ovr.news

Under the hood

How we rank stories

This section is currently available in English only.

Same for everyone

Every visitor to ovr.news sees the same stories in the same order. There is no personalization. No tracking. No engagement optimization. No A/B testing on article order.

Most news platforms rank content to maximize engagement: clicks, time on site, scroll depth. That creates filter bubbles — each reader sees a different version of reality shaped by their past behavior.

We rank content to surface what matters, not what performs. The algorithm is deterministic: given the same articles at the same time, every visitor sees the same result. This page publishes the full formula.

The ranking formula

Each article's position is determined by its display rank:

display_rank = score × 0.95age_days × recency_boost × corroboration_boost × diversity_boost

Where:

Worst-case stack. All three boosts can apply at once: an article less than 24 hours old, from a scientific source, with three or more corroborating outlets. The combined multiplier is 1.3 × 1.7 × 1.4 = 3.07×. This is rare and time-limited. Recency_boost decays after 24 hours. Diversity_boost is suppressed once the lens quota is met. Corroboration_boost requires upstream dedup to actually identify multiple sources. The full formula is published here precisely so this kind of worst-case can be checked.

In practice, this is what the formula looks like for a primary article that scored 8.0 — with and without corroboration (2 independent sources, 1.5× boost):

Time decay and corroboration effect on display rank for an article scoring 8.0
Age Decay factor Display rank With corroboration
Fresh (< 24h) 1.00 × 1.30 10.4 15.6
1 day 0.95 7.6 11.4
3 days 0.86 6.9 10.3
7 days 0.70 5.6 8.4
14 days 0.49 3.9 5.9

A corroborated 3-day-old article (10.3) still outranks a fresh uncorroborated one (10.4) only barely. By day 7, the fresh article wins. Corroboration gives a meaningful bump, not a permanent position.

Score threshold

An article needs a weighted average of at least 4.5 out of 10 to appear on the site. Below that threshold, the article is filtered out during the build. This is a quality floor, not a ceiling — most articles that make it to the site score considerably higher.

Editorial rules

After ranking, a rule-based editorial layer makes final adjustments. These are not AI. They are deterministic checks with configurable thresholds. Every action is logged with a reason.

Homepage interleaving

The homepage draws from all five lenses using round-robin interleaving: one from Thriving, one from Belonging, one from Recovery, and so on. This ensures every lens gets represented, regardless of which lens has the highest-scoring articles that day.

Within each lens, articles appear in display rank order. Between lenses, representation is equal.

What we don't do

The scoring filter definitions are published in llm-distillery on GitHub. The ranking formula is documented in full on this page.

Last updated: April 2026