Misinformation Effect: How False Memories Get Implanted (2D)

Witnessed Event

You watch a car pass a stop sign and speed off. Later, an interviewer asks a leading question that implies it was actually a yield sign.

Recall Conditions

Time since event 2.0 days

Suggestion strength (leading question) 55%

Misinformation exposures 2

True detail False (implanted) detail Suggestion source

Live Readouts

Trace fidelity100%

P(false detail recalled)0%

Source-monitoring confusion0.00

How it works

The scene shows a "memory trace" as 24 feature-binding nodes (color, sign shape, speed estimate, sequence of events…) arranged in a ring around a hippocampal engram core. A witnessed event first encodes them all as true (green). The model follows the misinformation effect (Loftus & Palmer, 1974) inside the source-monitoring framework (Johnson et al., 1993): a trace only gets overwritten if its original signal has faded enough that a suggested detail can't be told apart from a real memory.

S(t) = e^(-t / τ)              retained original-trace strength
p = L · (1 − S(t))             P(one exposure overwrites a node)
P(false | n) = 1 − (1 − p)^n   P(a detail is corrupted after n exposures)

t — days since the witnessed event; τ is the trace's decay constant (3 days baseline, ×2.2 with retrieval practice — the "testing effect" protects memories from later interference).
L — suggestion strength: how leading the post-event question is (0 = neutral wording, 1 = strongly implies the wrong detail).
n — number of separate misinformation exposures (repeated interviews reinforce the false detail).
Each "Run Recall Test" samples every node against p once per exposure; nodes that flip stay flipped — memories don't spontaneously self-correct once overwritten, matching real memory-conformity studies (about a third of Loftus's subjects "recalled" broken glass that was never there after one leading question).

Real-world relevance: this is why eyewitness testimony is treated cautiously in court, why police interview protocols (e.g. the cognitive interview) avoid leading questions, and why repeated post-event exposure to media coverage or co-witness discussion measurably corrupts eyewitness accounts.

This 2D view lays the same 24 nodes out on a flat ring (a projection of the 3D shell) around the engram core, with the identical decay/suggestion math as the 3D sibling — drag to pan and scroll to zoom the trace.

Trace encoded — 24/24 details true. Adjust conditions, then run a recall test.

Drag to pan · scroll / pinch to zoom