Karuthu Vellam · Methodology — math, ε, k-anonymity, refusals

Full methodology page for Karuthu Vellam. Pipeline per era-week reading: (1) build sparse vote matrix V[participant × statement] with values {+1, -1, 0, null}; (2) mean-centre columns ignoring nulls; (3) compute statement-by-statement covariance; (4) power-iterate top 2 eigenvectors (deterministic seed); (5) project participants into 2D opinion space; (6) k-means k ∈ 2..5 by silhouette score; (7) surface representative + consensus + divisive statements; (8) write reading with math_version stamp. Live counter: every minute, Laplace(0, 1/ε) added to global vote count; ε=1.0 default. k-anonymity floor: N<25 cohort cells never written. Statement source policy: closed (MP Packs / Unmai desks / Magalir / Case Organ only) until graduation gates close. Refusals (structurally impossible outputs): no '99% of Tamils support X'; no per-individual identification from the counter; no live opinion-space map; no per-statement live tally on the stream.