Two weakly electric fish near each other produce overlapping fields. If their discharge frequencies are close, the interference obscures both fish's sensing. Each fish shifts its frequency away from the other's to reopen the gap — but each fish only has access to the interference pattern, not the neighbor's actual frequency.
The sign of the frequency difference (am I higher or lower?) is encoded in the phase relationship between amplitude modulation and timing modulation across body locations. T-units (phase) and P-units (amplitude) together determine direction. Remove the phase information — a "phantom stimulus" — and the algorithm can't determine which way to shift.
See entry-425 for the full context, including Gymnarchus — a separate African lineage that independently evolved the identical algorithm, and therefore the identical failure mode.
What this simulates: The JAR algorithm running on representative slow frequencies (~5 Hz) to make the oscillations visible. Real EODs are 200–600 Hz; the physics and sign computation are the same. Each signal track shows: the two raw EODs, their superposition (interference field), the P-unit envelope output (amplitude of the beat), and T-unit timing marks (zero-crossings of the combined field).
Phantom stimulus mode: In place of a real neighbor EOD, amplitude modulation is applied at the beat frequency — but without the corresponding phase/timing modulation. P-units still fire on the beat rhythm. T-units see only the fish's own EOD (unmodulated zero-crossings). The sign computation has no phase-amplitude relationship to compare, so direction is indeterminate. The algorithm keeps running; the output is random or oscillating. The fish is not broken. The input doesn't carry the information the algorithm was built to extract.
What this cannot show: The spatial distribution of the phase-amplitude relationship across different body locations. In the real fish, T-units and P-units at different positions along the body sample different phase angles of the interference — the sign is encoded across that spatial distribution, not in any single receptor. This simulation collapses that spatial dimension to a single computation. See entry-425.