Methodology

Scoring algorithm v1.0

Every phone is scored in seven blocks: Smoothness (benchmark, RAM, storage, refresh rate), Battery (regulated EU-label endurance hours first, capacity only as a low-confidence fallback, charging speed, efficiency class, cycle life), Camera hardware (a physical light-capture proxy from sensor size and aperture, optical stabilization, telephoto reach, ultrawide, video), Screen (brightness, panel type, sharpness, refresh), Build & comfort (weight ergonomics, IP rating, drop standards, protective glass), Longevity (remaining vendor software support, 5G, eSIM, modern Wi-Fi) and Basics (NFC, 5G, USB-C, fingerprint, gyroscope). The overall score is their weighted average (20/18/17/14/12/11/8 %).

Values are mapped through fixed, published anchor curves — "45 hours of endurance is excellent" — never normalized against the rest of the catalog. That keeps scores stable when phones are added, lets extreme spec-sheet numbers saturate instead of distorting everyone else, and makes every anchor open to criticism. The full anchor set ships as a public JSON (/data/anchors.json) and is versioned; this is algorithm 1.0.0.

Dealbreakers cap the overall score with a visible reason instead of hiding inside weights: no NFC caps at 75, no 5G at 82, both missing at 70, microUSB at 75, under 64 GB storage at 75, and over 280 g or 14 mm at 80 for non-rugged phones.

Missing data never silently counts as zero: each value carries a confidence, unknown values shrink toward a neutral 50, and every score shows its data evidence. Phones below 70 % evidence are excluded from rankings. Camera scores measure hardware potential — sensor, optics, stabilization — and are labelled as such; computational photography cannot be judged from specifications, and megapixels alone earn nothing.

Devices are classed by signals (feature phones, foldables, rugged phones with MIL-STD, IP69, thermal cameras or outsized batteries); feature phones are not scored and rugged phones compete in their own ranking. The scoring system was developed through the cooperation of multiple AI models and validated in several rounds of qualitative testing, so that it is as useful as possible to visitors who come here for advice on picking a new device.