3D Witness Confidence Tower

Confidence Tower

Witness count (N) 6

Individual reliability 0.75

Witness correlation ρ 0.00

Naive tower confidence

–

Capped (corrected) confidence

–

Effective witnesses N_eff

–

Overcounted layers

–

How it works

Each witness node in the ring contributes one stacked layer to the central tower — its log-likelihood ratio LLR = log(r/(1-r)) for reliability r. Stacked without correction, N independent-looking layers give the naive (red-capped) tower height:

log-odds_naive = Σ LLR_i   (correct only if witnesses are truly independent)

When a shared contamination source (glowing orb above the ring, brighter beams as ρ rises) links the witnesses — a rumor, viral post, or pre-report conversation — their testimonies overlap. The gold disc marks where the tower's height gets capped once you account for that:

deff = 1 + (N - 1)·ρ,  N_eff = N / deff
log-odds_corrected = N_eff · mean(LLR)

The blue portion of the tower below the gold disc is the honestly-supported height; the translucent red portion above it is evidence the naive sum over-counts because it treats correlated testimonies as if each were a fresh independent data point. Drag ρ to 1 and watch the cap collapse the tower down to roughly one witness's worth of height, no matter how many nodes ring the base.

Witness node / supported layer Overcounted (naive-only) layer Correction cap (N_eff height)

⚠ Couldn't load the 3D engineThree.js failed to load from the CDN. Check your connection and reload.