Team Consensus Map and Trust Matrix (2D)

Plan view of who listens to whom, and the trust matrix W that decides the culture.

Switch to 3D

Presets

Team

Matrix view

Clock

Model check

The science

1. DeGroot learning: how a team's opinions converge

Twelve people each hold an opinion x_i between 0 (keep the old way of working) and 1 (adopt the new way). Each round every person replaces their opinion by a weighted average of everyone's: x(t+1) = W x(t). Row i of the heat map says how much person i listens to person j; the diagonal is stubbornness. If everyone can be reached from everyone (and the chain is aperiodic) all opinions converge to x* = pi . x(0), where pi W = pi, sum(pi) = 1 (DeGroot, 1974). The entries of pi are the influence shares, drawn as node size.

2. What the 2D view adds

The matrix view shows W^k: as k grows every row turns into the same row pi, so "the team forgets who it started as" and all that is left is the influence vector. The slower this happens (two weakly linked teams), the longer the camps keep arguing: disagreement shrinks by the second eigenvalue of W each round.

3. Model checks

(1) simulated consensus equals pi . x(0); (2) a random walker following the trust weights visits people with frequency pi; (3) pi . x is invariant and, for the open plan, matches the closed form pi_i ~ a_i (A - a_i); (4) measured shrink per step equals the spectral radius of W - 1 pi; (5) 2D-only: every row of W sums to 1 and every row of W^400 equals pi.

Honest simplifications

Real culture is not a single number and people do not simply average: they can disagree more after talking, copy only similar colleagues, respond to incentives, leave, or pretend. Trust weights are fixed here; real ones change with experience. The model only says what linear social learning does with a given pattern of trust. No real group is represented.

Opinions over time
Who is influential (left eigenvector pi)
Disagreement (log scale)
Made with 🟠 Blender