSSMonster estimate reading affinity for new, unread articles from the feed. They are supporting observations and should not be confused with the final trust score itself.
Recalculating Feed Scores
Feed trust is calculated as a batch operation rather than after every click or read. This keeps the score stable and avoids expensive recalculation during normal reading.
From the interface:
- Open Settings → Feeds.
- Select Recalculate Scores.
The interface recalculates active feeds owned by the signed-in user and then reloads the feed table. The table displays each feed’s trust score separately from its crawl reliability.
For a command-line installation, run from the server directory:
npm run feedtrust
The command processes all active feeds. Run it periodically with the scheduler appropriate to your installation if you want scores refreshed automatically.
How Feed Trust Affects RSSMonster
Trust Sorting
Use the Trust sort in the interface or this search expression:
sort:trust
Articles are ordered by feed trust from highest to lowest, then by newest publication date and article ID. Trust sorting changes order; it does not hide articles from lower-scoring feeds.
See the Search Guide for combining Trust sorting with other filters or using it in a Smart Folder.
Article Quality
Feed trust and duplication history can gently adjust an article’s computed quality after a feed has enough attention samples. Confidence begins above 10 attention samples and reaches full strength at 100.
At full confidence, feed evidence can boost the article-quality multiplier by up to 10% or reduce it by up to 15%. The feed duplication rate contributes a penalty of up to 10%. The final multiplier remains bounded between 0.85 and 1.10, so feed reputation refines article quality rather than replacing the article’s own content signals.
Because article quality is one input to Recommended ranking, this creates a small indirect feed-trust influence even when explicit trust prioritization is disabled.
Prioritize High-Trust Coverage
The Unread and Daily Briefing preferences include Prioritize high-trust coverage. When enabled, RSSMonster adds the feed’s bounded trust score during runtime ordering. This makes trusted feeds more prominent in Recommended and other supported ordering modes without imposing a minimum-trust cutoff.
The setting affects priority, not eligibility: relevant coverage from a lower-scoring feed can still appear.
Trusted-Source Signal
Articles from feeds with a trust score above 0.85 can display a Trusted source signal. This is a product-level indication that the feed has developed strong behavioral trust; it is not an independent verification of the article’s factual accuracy.
Interpreting Feed Trust
Feed trust is most useful comparatively: it helps answer which subscribed sources have recently been original, consistently useful, and engaging for a particular user. Avoid treating small score differences as precise judgments. Sample size, recent behavior, and smoothing deliberately make the score evolve over multiple recalculations.
Most importantly, a low trust score does not remove a feed or filter its articles. RSSMonster keeps the underlying sources accessible and uses trust as one transparent signal among article quality, freshness, personal interest, event coverage, source diversity, and corroboration.