Every article you see on EVERYWEAR has been scored, ranked, and categorised by an automated pipeline running daily across 60 RSS sources. Here is exactly how it works — the scoring formula, the category logic, the source selection, and the prediction system.
Every article receives an EWEAR score from 0 to 100. This is a composite of five components, each measuring a different dimension of editorial value. The score determines article placement across all sections of the site — leaderboard, Best feed, category rankings, and the weekly briefing.
A score of 70+ indicates a strong, highly relevant article. Top scores typically land between 80 and 85. The theoretical maximum is 100, but multi-component scoring means very few articles approach it.
Every article is classified against eight categories using keyword matching across titles, previews, and URLs. An article can match multiple categories simultaneously — a review of bone-conduction earbuds that also tracks heart rate might sit in both Hearables and Fitness & Health.
The category vocabulary is hand-curated and updated regularly. Brand names are handled separately — a Garmin article defaults to Smartwatches, a Meta article defaults to AR/VR/XR — but only when wearable context words are also present, preventing off-topic brand coverage from slipping through.
The pipeline fetches from 60 RSS feeds daily, spanning three tiers:
Sources are evaluated continuously via a source quality tracker that computes the average EWEAR score of all articles from each publication. Sources that consistently produce low-scoring articles are candidates for removal. New sources are added when a category is underserved.
Each day's editorial signal — the WTI Signal (Wearable Technology Intelligence) — is generated by Claude Sonnet, Anthropic's language model, using the day's top five articles by EWEAR score plus the current trending topics. The model is prompted to produce a 2–3 sentence editorial that identifies the underlying story, not just the surface events.
The WTI Signal is not a summary. It is an editorial interpretation — what the day's coverage means for the industry, what's being missed, and where the industry is heading. If the AI step fails, a rule-based fallback generates a plain signal from the top article's metadata. The source is always disclosed.
Article-level "So what?" implications — the one-line interpretations visible in each card's score breakdown — are also generated by Claude Sonnet across the top 10 articles daily.
EVERYWEAR maintains a public prediction ledger — specific, falsifiable calls about where wearable technology is heading. Each prediction has a deadline, a confidence level (0–100%), and a signal explaining what evidence informed it.
Predictions are resolved manually when sufficient evidence exists — either a confirmed announcement or a confirmed miss. The pipeline flags potential resolutions daily by scanning top articles for entity matches and resolution-signal language, suppressing speculation words (reportedly, rumour, leak). All resolved predictions remain on the ledger with their outcomes.
The hit rate is public and updated as predictions resolve. EVERYWEAR's track record is part of its editorial accountability — wrong calls stay on the record.