Micro-Targeting Field Map: Segmentation & Defense (2D)

Campaign Parameters

Targeting precision 50%

Campaign budget 60%

Population media literacy 30%

Live Readouts

Opinion shift

0.0%

Reached

0.0%

Detection risk (platform anomaly flag)

0.0%

This map is a population field over an ideology (horizontal) × persuadability (vertical) plane — brightness is population density, and background texture is the media-literacy field (grainier = more literate). The dashed ring is a cross-section of the campaign's 3D targeting kernel sliced at the centroid's literacy level. After launch, opinion shift diffuses outward frame by frame — a social-contagion spread across ideologically-similar neighbors, not present in a single-shot targeting hit.

How it works

Every grid cell (x,s) represents ideology x ∈ [-1,1] and persuadability s ∈ [0,1]; a per-cell scalar field ℓ(x,s) (media literacy, ∈[0,1]) is generated by upsampling a coarse random control grid, and a population-density field w(x) follows the same distribution the underlying agent model draws ideology from — a sum of three uniforms, which by the central limit theorem is closely approximated by a Gaussian(0, 1/3):

x = (u1+u2+u3 − 1.5)/1.5,  ui ~ U(0,1)  ⇒  x ≈ N(0, (1/3)²)
w(x) = exp(−x²/(2·(1/3)²))

The campaign targets a fixed centroid c = (ideology 0, persuadability 0.82, literacy 0.18) with the identical Gaussian match kernel as the 3D model, now evaluated per grid cell against the cell's own literacy-field value:

d(x,s)  = ‖(x,s,ℓ(x,s)) − c‖
match   = exp(−d² / (2σ²))
σ = lerp(1.8, 0.25, precision)

Effective influence and the resulting opinion write into the field exactly as in the point-agent model:

resist  = clamp(0.5·ℓ(x,s) + 0.5·literacySlider, 0, 1)
transp  = adTransparencyOn ? 0.5 : 1.0
influence(x,s) = budget · match(x,s) · (1 − resist) · transp
opinion(x,s)  += influence(x,s) · s        (clamped to [−1,1])

What's new here versus a flattened camera view: once the campaign lands, the opinion field diffuses for a few dozen animation frames via an explicit 2D heat-equation step (a discrete Laplacian with a zero-flux/no-leak boundary), modelling word-of-mouth spread across the social graph to ideologically nearby, similarly-persuadable neighbors who weren't directly targeted:

opinion[i,j] += D · (opinion[i-1,j]+opinion[i+1,j]+opinion[i,j-1]+opinion[i,j+1] − 4·opinion[i,j])

Detection risk keeps the same platform-classifier proxy as the 3D model — a narrow, high-budget blast is easy to flag, more so with mandatory disclosure:

concentration = 1 − σ/1.8
detection = clamp( budget · concentration · (adTransparencyOn ? 1.6 : 1.0), 0, 1 )

↑ persuadability

ideology →

⚠ FLAGGED — anomalous targeting pattern

Unaffected cell Opinion shifted Targeting cross-section