A translucent scan front expands outward from the center of a synthetic neural network — a stand-in for the "incredibly detailed and accurate scanning technology" a real mind upload would need. As the front reaches a neuron, that neuron is captured only with probability equal to the scan resolution; at low resolution, many neurons the front sweeps past are missed entirely and stay biological-only (dim), while others are digitized (bright violet). A synapse only survives the copy if both neurons it connects were captured — so connectivity degrades faster than raw neuron count as resolution drops, since one missed neuron silently deletes every edge attached to it.
capture(neuron) = true with P = resolution (once swept)
coherence = (Nmapped / Ntotal) × (Emapped / Etotal)
- Scan resolution — probability a swept neuron is faithfully captured; the article's "atomic resolution" ideal is 100%, but every real scanning technology falls short of it.
- Scan speed — how fast the front expands; a faster scan finishes sooner but doesn't change the resolution — speed and fidelity are independent knobs here, just as a quick sensor pass over a brain wouldn't itself sharpen its accuracy.
- Network complexity — number of modeled neurons and their nearest-neighbor synapses; a richer network has more edges that can be broken by a single missed neuron.
- Digital mind coherence — the fraction of neurons and connections that survive the copy, multiplied together. It stays far below the neuron-only fidelity number whenever gaps are scattered through the network rather than clustered — a rough proxy for whether the reconstructed network would still function as one connected mind rather than a shattered echo of it.
Real-world relevance: even a scan that captures 90% of neurons individually can lose a much larger share of the synaptic wiring between them, which is exactly the gap between "we digitized most of the tissue" and "we preserved the mind" that the hard problem of consciousness makes impossible to close with resolution alone.