Registration detector mistook unrelated words for account barriers
2026-10-01 — Added word boundaries and brand controls. Fresh Register samples showed dismissible consent; access decisions remained evidence-based.
Corrections
Confirmed measurement and grading defects, their repairs, and the evidence supporting them. A clean repeat does not silently erase an unresolved barrier.
Hidden text and consent UI polluted extracted article evidence
2026-10-01 — Removed computed-hidden and consent text before extraction while retaining visible article discussion. Repaired and verified Aniagotuje and TU Delft measurements.
Homepage typography was being scored as article readability
2026-09-09 — Withheld prose scoring for homepage-only evidence. Article discovery restores readability only after qualifying reviewed samples.